Progress in Implementing the European Union Coordinated Plan on Artificial Intelligence (Volume 2)

4. AI in manufacturing

Abstract

Artificial intelligence (AI) is increasingly recognised as a catalyst for transforming the manufacturing sector, offering powerful tools to optimise production processes, boost efficiency and enhance resilience across value chains. Yet, the sector’s adoption of AI remains modest and highly fragmented. This chapter outlines the key conditions necessary for responsible, effective and scalable adoption of AI solutions in the European Union. To that end, it examines the strategic role of AI in EU manufacturing, and reviews current AI uptake in the sector with a focus on predictive maintenance, quality assurance and supply chain optimisation. The chapter is based on a literature review and interviews with EU business associations and enterprises between December 2024 and April 2025.

Introduction

The manufacturing sector plays a key role in the economy of the European Union (EU). It employs more than 30 million people and accounts for about 16% of the total value added generated by all economic activities in the European Union. As the European Union strives to modernise its industrial base and strengthen competitiveness, it could harness artificial intelligence (AI) to help update operations, enhance productivity and support resilience.

AI is increasingly recognised as a catalyst for transforming the manufacturing sector, offering powerful tools to optimise production processes, boost efficiency and enhance resilience across value chains. Yet, AI adoption in the EU manufacturing sector remains modest and highly fragmented. Industries such as pharmaceuticals and electronics have emerged as frontrunners in use of AI. However, adoption remains limited in more traditional industries such as food processing, textiles and basic metals, despite their economic weight and employment share.

This chapter provides an overview of how the EU manufacturing sector is using AI, outlining the key conditions necessary to enable the responsible, effective and scalable adoption of AI solutions. It is structured in two parts:

• Overview of the EU manufacturing sector and the strategic role of AI: a profile of the EU manufacturing sector discusses AI use-cases in manufacturing and provides evidence on the level of AI uptake.

• Spotlight on selected AI use-cases in manufacturing: the chapter examines three high-impact AI use-cases in detail – predictive maintenance (PdM); quality assurance and control; and supply chain optimisation. Each section draws on targeted literature reviews and expert interviews, highlighting practical examples of AI deployment. The analysis identifies common barriers, assesses competitiveness potential and provides targeted policy recommendations to support responsible, effective and scalable adoption of AI technologies across the EU manufacturing sector.

Methodological considerations are discussed in Chapter 1. A summary of the key recommendations is provided next.

Key recommendations to enhance AI uptake in manufacturing in the European Union

Data access and sharing

• Foster data sharing within and across sectors: promote interoperable rules, standards and mechanisms to enable broader and more secure data exchange, particularly among SMEs. Increase awareness and incentivise enterprises to join these initiatives by demonstrating benefits while mitigating security concerns (e.g. by facilitating secure data storage) and identifying successful adoption examples. Encourage industry consortia to tackle data-sharing challenges and best practices and develop training and certification.

• Ensure secure and simple data access management: address data-sharing concerns of enterprises by ensuring secure data storage and protection compliant with the General Data Protection Regulation (GDPR). Develop a simple but trustworthy way for enterprises and relevant stakeholders to access data, while ensuring that enterprises have the necessary skills (see below).

Infrastructure

• Address AI infrastructure needs: as part of renewed efforts to help develop dynamic AI ecosystems by bringing together compute, data and talent (e.g. AI Factories), the European Union should help address the advanced simulation data, storage and processing needs of manufacturing enterprises. Support should be provided to establish dedicated AI compute and development centres with HPC capabilities focused on simulation data for manufacturing enterprises.

• Strengthen domestic AI ecosystems: support research collaboration and strategic partnerships. Help bridge the gap between academic AI research and practical enterprise applications to accelerate AI deployment in manufacturing. Encourage partnerships that drive innovation and ensure AI solutions align with real-time needs. Simplify applications for public funding, especially for SMEs and start-ups, to improve the cost-benefit balance.

Regulatory frameworks

• Develop practical guidelines and standards: provide clear, practical and sector-specific guidance on the EU AI Act and related regulation. This must address concerns about high-risk classification, conformity assessment, technical documentation and monitoring.

• Reduce cumulative regulatory burden and fragmentation: simplify compliance across GDPR and AI regulations and foster industrial AI collaboration, while preserving competition. Advocate for the EU AI Board to help ensure consistent AI regulation enforcement and minimise regulatory fragmentation.

Skills and trust

• Strengthen AI workforce development: allocate funding towards AI upskilling and reskilling programmes for the manufacturing workforce. These programmes should be co-created with industry associations, chambers of commerce and social partners to ensure relevance.

• Build trust and awareness in AI adoption: promote communication campaigns and success stories to enhance societal confidence in AI use. Encourage mentorship, training and knowledge exchange to make the most of AI advancements and available traditional manufacturing expertise.

Overview of the EU manufacturing sector and the strategic role of AI

Artificial intelligence (AI) is increasingly recognised as a catalyst for transforming the manufacturing sector, offering powerful tools to optimise production processes, boost efficiency and enhance resilience across value chains. Yet, the sector’s adoption of AI remains modest and highly fragmented. This section provides an overview of EU manufacturing in the European Union, including the sectoral make-up and examples of the fields where AI is applied.

Key characteristics of the manufacturing sector

Sectoral make-up

The manufacturing sector accounts for a considerable share of economic activity in the European Union (EU). In 2023, manufacturing contributed to 16% of total value added of all economic activities in the European Union, significantly higher than shares for human health and social work (7%), transportation and storage (5%), and for agriculture and forestry (2%) (OECD, 2025[1]). There is nevertheless considerable heterogeneity across manufacturing industries. For instance, food, beverages and tobacco products accounted for 13% of total manufacturing value added in 2021 (OECD, 2025[1]). Meanwhile, motor vehicles, trailers, semi-trailers and other transport equipment represented only 1% of total manufacturing value added, reflecting the high reliance on intermediate inputs for such industries.

The fabricated metal products industry represented almost one-fifth of the manufacturing market. In 2023, fabricated metal products accounted for nearly 20% of all enterprises in the manufacturing sector (Figure 4.1). It was followed by the food processing industry (13%), and the repair and installation of machinery and equipment industries (11%). Food processing and fabricated metals also accounted for a considerable share of total manufacturing employment (14% and 12%, respectively), followed by machinery and equipment (10%) and motor vehicles (8%), as shown in Figure 4.2.

Figure 4.1. Number of enterprises by manufacturing industries, 2023

Notes: The chart reflects data on the total number of enterprises by manufacturing industry, using the 2-digit NACE Rev.2 classification and covering all enterprise size classes across all EU Member States. The total number of enterprises by industry is shown in the bar and read on the left-hand side scale. The share of each manufacturing industry in the total number of enterprises in the manufacturing sector is displayed as a diamond and reads on the right-hand side scale. Latest available data, reflecting estimated and provisional statistics for 2023. See Eurostat (2025[2]) for further details on “Enterprises by detailed NACE Rev. 2 activity and special aggregates”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Figure 4.2. Number of employees by manufacturing industry, 2023

Notes: The chart reflects data on the total number of employees by manufacturing industry, using the 2-digit NACE Rev.2 classification and covering all enterprise size classes across all EU Member States. The total number of employees by industry is shown in the bar and read on the left-hand side scale. The share of each manufacturing industry in the total number of employees in the manufacturing sector is displayed as a diamond and reads on the right-hand side scale. Latest available data, reflecting estimated and provisional statistics for 2023. See Eurostat (2025[2]) for further details on “Enterprises by detailed NACE Rev. 2 activity and special aggregates”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

The manufacturing sector is the largest employer in the European Union, with more than 30 million employees in 2023. This is more than more than twice the size of the construction sector (14 million employees) and three times the transportation sector (10 million employees) (Eurostat, 2025[3]). Manufacturing is followed closely by the wholesale and retail sector (just short of 30 million employees).

In the manufacturing sector, industries have both distinct production processes and different business models (Box 4.1). This heterogeneity entails different incentives and challenges for the adoption of AI, according to use-case. Interviews for this chapter included a mix of continuous and discrete manufacturing models that apply AI. However, a dedicated analysis and different interview design is needed to explore differences between models in more depth. More detailed statistics would further help assess any differences in the extent of (and scope for) AI adoption between continuous and discrete manufacturing. Such granular data are not available.

Box 4.1. Continuous and discrete manufacturing

The manufacturing model for end-products that are relatively homogeneous and commoditised (e.g. steel, food, chemicals, paper) relies on economies of scale, continuous manufacturing and low- unit margins. However, this model is different for highly differentiated and customised products. Such products require firms to adjust for small batches (e.g. speciality semiconductors) or even for each unit (e.g. ships, airplanes), according to customer requirements.

• Continuous and massified production involves many repetitive tasks that can be easier to automatise. This is especially relevant as downtime in this type of manufacturing can be expensive, given how the economic model is based on volume. While AI adoption can entail a significant upfront investment, it might be justified by lower unit costs and thus longer-term gains.1 For example, AI applications that support seamless integration with the supply chain, alongside optimised downtime for repair and maintenance might be important for this model.

• Batch or discrete manufacturing models require a different machinery for different products/contracts. Production steps and tasks might vary significantly, entailing a more complex manufacturing system and a case for AI to help optimise the production process across batches/units. AI can improve scheduling and resource allocation, manage variability in workflows and support faster design-to-production cycles through generative AI and AI-assisted engineering tools. Moreover, the benefits of AI adoption could justify the marginal cost for more advanced and expensive equipment needed to deploy sensors and other measurement instruments in production.

While this chapter focuses on manufacturing, the line between manufacturing and services is increasingly blurry, notably for discrete manufacturing. AI adoption across manufacturing industries might reflect this blurred line in the form of different types of AI applications, such as generative AI for concurrent engineering. Discrete manufacturing has an important service component because design and engineering are intrinsically linked to the manufacturing phase for each product. Often, the design and engineering of the product are discussed and agreed with the client and custom-made for each production batch/unit. These are provided by the client or a third party, such as in semiconductor manufacturing.

Note:

1. Long-term cost savings can justify larger capital investment in continuous manufacturing systems in the pharmaceutical industry (Rossi, 2022[4]), where it is possible to opt between continuous or discrete production processes. To the knowledge of the authors, no research focuses on the investment required by AI adoption. However, similarities could be drawn to physical capital investment (notably for hardware such as sensors and metrology equipment).

Sources: (De Backer, Desnoyers-James and Moussiegt, 2015[5]); (OECD, 2025[6]).

AI use-cases

AI encompasses a diverse set of technologies with wide-ranging potential to transform manufacturing processes, products and systems. Early forms of AI have been used in manufacturing for more than 40 years, even if limited to a handful of applications (Nolan, 2021[7]).

Within the manufacturing sector, AI can support a variety of core functions – from forecasting and event detection to optimisation, recognition and interaction. These applications span different stages of the industrial value chain. For example, AI can be used to detect anomalies and forecast equipment failure in predictive maintenance (PdM). It can also help estimate energy needs to optimise production schedules through energy consumption forecasting. In supply chain operations, AI supports demand forecasting, inventory management, and routing decisions. It also plays a role in real-time process optimisation and job scheduling to reduce production costs and improve throughput. AI techniques enable generative design and simulation through digital twins, allowing manufacturers to explore new geometries and operational scenarios. Additionally, AI contributes to quality assurance by detecting visual defects; to workplace safety by identifying hazards; and to robotics and automation by enabling object recognition, path planning and human-machine interaction.

These use-cases illustrate the varied nature of AI in manufacturing and can be categorised according to their primary functions (Table 4.1) using the OECD Framework for the Classification of AI Systems (OECD, 2022[8]). The use-cases entail different levels of maturity, including relatively new applications (e.g. concurrent design, engineering and manufacturing) and promising applications (e.g. holistic optimisation). These complement more mature applications (e.g. predictive maintenance; quality control) where widespread adoption in manufacturing creates the potential for considerable gains. The use-cases treated in more depth (Part 2) were selected in co‑ordination and with the agreement of the European Commission (see methodology in Chapter 1 for details).

Table 4.1. Examples of application fields of AI in manufacturing

AI application field	AI systems tasks	Description and examples	Type of learning/reasoning	Challenges and barriers reported in literature
Predictive maintenance	Event detection; forecasting	Detects anomalies and forecasts component failure to schedule repairs pre-emptively. Examples: Proactive gearbox service in rotating machinery; sensor-based alerting for incipient motor failures.	Unsupervised anomaly detection (e.g. outlier analysis), supervised regression for time-to-failure estimates.	Data sparsity for rare failures; sensor noise; difficulty in generalising models across equipment types; lack of real-time data integration.
Dynamic production and maintenance scheduling	Forecasting; goal-driven optimisation; event detection	Optimises production through dynamic production scheduling and maintenance planning, while minimising expected total downtime and maintenance costs by accounting for opportunistic grouping of similar product batches and component maintenance activities, including breakdown costs associated with failure risk. This use-case integrates process optimisation with predictive maintenance – see corresponding table entries. Examples: production scheduling for different and/or complex products.	Deep reinforcement learning enables real-time decisions, using live data from a digital twin, including condition monitoring and production progress information.	Data quality challenges, model transparency, real-time decision demands, legacy system integration.
Energy consumption forecasting	Forecasting	Predicts energy loads and usage patterns for scheduling heavy operations. Examples: Co‑ordinating machine usage during off-peak times, responding to real-time grid signals, balancing local energy generation with demand.	Recurrent Neural Networks (RNN) (i.e. long short-term memory recurrent neural networks); Support Vector Machine regression; baseline forecasting.	Inaccurate or missing data; variability in operational conditions; difficulty in capturing external factors like weather or grid events; need for interpretability.
Supply chain optimisation	Forecasting; goal-driven optimisation	Forecasts demand and optimises replenishment, routing or scheduling under constraints. Examples: Demand forecasting for raw materials; dynamic warehouse routing; and inventory planning to avoid stockouts or overstock.	Time-series regression or RNN for demand; reinforcement learning (RL) or optimisation for route and inventory decisions.	Complexity of global supply chains; data silos and integration issues; geopolitical or pandemic-related disruptions; lack of trust in AI-driven decisions.
Holistic optimisation	Forecasting; goal-driven optimisation; pattern recognition; event detection	Provides an integrated approach to the supply chain and internal production processes combining AI, generative AI, digital twins and optimisation techniques. Digital twins create dynamic, data-driven virtual models of physical assets and processes. It offers enhanced visibility, real-time monitoring, predictive maintenance and improved decision-making capabilities. This use-case integrates process optimisation with supply chain optimisation – see corresponding table entries. Example: Simulating supply chain disruptions allows manufacturers to proactively assess potential risks and create contingency strategies and adjusting production schedules accordingly.	Digital twins and optimisation techniques; Data from radio frequency identification tags, barcodes and Internet of Things sensors, neural networks and regression analysis.	Full potential of digital twins includes challenges to data quality, scalability, security and interoperability. A holistic approach to all manufacturing systems also entails the challenges of several other AI use-cases (e.g. predictive maintenance, process optimisation, supply chain optimisation, etc.).
Process optimisation	Goal-driven optimisation	Tunes process parameters in near real time to minimise scrap or cost. Examples: Dynamic control of injection pressure to cut rejects, adjusting welding temperature for stronger joints.	RL for continuous control, evolutionary algorithms (e.g. Genetic Algorithm and Particle Swarm Optimisation for multi-objective solutions).	High cost of failed experiments; difficulty of simulating complex processes accurately; resistance from operators; slow feedback loops.
Scheduling and dispatch	Goal-driven optimisation	Allocates jobs or dispatches automated guided vehicles to minimise makespan or idle times. Examples: Job-shop scheduling in discrete manufacturing, forklift pathing in large facilities, controlling queue lengths in multi-machine setups.	RL for sequential decision making; heuristic/ evolutionary algorithms for multi-objective scheduling.	Need for real-time responsiveness; legacy IT systems.
Generative design, research and development; concurrent design, engineering and manufacturing	Reasoning with knowledge structures; goal-driven optimisation	Explores new product geometries or simulates manufacturing processes with “what-if” scenarios. This also includes generative AI for concurrent design and engineering of products, their manufacturing processes and product life cycle. Examples: Lightweight part design in 3D printing, digital twins that optimise line speed or resource usage under varied conditions. Digital twin used to plan production and assembly stages.	Generative Adversarial Networks (GANs), physics-based machine learning or symbolic reasoning for design exploration. Variational autoencoders and transformer-based models for concurrent engineering.	High computational cost; need to adapt to domain-specific requirements; lack of standardisation in digital twin frameworks; low user familiarity with design automation. Accurate detection of defects for maintaining high product quality and ensuring customer satisfaction.
Quality assurance	Recognition	Identifies defects in real time (visual or sensor-based) and flags substandard items on the line. This includes AI integrated with cyber-physical systems (CPS) for smart manufacturing and maintenance. Examples: Automated surface inspection of weld seams, castings or circuit boards to reduce scrap and manual checks.	Supervised classification, typically with convolutional neural networks (CNNs) for image recognition.	Need for large, labelled datasets; changing lighting/visual conditions; difficulty in detecting subtle defects; high false positive/negative rates.
Safety and security	Recognition; event detection	Spots unsafe events or detects hazards in the workplace. Examples: Intrusion detection in industrial operational technology systems, camera-based personal protective equipment checks, real-time alerts on unusual sensor reading. CPS systems to detect human presence near heavy machinery and disable machines to prevent accidents.	Unsupervised anomaly detection, CNN-based classification (vision), occasional GAN-based anomaly modelling; AI integrated with CPS, Internet of Things technologies, robotics and machinery.	Privacy and ethical concerns; false alarms; adversarial attacks on vision systems; high set-up and maintenance costs.
Robotics and automation	Recognition; interaction support	Recognises surroundings and/or interacts with humans for collaborative tasks, including through integration with CPS systems. Examples: Vision-guided bin picking, collision-avoidance for collaborative robots (cobots), voice or gesture commands in flexible assembly lines.	CNNs (object recognition), RL (path planning), and occasionally Natural Language Processing /gesture recognition.	Safety certification; unpredictable human behaviour; integration with workflows; latency in real-time decision making.

Source: Sierla et al. (2018[9]), “Automatic assembly planning based on digital product descriptions”, https://doi.org/10.1016/j.compind.2018.01.013; Radanliev et al. (2020[10]), “Artificial intelligence in cyber-physical systems”, https://doi.org/10.1007/s00146-020-01049-0; Shrivastav (2022[11]), “Barriers related to AI implementation in supply chain management”, https://www.igi-global.com/pdf; Agrawal et al. (2023[12]), “Digital twin: Where do humans fit in?”, https://doi.org/10.1007/s00146-020-01049-0; Hoffman and Reich (2023[13]), “A systematic literature review on artificial intelligence and explainable artificial intelligence for visual quality assurance in manufacturing”, https://doi.org/10.3390/electronics12224572; Mypati et al. (2023[14]); “A critical review on applications of artificial intelligence in manufacturing”, https://doi.org/10.1007/s10462-023-10535-y; Plathottam et al. (2023[15]), “A review of artificial intelligence applications in manufacturing operations”, https://doi.org/10.1002/amp2.10159; Soori, Arezoo and Dastres (2023[16]), “Digital twin for smart manufacturing: A review”, https://doi.org/10.1016/j.smse.2023.100017; Culot, Podrecca and Nassimbeni (2024[17]), “Artificial intelligence in supply chain management: A systematic literature review of empirical studies and research directions”, https://doi.org/10.1016/j.compind.2024.104132; Espina-Romero et al. (2024[18]), “Challenges and opportunities in the implementation of AI in manufacturing: A bibliometric analysis”, https://doi.org/10.3390/sci6040060; Espinosa-Jaramillo (2024[19]), “Digital twins in supply chain operations bridging the physical and digital worlds using AI”, https://doi.org/10.52783/jes.5434; Gao et al. (2024[20]), “Artificial Intelligence in manufacturing: State of the art, perspectives, and future directions”, https://doi.org/10.1016/j.cirp.2024.04.101; Liu, Pan and Ballot (2024[21]), “Unveiling the potential of digital twins in logistics and supply chain management: Services, capabilities, and research opportunities”, https://doi.org/10.1016/j.dte.2024.100025; Ouahabi et al. (2024[22]), “Leveraging digital twin into dynamic production scheduling: A review”, https://doi.org/10.1016/j.rcim.2024.102778; Benhanifia et al. (2025[23]), “Systematic review of predictive maintenance practices in the manufacturing sector”, https://doi.org/10.1016/j.iswa.2025.200501; Kermati Feyz Abadi et al. (2025[24]); “Leveraging AI for energy-efficient manufacturing systems: Review and future prospectives”, https://doi.org/10.1016/j.jmsy.2024.11.017; Khurram et al. (2025[25]), “Artificial Intelligence in manufacturing industry worker safety: A new paradigm for hazard prevention and mitigation”, https://doi.org/10.3390/pr13051312; Ouahabi et al. (2025[26]), “Dynamic production scheduling and maintenance planning under opportunistic grouping”, https://doi.org/10.1016/j.cie.2024.110646; Urrea and Kern (2025[27]), “Recent advances and challenges in industrial robotics: A systematic review of technological trends and emerging applications”, https://doi.org/10.3390/pr13030832; Shafiee (2025[28]), “Generative AI in manufacturing: A literature review of recent applications and future prospects”, https://doi.org/10.1016/j.procir.2025.01.001.

The rapid rise of generative AI (GenAI) presents significant opportunities across the manufacturing value chain. Potential applications include accelerating product design, as well as research and development (R&D) by generating prototypes or simulating scenarios; automating software code generation for embedded systems or control software; enhancing customer service through intelligent chatbots; creating technical documentation or marketing content; and boosting general office productivity. In 2024, European enterprises invested on average USD 110 million in initiatives to deploy and advance GenAI. The widespread adoption of GenAI could potentially contribute an additional 8% of Europe’s gross domestic product over ten years (ERT, 2025[29]).

Level of AI uptake in the manufacturing sector in the European Union

AI could boost productivity and drive economic growth, but it does not affect all industries equally due to the different nature of their activities, exposure and potential for AI adoption. Recent research shows that uneven rates of AI adoption across industries could limit the potential aggregate (economy-wide) productivity growth driven by AI (Filippucci, Gal and Schief, 2024[30]). These estimates are based on adoption rate differences (and productivity differences) across industries in the United States. As such, a better understanding of the factors influencing AI use in each industry would be needed to reap the full benefits of AI (Calvino et al., 2024[31]).

The adoption of AI technologies in the manufacturing sector has increased in recent years. The share of enterprises using AI rose from 7% to 11% between 2021 and 2024 among enterprises with ten or more employees (Figure 4.3). Similarly, AI talent concentration in manufacturing has increased between 2018 and 2024. However, the share of workers with AI skills remains low in the sector – below 2% for most EU Member States (Figure 4.4). Differences in talent concentration across EU Member States might also be driving different rates of AI adoption in manufacturing enterprises (Figure 4.3). AI uptake in the manufacturing sector (10.6% of enterprises) remains below overall AI adoption across all sectors in the European Union, where 13% of enterprises with ten or more employees reported using AI (Figure 4.5). Compared to other main economic activities, manufacturing continues to lag in AI adoption. This is especially the case in relation to the information and communication sector, where nearly half of the enterprises have integrated AI into their operations. Nevertheless, AI use in manufacturing is more widespread than in construction or accommodation and food services (both at 6%). These differences reflect variations in the digital intensity of economic activities and the extent to which tasks can be automated or enhanced through AI-based solutions (Calvino et al., 2024[31]).

Figure 4.3. Enterprises using AI technologies in manufacturing in EU Member States

Note: The figure shows the percentage of enterprises with at least ten employees in the manufacturing sector using at least one AI technology in 2021 and 2024.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Figure 4.4. AI talent concentration in the manufacturing sector in EU Member States

Notes: This chart shows the concentration of LinkedIn members with at least two AI engineering skills or who perform or have previously performed an AI occupation per country, industry and in time. Please see the methodological note for more information.

Source: (OECD.AI, 2025[32]), AI talent concentration by country and industry, calculations based on data from LinkedIn Economic Graph, last updated 2025-04-07, https://oecd.ai/.

Figure 4.5. Enterprises using AI technologies by economic activity in the European Union, 2024

As a percentage of enterprises with ten or more employees

Notes: “All activities” include all economic sectors except agriculture, forestry and fishing; mining and quarrying; and the financial sector. The percentage of enterprises using AI in a given industry and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the specific industry. Includes only enterprises with at least ten employees. AI technologies include performing analysis of written language (text mining); generating written or spoken language (natural language generation); automating different workflows or assisting in decision making (AI-based software robotic process automation); converting spoken language into machine-readable format (speech recognition); identifying objects or persons based on images (image recognition, image processing); machine learning (e.g. deep learning) for data analysis; and enabling physical movement of machines via autonomous decisions based on observation of surroundings (autonomous robots, self-driving vehicles, autonomous drones). See Eurostat (2025[2]) for further details on the data obtained through the “EU survey on ICT usage and e-commerce in enterprises”.

Source: Eurostat (2025[33]), “Statistics explained: Use of artificial intelligence in enterprises”, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises.

Within the manufacturing sector, AI use has increased across all industries, although the pace of uptake varies considerably. Industries such as pharmaceuticals and electronics are emerging as frontrunners, with 26% and 25% of enterprises, respectively, reporting use of AI technologies. In contrast, adoption remains relatively limited in industries such as textiles and apparel (6%), food processing (7%), basic metals (9%) and wood and paper (9%) (Figure 4.6).

The limited use of AI in food processing is noteworthy. The industry is among the largest manufacturing employers in the European Union; accounts for a large share of all EU businesses; and contributes substantially to the total value of the manufacturing sector. Legacy infrastructure and equipment that are incompatible with AI technologies might be part of specific technological needs or barriers unique to these industries (Calvino et al., 2024[31]).

Most AI technologies used in manufacturing are developed by third parties. In 2024, approximately 6% of manufacturing enterprises reported using off-the-shelf commercial software or systems and 3% used technology developed by external providers (Eurostat, 2025[2]). Just over 4% of manufacturing enterprises were using commercial AI software or open-source AI software that had been internally adapted to their needs. Conversely, less than 2% of manufacturing enterprises developed AI solutions internally in-house. Since internal AI typically requires specialised expertise and substantial investment, it is more common in larger enterprises.

Figure 4.6. Enterprises using AI technologies in the manufacturing industry in the European Union

As a percentage of enterprises with ten or more employees

Notes: The percentage of enterprises using AI in a given industry and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the specific industry. Includes only enterprises with at least ten employees. AI technologies include performing analysis of written language (text mining); generating written or spoken language (natural language generation); automating different workflows or assisting in decision making (AI-based software robotic process automation); converting spoken language into machine-readable format (speech recognition); identifying objects or persons based on images (image recognition, image processing); machine learning (e.g. deep learning) for data analysis; and enabling physical movement of machines via autonomous decisions based on observation of surroundings (autonomous robots, self-driving vehicles, autonomous drones). See Eurostat (2025[2]) for further details on the data obtained through the “EU survey on ICT usage and e-commerce in enterprises”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Outsourcing AI to third parties can reduce development costs, but it may pose other challenges. When systems must be tailored to the operations of an enterprise, sensitive data may need to be shared with third parties to train the AI. AI adoption is generally more prevalent in larger and younger (start-up) enterprises (Calvino and Fontanelli, 2023[34]), whereas smaller enterprises rely on external providers. For more established enterprises, corporate practices and traditions may also present barriers to internal development and implementation.

Access to large volumes of quality data is a key enabler of AI deployment and effectiveness. Enterprises with long-standing operations and extensive historical data hold a significant advantage in training and deploying AI systems. Conversely, newer enterprises, or those with limited access to data, may face growing barriers to competitiveness as data become increasingly strategic assets. However, alternative approaches are emerging that may mitigate these disadvantages. For example, simulation-based AI models, such as those developed by Phantasma Labs (2025[35]), use millions of synthetic training scenarios rather than relying on large historical datasets. Powered by reinforcement learning, such systems can be deployed quickly and require only minimal one-time data inputs. These models are particularly well-suited to small and medium-sized enterprises (SMEs), which may lack the data infrastructure or computing capacity for traditional AI development.

Compared to other economic activities, AI in manufacturing can have an especially important role in physical production. Figure 4.7 shows that 26% of all the manufacturing enterprises using AI in 2024 (compared to 23.5% across all sectors) used AI technology for optimising production processes. These included predictive maintenance, quality control, and autonomous drones and robots. Despite the potential productivity gains through process optimisation, only 2.8% of EU manufacturing enterprises use AI for this purpose. Industries that do invest in process optimisation are driven mostly by petroleum (10.7% of all enterprises in this industry) and pharmaceuticals (7.6%). Other industries include electronics (6.0%), transport equipment (5.9%) and electrical equipment (5.1%). More detailed survey data would help better monitor adoption of AI for specific use-cases within production, including those outlined in Table 4.1 above.

Figure 4.7. AI use-cases in EU manufacturing enterprises

As a percentage of enterprises using AI in manufacturing/all sectors

Note: “All activities” include all economic sectors except agriculture, forestry and fishing; mining and quarrying; and the financial sector. The data show the number of enterprises using AI for each specific application, as a percentage of all the enterprises (with at least ten employees) using at least one AI technology in that sector. See Eurostat (2025[2]) for further details. ICT: Information and communication technology.

Source: Eurostat (2025[33]), “Statistics explained: Use of artificial intelligence in enterprises”, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises.

AI for information and communication technology (ICT) security is also a relatively important use- case in EU manufacturing (24.3% of all enterprises using AI ), compared to all economic activities (22.4%). AI in ICT is mostly driven by the petroleum and pharmaceuticals industries. AI-driven logistics are less common but still relatively more important in manufacturing (9.4%) than across all economic activities (6.1%). EU manufacturing enterprises that use AI are less likely to use it for marketing, business administration, accounting and R&D.

In several promising AI applications, adoption appears to be particularly low but would be suited for enhancing efficiency and productivity in manufacturing industries. The adoption of AI in manufacturing varies significantly across different application areas, with a noticeable concentration in language-related functions rather than core production processes. Text mining is the most widely used AI application, adopted by 4.6% of manufacturing enterprises (Figure 4.8). Other uses related to language and text, such as natural language generation (3.5%) and speech recognition (2.9%). These functions are typically associated with customer-facing services, knowledge management or administrative tasks.

By contrast, AI applications more closely tied to manufacturing operations are used less frequently. Workflow automation and assistance in decision making (e.g. robotic process automation) are key for fabrication processes and supply chain optimisation, but only 3.2% of manufacturing enterprises used them in 2024. Meanwhile, only 2.7% of manufacturing enterprises used image recognition and image processing, despite them being instrumental in quality assurance. Machine learning (ML) for data analysis and robot automation (autonomous decisions by robots) can help optimise fabrication processes and supply chain management. However, only 2.7% and 1.5% of manufacturing enterprises, respectively, used them in 2024.

Figure 4.8. Enterprises using AI technologies by function in the European Union, 2024

Note: “All activities” include all economic sectors except agriculture, forestry and fishing; mining and quarrying; and the financial sector. The percentage of enterprises using AI in a given sector and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the specific sector. Includes only enterprises with at least ten employees. See Eurostat (2025[2]) for further details.

Source: Eurostat (2025[33]), “Statistics explained: Use of artificial intelligence in enterprises”, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises.

The above-mentioned patterns suggest a mismatch between the transformative potential of AI in manufacturing and its actual deployment, particularly in areas critical to productivity gains. The most significant long-term impacts are expected to stem from scalable and strategically transformative applications. However, interview findings indicate that most manufacturing sector firms that have adopted AI focus on improving existing products or processes. Similarly, in a recent survey, 94% of French firm managers of SMEs (across sectors) cited at least one method of optimising systems with the help of AI. Conversely, only slightly more than half cited at least one AI-based method of developing their business by, for example, gaining market share or accessing new markets (BPI France, 2025[36]).

A closer examination of two AI applications in manufacturing – image recognition and processing, and robotic process automation – reveals considerable heterogeneity in adoption across manufacturing industries. Fuel refineries (“coke and petroleum”), pharmaceuticals, electronics and transport equipment reported the highest uptake of AI. More than 5% of enterprises in these industries reported use of AI imagery (Figure 4.9) or robotic automation technologies (Figure 4.10) in 2024. By contrast, adoption remains relatively limited in industries such as food processing, textiles, basic metals, and wood and furniture, where fewer than 3% of enterprises reported using these technologies. This disparity may reflect a combination of sector-specific technological constraints, differences in digitalisation and the influence of established corporate practices in these more traditional industries.

These findings highlight the considerable untapped potential for productivity gains through AI adoption in manufacturing. A recent survey of 247 industrial enterprises in Germany, Austria and Switzerland suggested that consistent use of GenAI in manufacturing could boost profitability by up to 10.7%. This, in turn, would translate into an additional EUR 27 billion in profit (Frankfurter Allgemeine, 2025[37]). However, the same study found that only 0.74% of the 10.7% potential profitability gains have been realised to date, underscoring the gap between expectations and implementation. While many enterprises express high expectations for the opportunities presented by GenAI, the implementation or development of new business models has yet to be fully embraced (Frankfurter Allgemeine, 2025[37]). The greatest impact is projected in sales and marketing, where AI could increase profitability by up to 2.4%, followed by R&D and planning, each with a potential gain of 1.7%. In contrast, supporting functions such as information technology, where AI is most widely used, appeared to offer comparatively smaller marginal gains (Frankfurter Allgemeine, 2025[37]).

Figure 4.9. AI use for image recognition across manufacturing industries in the European Union, 2021 and 2024

As a percentage of enterprises with ten or more employees

Notes: The percentage of enterprises using AI in a given industry and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the specific industry. Includes only enterprises with at least ten employees. Data for the coke and petroleum industry (NACE2 C19) in 2024 are not available. AI for robotic process automation refers to AI technologies automating different workflows or assisting in decision making (AI-based software robotic process automation) – see Eurostat (2025[2]) for further details on the data obtained through the “EU survey on ICT usage and e‑commerce in enterprises”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Figure 4.10. AI use for robotic process automation across manufacturing industries in the European Union, 2021 and 2024

As a percentage of enterprises with ten or more employees

Note: The percentage of enterprises using AI in a given industry and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the specific industry. Includes only enterprises with at least ten employees. Data for the coke and petroleum industry (NACE2 C19) in 2024 are not available. AI for robotic process automation refers to AI technologies identifying objects or persons based on images (image recognition, image processing) – see Eurostat (2025[2]) for further details on the data obtained through the “EU survey on ICT usage and e‑commerce in enterprises”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Several key barriers continue to constrain AI adoption in manufacturing. Skills shortages, data limitations and regulatory complexity are among the most frequently cited challenges to AI adoption in manufacturing. In 2024, more than 7.5% of manufacturing enterprises reported a lack of the relevant expertise as one of the main reasons for not using AI (Figure 4.11). Other key barriers noted were data availability or quality (5.0%) and (in)compatibility of equipment, software and systems with AI technology (4.8%). Concerns about legal consequences (4.9%) and with data protection and privacy (4.4%) were also important considerations for not using AI. These findings suggest that efforts to better communicate on regulation and to clarify legal implications would help mitigate such concerns. Costs-related barriers appeared as less prominent, with 3.9% of manufacturing enterprises indicating they did not use AI technologies because of seemingly high costs; this might reflect computational costs, among others. Among other constraints, 2.0% cited ethical considerations and 2% indicated that AI technologies were not useful for them.

These results are well aligned with a survey of 247 industrial enterprises in Germany, Austria and Switzerland (Frankfurter Allgemeine, 2025[37]). The survey identified the biggest obstacles to broader adoption as lack of data (25%), a shortage of AI specialists and insufficient IT infrastructure (both at 24%). Together, these results highlight the need for a supportive ecosystem that combines investment in digital infrastructure and skills with regulatory clarity to enable more widespread and effective use of AI in manufacturing.

Figure 4.11. Perceived barriers to AI adoption in manufacturing in the European Union, 2024

As a percentage of enterprises with ten or more employees

Notes: The percentage of enterprises using AI in a given sector and year is defined as the number of enterprises using at least one AI technology relative to all enterprises in the manufacturing sector. Includes only enterprises with at least ten employees. See Eurostat (2025[2]) for further details on the data obtained through the “EU survey on ICT usage and e-commerce in enterprises”.

Source: Eurostat (2025[2]), “Artificial intelligence by NACE Rev. 2 activity”, https://doi.org/10.2908/ISOC_EB_AIN2.

Spotlight on selected AI use-cases in manufacturing

The following sections explore in depth three use-cases that exemplify the transformative potential of AI in manufacturing: i) predictive maintenance; ii) quality assurance; and iii) supply chain optimisation. These applications have the level of maturity and track record required to analyse and understand the extent of their adoption in manufacturing. Despite their level of maturity of these use-cases, the data described here suggest plenty of room exists to broaden adoption across manufacturing industries. This is especially case among more “traditional” industries where additional efforts might be needed to attain the level of digital maturity required to adopt AI. Encouraging uptake of these applications in manufacturing and addressing the barriers and challenges described later in Part 2 could entail important economic benefit.

Better data and additional analyses would be required to fully map the level of adoption of AI across manufacturing firms in the European Union for the use-cases considered in this report and estimate the potential economic benefit from enhanced adoption across different industries. The EU survey on ICT usage and e‑commerce in enterprises can play an important role by providing increasingly granular data that allow monitoring and estimate the impact of different AI applications and use-cases.

Main AI use-cases

AI technologies are increasingly embedded across core manufacturing processes, including casting, machining, forming, welding and additive manufacturing, where they can support real-time monitoring, process optimisation, defect detection and predictive analytics. These applications illustrate how AI is affecting both individual production steps and entire manufacturing processes by improving quality, efficiency and responsiveness (Plathottam et al., 2023[15]; Soori, Arezoo and Dastres, 2023[16]).

Table 4.2 provides an overview of the three selected high impact use-cases. It includes how AI is applied, the types of data required, key operational impacts and their potential to strengthen EU manufacturing competitiveness. The subsequent sections examine each use-case in greater detail, integrating examples of implementation in practice from the interviewed enterprises.

Table 4.2. Summary table based on literature review

Use-case	Predictive maintenance	Quality assurance/control	Supply chain optimisation
How AI works	AI models analyse sensor data from machines to detect anomalies, diagnose faults and estimate component degradation. Predictive models learn patterns from historical failure data to anticipate breakdowns and optimise maintenance timing. AI is integrated into real-time monitoring, decision support and alert systems to reduce unplanned downtime.	AI systems use image data, sensor signals and process variables to detect, classify and localise defects in real time. Other models analyse upstream production data to identify process deviations and adjust parameters before quality issues occur. Visual inspection and classification rely on computer vision, while process optimisation uses pattern recognition and forecasting techniques.	AI systems learn from demand signals, inventory levels, supplier data and logistics metrics to improve forecasting, automate procurement and optimise transport. Algorithms predict demand shifts, flag supplier risks and dynamically plan routes or schedules. Decision making is supported across the chain from sourcing and production to delivery, using real-time and historical data.
Data required	Time-series sensor data (vibration, temperature, pressure), equipment health logs, maintenance records and failure data. Often requires high-frequency sampling and labelled fault data for training.	High-resolution image data (e.g. surface scans), vibration/ temperature/pressure readings, product specifications, defect labels and process parameter logs. Data must be consistent across production batches and may include rare or edge-case examples.	Historical sales, point of sale data, weather forecasts, supplier performance metrics, logistics tracking data, telematics (fleet/warehouse), inventory levels and contract metadata. Real-time and unstructured data (e.g. news, e‑mails) increasingly complement traditional planning datasets.
Main impacts	Reduced unplanned downtime, extended equipment life, improved maintenance planning, lower repair costs, increased operational efficiency. Enhanced asset availability and production continuity, especially in capital-intensive sectors.	Reduced defect rates, faster detection and correction, improved yield and consistency, lower rework/scrap rates, enhanced traceability and more stable process performance. Also enables real-time visual inspection and automated documentation, improving quality assurance workflows.	Improved demand-supply balance, reduced inventory costs, greater sourcing resilience, better logistics co‑ordination, faster fulfilment and adaptive response to disruptions. AI enables end-to-end visibility, dynamic decision making, and sustainability-oriented optimisation (e.g. energy-efficient routing, greener sourcing).
Competitiveness potential	Strengthens industrial reliability and asset-intensive sectors (e.g. steel, energy, aerospace). Supports predictive services and AI-integrated manufacturing. Boosts EU industrial resilience through better uptime and lifecycle management.	Enhances product quality and certification, supports advanced manufacturing competitiveness and reduces dependency on manual inspection labour. Aligns with EU goals on industrial excellence and digitalised quality control.	Boosts resilience of EU manufacturing supply chains, supports green and circular economy strategies, enhances responsiveness to disruptions (e.g. pandemics, trade wars) and strengthens sovereignty in supply networks. Builds competitive advantage through agile, data-driven supply chain management.

Predictive maintenance

PdM represents one of the most important applications of AI in manufacturing (Jakubowski et al., 2024[38]). PdM leverages AI to monitor equipment health, anticipate failures and optimise maintenance schedules based on real-time data analysis (Ucar, Karakose and Kırımça, 2024[39]). In contrast to traditional preventive approaches, which rely heavily on fixed intervals or manual inspections, AI-driven PdM integrates continuous sensor-based monitoring and advanced analytics. This increases accuracy, reduces operational downtime and minimises maintenance costs (Rojek et al., 2023[40]; Jakubowski et al., 2024[38]).

The implementation of AI-driven PdM typically involves the following six interconnected components (Figure 4.12):

Figure 4.12. AI-driven predictive maintenance process

Source: Based on Ucar, Karakose and Kirimça (2024[39]), “Artificial intelligence for predictive maintenance applications: Key components, trustworthiness and future trends”, https://doi.org/10.3390/app14020898.

PdM should be distinguished from other types of planning, such as use- or time-based maintenance that depend on predefined schedules. PdM, or “condition-based” maintenance, has been practised in industrial settings for several decades. However, the use of AI allows for more accurate detection of complex failure patterns and the analysis of larger, more diverse data streams, helping maintenance engineers make better decisions. Implementing AI-based PdM can be costly because of the required sensors and instruments required to monitor the manufacturing equipment, in addition to compute costs.

Some industries might be able to justify adoption of AI-driven PdM on economic grounds. Industries using costly and delicate equipment for complex manufacturing processes (e.g. extreme ultraviolet lithography machines in manufacturing semiconductors) might ultimately save money by investing in AI-driven PdM. The same is true for those relying mostly on continuous manufacturing with critical machinery working around the clock to drive down average costs (e.g. cold rolling mill for flat steel manufacturing).

Conversely, more “traditional” industries need to carefully analyse the trade-offs of investing in AI-PdM. Traditional industries relying on legacy equipment and/or on less complex machinery might find the upfront investment required for PdM (e.g. retrofitting machinery with sensors and measuring instruments) as relatively high. They need to compare it to the value of the manufacturing machinery itself and with the costs of reactive repairs or replacing machinery (e.g. spinning equipment in textiles manufacturing) (Cottongins, 2025[41]). It would therefore be important to help firms across all industries, particularly SMEs and firms in more “traditional” industries, to quantify the long-term potential benefits of AI-driven predictive maintenance, and to understand in which contexts such investments are economically justified, given the characteristics and criticality of their equipment.

Main use-cases reported in literature

Academic literature identifies three key applications of AI in predictive maintenance: i) condition monitoring and anomaly detection; ii) fault diagnosis; and iii) prediction of remaining useful life (RUL). Condition monitoring systems apply AI-based methods to sensor data, enabling early anomaly detection and failure prevention (Fordal et al., 2023[42]; Surucu, Gadsden and Yawney, 2023[43]). Fault diagnosis techniques employing hybrid AI methods can enhance root cause identification, further refining maintenance interventions (Rojek et al., 2023[40]). Additionally, AI algorithms, particularly deep-learning models, have improved the accuracy of predicting RUL, thereby supporting more targeted maintenance scheduling (Jakubowski et al., 2024[38]; MPO, 2024[44]).

Condition monitoring and anomaly detection

Condition monitoring and anomaly detection form the foundation of many AI-based predictive maintenance (PdM) systems (Surucu, Gadsden and Yawney, 2023[43]). These systems rely on data collected from a wide range of industrial sensors that measure key operational parameters (e.g. temperature, vibration, acoustic emissions and pressure) in real time. AI models are trained on this sensor data to recognise subtle deviations from normal operating patterns data, which may indicate early-stage mechanical issues such as wear, imbalance or misalignment. When such anomalies are detected, the system can issue early warnings, prompting maintenance teams to investigate and address potential issues before they turn into costly failures (Keleko et al., 2022[45]; Raza, 2023[46]; Banjeree, Ashok and Sharma, 2024[47]).

Sensor fusion further enhances anomaly detection capabilities by integrating data from multiple sources – such as vibration, thermal imaging, and ultrasonic sensors – to provide a more complete and reliable assessment of equipment condition (Raza, 2023[46]; Wiese, 2024[48]). Studies show that combined sensor data improves the detection of faults that might otherwise go unnoticed when relying on a single data stream (Banjeree, Ashok and Sharma, 2024[47]; Wiese, 2024[48]). For example, the integration of thermal and acoustic signatures can help distinguish between benign process variability and early indicators of failure. These systems are increasingly deployed within Internet of Things frameworks, allowing edge-based processing and decentralised decisions across distributed production assets (Ucar, Karakose and Kırımça, 2024[39]).

A further application of integrated AI and sensor technology is demonstrated by Sensorminds (2024[49]), a company offering smart sensors and real-time dashboards for machine performance monitoring. The sensors help overcome the challenge of data acquisition, even with outdated machinery. Meanwhile, a live dashboard features alarms to monitor performance against set thresholds, such as identifying underperforming production.

In addition to enhancing fault detection, AI-enabled condition monitoring supports a shift from scheduled, time-based maintenance to dynamic, condition-based strategies (Surucu, Gadsden and Yawney, 2023[43]). This can reduce unnecessary inspections and extend component (Rojek et al., 2023[40]; Jakubowski et al., 2024[38]).

However, implementation challenges remain. False positives can reduce user trust and cause operational inefficiencies, while false negatives risk allowing critical issues to go undetected (López et al., 2023[50]). Ongoing research seeks to address these limitations by refining model accuracy, improving data quality and developing hybrid approaches that combine AI-based diagnostics with expert rules or physics-based models (Raza, 2023[46]; Ucar, Karakose and Kırımça, 2024[39]).

Fault diagnosis and classification

Fault diagnosis and classification is a key application of AI in PdM. This process enables systems both to detect a problem and to identify the type of fault and where it originates (Singh et al., 2023[51]). This added layer of insight supports more targeted and efficient maintenance interventions, helping reduce troubleshooting time, avoid unnecessary part replacements and improve root cause analysis (Keleko et al., 2022[45]; Rojek et al., 2023[40]).

Fault classification models are typically trained on historical or simulated sensor data to distinguish between different failure modes (Fernandes, Corchado and Marreiros, 2022[52]). Depending on the use‑case, the goal may be to classify faults at the component level (e.g. bearing failure vs. shaft misalignment), process level (e.g. temperature deviations in a chemical reactor) or system level (e.g. degraded performance in a robotic arm). Various ML and deep-learning methods are used to support this task (Singh et al., 2023[51]).

In practice, fault classification is especially useful in settings where multiple failure types can produce overlapping symptoms (Fernandes, Corchado and Marreiros, 2022[52]). For example, in rotating machinery, an AI model may be trained to differentiate between imbalance, misalignment, lubrication failure and bearing defects based on subtle differences in vibration signatures (Fordal et al., 2023[42]). In automated assembly lines, fault classification systems can help pinpoint whether defective output is due to sensor drift, actuator failure or process deviation (Singh and Desai, 2023[53]). These distinctions are critical for ensuring that maintenance responses are both accurate and efficient.

More advanced systems also integrate fault diagnosis with real-time decision support (Ucar, Karakose and Kırımça, 2024[39]). For example, when a fault is detected, the AI system may classify the fault but also suggest possible corrections or link the issue to known root causes observed in past maintenance records (Rojek et al., 2023[40]). Nowos, a company specialising in battery maintenance and circularity, provides a practical application of such integrated diagnostics. Nowos employs AI both to support self-repair functionalities in its “Sheet Trench” battery systems and to assist engineers in real time during maintenance tasks. AI agents analyse performance data to classify battery faults and offer actionable insights, helping technicians make timely decisions. The system also provides clients with deeper behavioural and usage insights, supporting PdM that extends battery lifespan and reduces overall costs.

As with other PdM applications, the performance of fault classification models depends heavily on the availability and quality of labelled training data. Challenges include the uneven frequency of faults (with some types occurring only rarely), interference from sensor noise and the difficulty of getting precise fault labels in real-world operating conditions (Nath, Udmale and Singh, 2021[54]).

Remaining useful life prediction

In addition to identifying and classifying faults, PdM systems increasingly aim to estimate the lifespan of a component (Taşcı, Omar and Serkan, 2023[55]). This task, known as RUL prediction, is an important application of AI in manufacturing. Accurate predictions can help manufacturers reduce downtime, avoid unnecessary replacements and optimise the planning and timing of maintenance (Fordal et al., 2023[42]; Jakubowski et al., 2024[38]).

RUL models are trained on time-series sensor data to capture degradation patterns and estimate how long a component will remain operational. Modelling approaches include regression techniques, ML algorithms, and deep-learning architectures designed for sequential data (Taşcı, Omar and Serkan, 2023[55]; Ucar, Karakose and Kırımça, 2024[39]). For example, in the steel industry, deep-learning models have been trained to predict bearing wear in continuous casting machines using vibration and acoustic signals. This improves failure forecasting and enables earlier interventions (Jakubowski et al., 2024[38]).

Nevertheless, RUL prediction remains technically demanding (Wang, Zhao and Addepalli, 2020[56]). One major challenge lies in the limited availability of labelled failure data, especially in systems with high reliability where breakdowns are rare (Ucar, Karakose and Kırımça, 2024[39]). Another difficulty is generalising across equipment operating under different conditions: a model trained on one machine type or usage profile may not perform well in another context without adaptation. As a result, research increasingly focuses on developing more generalisable and data-efficient models that can perform well under real-world constraints (Raza, 2023[46]; Banjeree, Ashok and Sharma, 2024[47]).

Emerging applications of AI in predictive maintenance

While core AI applications in PdM focus on monitoring, failure prediction and diagnostics, recent research highlights several emerging directions that may shape the future of PdM in manufacturing environments. As AI technologies mature, their integration with digital twins and explainable systems is gaining momentum (Goldmann et al., 2023[57]; Mikołajewska et al., 2025[58]).

Digital twin-enabled predictive maintenance

As AI capabilities evolve, the integration of digital twins into PdM workflows is receiving growing attention (Mikołajewska et al., 2025[58]). Digital twins are virtual representations of physical systems that replicate their behaviour in real time by combining sensor inputs with simulation models. In the context of PdM, they enable continuous monitoring and prediction of equipment health under dynamic operational conditions. AI models embedded within digital twins can help identify degradation patterns, test different maintenance scenarios and estimate the impact of potential interventions without disrupting production (Mikołajewska et al., 2025[58]).

These systems hold promise in complex, high-risk environments such as aerospace, energy and heavy manufacturing, where unplanned downtime carries significant cost (Mikołajewska et al., 2025[58]). By simulating wear and failure under varying loads or usage conditions, digital twins can offer deeper insights into equipment performance than traditional monitoring tools. This allows maintenance teams to anticipate failures and act accordingly.

However, challenges remain. Effective deployment requires consistent, high-quality data streams and robust integration between physical and virtual systems (Huang et al., 2023[59]). Inaccurate or delayed sensor inputs can reduce reliability and undermine trust. In addition, the creation and maintenance of high‑fidelity digital twins demand considerable technical and computational resources. These demands may limit their scalability across diverse equipment fleets or legacy assets (Mata et al., 2025[60]).

Explainable and interpretable AI for predictive maintenance

As AI systems play a larger role in maintenance operations, a lack of transparency in their decision-making processes can present barriers to adoption (Ahmed, Jeon and Piccialli, 2022[61]). Explainable AI (XAI) techniques thus aim to make these systems more interpretable for human users. In so doing, they help teams understand how predictions are generated and increase trust in AI-supported recommendations (Scaife, 2024[62]; Ucar, Karakose and Kırımça, 2024[39]).

In the PdM context, XAI is being used to clarify which input features influence predictions or to visualise how sensor patterns relate to specific fault types (Ahmed, Jeon and Piccialli, 2022[61]). Common tools include model-agnostic techniques that break down predictions into feature contributions, enabling users to trace outputs back to their source. These explanations can support more informed decisions and facilitate collaboration between AI systems and human operators (Brito et al., 2022[63]).

Quality assurance/control

While PdM focuses on anticipating equipment failure to ensure the reliability of production machinery, quality assurance and control targets the end product. In so doing, it ensures that manufactured items meet defined standards of quality, safety and performance. Quality assurance is concerned with preventing defects through robust process design and monitoring, whereas quality control entails the detection and correction of defects through inspection and testing.

AI technologies can be used to support both quality assurance and control. In manufacturing, AI-powered QA/QC systems help automate inspection processes, detect product defects in real time and analyse upstream data to optimise process parameters before quality issues arise (Rožanec et al., 2022[64]; Singh and Desai, 2023[53]). Compared to manual inspection, these systems offer greater consistency, speed and scalability, making them particularly valuable in high-throughput sectors such as automotive, electronics and aerospace (Hütten et al., 2024[65]).

Main use-cases of AI in QA/QC

The literature identifies three main use-cases of AI in QA/QC: i) automated visual inspection; ii) defect classification and localisation; and iii) process parameter optimisation. Visual inspection systems automate the detection of surface flaws, dimensional inconsistencies or missing components directly on the production line (Lee et al., 2021[66]; Singh and Desai, 2023[53]). Classification and localisation models build on this by identifying whether a product is defective, what the defect is and where it occurs (Çallı, Albak and Öztürk, 2022[67]; Khinvasara, Ness and Shankar, 2024[68]). Meanwhile, process optimisation models aim to reduce quality issues altogether by identifying and adjusting parameters that correlate with defect occurrence (Pfrommer et al., 2018[69]; Kuric et al., 2022[70]).

Automated visual inspection

AVI is one of the most established applications of AI in quality control, particularly suited to high-volume, high-speed production environments (Hütten et al., 2024[65]). These systems rely on cameras installed along production lines to capture images of parts or products. AI models trained on annotated defect data then analyse these images for anomalies such as cracks, misalignments or missing components (Singh and Desai, 2023[53]). For example, Lee et al. (2021[66]) reported detection accuracies exceeding 99% for AI-based welding inspection systems operating under real industrial conditions.

Unlike traditional inspection methods based on hand-crafted rules or manual checks, AI-powered AVI systems learn to recognise complex visual patterns and subtle variations in shape, texture or colour. This makes them particularly effective in handling product variability and shifting production conditions (Rožanec et al., 2022[64]; Hütten et al., 2024[65]). In electronics manufacturing, for instance, AVI is used to scan printed circuit boards for soldering defects, missing components or contamination (Singh and Desai, 2023[53]). Deploying such systems at the edge (i.e. directly on or near the production equipment) further enables real-time feedback and faster response times without relying on centralised processing infrastructure (Liu et al., 2024[71]).

A key challenge in deploying these systems lies in obtaining sufficient annotated data for all relevant defect types, especially rare ones. To address this, some manufacturers use synthetic training data, semi‑supervised learning or active learning approaches where the model requests human annotation only for uncertain cases (Hütten et al., 2024[65]). For instance, the Czech company Sensorminds has developed a vision-based manufacturing execution system (MES) that monitors defects in fabric printing. Alongside real-world inspection, MES generates synthetic images and videos using camera sensor inputs to train and validate AI models offline. This approach has improved detection rates by 50% compared to manual inspection. At the same time, it enables flexible experimentation with defect types that are difficult to replicate in real production (Sensorminds, 2024[49]).

Defect classification and localisation

Once a product is flagged as defective, it is often necessary to determine the type and location of the defect to support rework, sorting or deeper process analysis. For example, in welding, AI systems trained on labelled data can classify defects into categories such as good weld, incomplete penetration, burn through, misalignment and undercut (Cardellicchio et al., 2024[72]). Classification performance further improves when combined with spatial localisation, allowing the system to indicate the location of the defect on the product surface or structure (Çallı, Albak and Öztürk, 2022[67]).

These capabilities are especially relevant in industries where certain defect types are safety-critical or process-specific. For example, in aerospace manufacturing, localised surface anomalies may indicate more serious structural risks and must be tracked carefully (Shafi et al., 2023[73]). In the food sector, Moba Group has demonstrated how AI-based classification and localisation can transform quality control and operational efficiency. Traditionally, crack detection in eggs relied on manual inspection methods such as tapping. Moba replaced this with AI-powered video analysis, enabling real-time detection of cracks and assessment of additional quality indicators like shell strength, internal composition and coloration. The system helps identify abnormalities early, reducing risk of contamination and improving food safety (Moba Group, 2025[74]).

Beyond inspection, Moba has extended AI use to other areas of egg processing. For example, its visual systems can estimate weight without physical scales, eliminating the need for regular calibration and labour-intensive daily cleaning (Moba Group, 2025[74]). Previously, a team of four workers spent hours each day on these tasks; AI automation has substantially reduced this burden. By reducing dependence on manual labour, the company is also addressing broader workforce challenges, including the ongoing shortage of skilled personnel in routine inspection and maintenance roles. To further improve precision, Moba developed a patented lighting system that highlights weak spots and structural inconsistencies. Combined with AI-based classification, this approach improves assessment reliability and reduces false positives.

Some classification systems also support root cause analysis by linking defect patterns to likely process faults (Hoffmann and Reich, 2023[13]). For example, recurring material voids may point to an upstream feeding issue, while scratches in consistent locations may suggest equipment misalignment.

Nevertheless, challenges persist. Defect classification, for example, depends highly on the quality of labelled data, the resolution of sensor inputs and the consistency of inspection conditions. These make it hard to classify defects accurately (Yang et al., 2020[75]).

Process parameter optimisation

AI can also improve quality by acting upstream, i.e. identifying which process parameters are most likely to cause defects and adjusting them before issues arise. This use-case shifts the focus from inspection to prevention, helping to stabilise production quality and reduce scrap. By analysing historical production data, AI systems can learn patterns between variables, such as machine speed, material temperature or humidity, and use these insights to predict the occurrence of defects (Rathi and Rathi, 2020[76]).

Prevention is particularly valuable in complex or sensitive processes such as plastic moulding, metal forming or additive manufacturing (Fu et al., 2022[77]). Even small deviations in temperature, pressure or mixing ratios can lead to visual flaws, weak adhesion or part deformation. AI models can be used to avoid this and alert operators when parameters begin to drift from optimal ranges (Zimmerling et al., 2022[78]). For example, Addionics applies AI to simulate and optimise battery component designs, such as cathodes and anodes, based on specific application requirements. By evaluating a wide range of process-property relationships digitally, its system identifies ideal configurations before physical production, reducing errors and enhancing consistency (Addionics, 2024[79]). The company also applies AI to control electro-deposition in metal manufacturing, where maintaining precise chemical balances is critical to product quality.

Similarly, Comau (n.d.[80]) has used AI for non-destructive weld quality analysis during battery assembly. During laser welding, high-speed thermal footage is recorded to capture plasma formation at temperatures between 600-900 °C. Although the initial footage is collected without AI, it is later processed using models trained on historical production data. These models evaluate the thermal cooling curves of each weld to assess its mechanical and electrical resistance. This enables early defect detection without damaging the product. While effective, the method remains cost intensive, requiring advanced thermal cameras and powerful computing infrastructure.

In some settings, AI tools are also leveraged to support real-time decisions by recommending adjustments before the defect occurs. For example, in continuous production lines, models can suggest modifying roller speed or cooling time when surface roughness begins to increase (Kim et al., 2022[81]).

Emerging applications of AI in quality assurance and control

While most AI applications in quality control focus on inspection, classification and process monitoring, recent research highlights two emerging use-cases that expand the role of AI beyond detection of defects. These developments respond to growing industry demands for greater transparency, regulatory compliance and robust data governance. Specifically, they include the integration of XAI into inspection systems and the use of AI to support documentation and compliance.

Explainable and interpretable AI for QA/QC

As AI models are increasingly used to make decisions about product quality, questions arise about how those decisions are made, and whether human operators can understand and trust them. In many cases, particularly in visual inspection, AI systems may identify defects that are not immediately obvious to the human eye. This can lead to hesitation or uncertainty about whether to act on the model’s recommendation (Rožanec et al., 2022[64]).

XAI tools address this challenge by helping operators understand what the model has seen or why it flagged a product. For example, XAI can highlight the regions of an image that influenced a defect decision or rank the input variables that most strongly contributed to a classification (Hoffmann and Reich, 2023[13]). One common method is SHapley Additive exPlanations (SHAP), which quantifies the contribution of each input feature (e.g. pressure, temperature or image zone) to the model’s output. Senoner et al. (2022[82]) show that using SHAP helped engineers in semiconductor production uncover which process parameters were driving quality losses. This resulted in targeted improvements and a 21% yield increase.

Comau’s use of AI in robotic quality assurance highlights the importance of explainability in operational settings. Even though its AI models reach over 90% accuracy, unexpected errors – known as “hallucinations” – can still occur (Comau, n.d.[80]). In one case, a robot mistakenly pierced a package during a depalletising task because the AI misinterpreted the object. To reduce such risks, Comau introduced a deterministic monitoring system that checks and validates the output of AI before action is taken. Comau applies AI in novelty detection to define mathematical equations and constants that ensure consistent robot trajectories, which are crucial for maintaining uniform product properties. This approach both helps prevent unpredictable robot behaviour and simplifies the complex task of robot configuration and fine-tuning. As a result, operations become more streamlined, allowing workers to focus less on repetitive adjustments and more on higher-value, task-oriented activities.

AI-supported documentation and reporting

Another emerging application lies in automating documentation and reporting tasks in quality management. In many factories, inspection outcomes or defect classifications must still be recorded manually through data entry, image tagging or written summaries. AI systems can simplify this work by automatically logging decisions, generating labelled images and structuring inspection results for later review (Khinvasara, Ness and Shankar, 2024[68]).

Natural language processing (NLP) is increasingly being used to generate short descriptions of defects or quality events, reducing the reporting burden for frontline operators (Javaid, Haleem and Singh, 2023[83]). By automating this documentation process, AI can both streamline reporting and enhance consistency across production lines, ensuring that quality records are captured in real time (Singh, 2024[84]).

Supply chain optimisation

While PdM focuses on maintaining equipment reliability and quality control targets product conformity, supply chain optimisation concerns the broader co‑ordination of materials, information and resources across manufacturing networks. AI plays a growing role in this domain by enabling real-time insights, predictive analytics and dynamic decision making across procurement, production, logistics and distribution processes (Grover, 2025[85]; Odumbo and Nimma, 2025[86]). It can help manufacturers improve visibility, adapt to disruptions and synchronise flows across global value chains, making supply systems more resilient, cost efficient and responsive to change (Eyo-Udo, 2024[87]; Riad, Naimi and Okar, 2024[88]).

Main use-cases of AI in supply chain optimisation

Academic literature identifies three core applications of AI in supply chain optimisation: i) demand forecasting and inventory management; ii) procurement and supplier relationship management; and iii) logistics and transportation optimisation. Demand forecasting and inventory models can enhance forecasting accuracy and allow for dynamic inventory control by learning from historical data, real-time demand signals and external variables (Aktepe, Yanık and Ersöz, 2021[89]; Olorunyomi et al., 2024[90]). In procurement, AI can be used to automate transactional workflows, evaluate supplier risk and support strategic sourcing based on performance, cost and reliability indicators (Allal-Chérif, Simón-Moya and Ballester, 2021[91]; Riad, Naimi and Okar, 2024[88]). Meanwhile, logistics and transport applications focus on improving route planning and fleet co‑ordination through AI-powered tools that analyse operational data and adapt to real-time conditions (Eyo-Udo, 2024[87]; Quoc Khoa et al., 2024[92]).

Demand forecasting and inventory management

Demand forecasting and inventory management are among the most common applications of AI in supply chain optimisation (Quoc Khoa et al., 2024[92]). AI models can improve forecasting accuracy by analysing large volumes of structured and unstructured data (e.g. historical sales, seasonal patterns, promotions and weather forecasts) to identify demand trends at a granular level (Aktepe, Yanık and Ersöz, 2021[89]; Khadem, Khadem and Khadem, 2023[93]). This allows manufacturers to better anticipate customer needs, avoid stockouts and reduce costly overstocks.

In contrast to rule-based or statistical forecasting methods, AI models continuously learn and adapt as new data become available, helping to generate rolling forecasts that account for dynamic market conditions (Islam et al., 2024[94]). These capabilities are particularly valuable in volatile or highly seasonal markets, where traditional methods often fail to capture unexpected shifts in demand (Jones, 2025[95]).

On the inventory side, AI can support dynamic stock management by updating safety stock thresholds, re‑order points, and replenishment schedules based on real-time inputs such as lead time variability and demand volatility (Albayrak Ünal, Erkayman and Usanmaz, 2023[96]; Odumbo and Nimma, 2025[86]). For example, reinforcement learning approaches have been applied to identify optimal inventory policies that minimise costs while maintaining service levels across distribution centres (Albayrak Ünal, Erkayman and Usanmaz, 2023[96]).

Phantasma Labs (2025[97]), a German AI provider offering fully AI-driven enterprise resource planning (ERP) systems, provides a practical example of AI-enhanced responsiveness. Once implemented, these systems enable manufacturers to optimise planning and operations by dynamically reallocating resources in response to disruptions. For instance, in the event of equipment breakdowns, the system can identify alternative machinery, reschedule production and minimise downtime. Phantasma’s solution also supports strategic workforce allocation, allowing enterprises to maintain operations with as little as half of their workforce by prioritising key customers and adjusting plans in real time.

In other notable applications, Starbucks uses AI models to align inventory and staffing decisions with weather conditions and customer traffic patterns at individual locations (Nweje and Taiwo, 2025[98]). Similarly, research has shown that manufacturing enterprises that adopted AI-based demand forecasting tools during COVID-19 adapted quickly to demand spikes and manage supplier constraints (Jones, 2025[95]).

Despite these benefits, implementation challenges remain. Many enterprises struggle with siloed or incomplete data, poor integration between planning systems or organisational resistance to automated decision making (Nweje and Taiwo, 2025[98]).

Procurement and supplier relationship management

Procurement and supplier relationship management are critical functions for ensuring cost efficiency, reliability and agility in supply chains. AI technologies can be used to automate procurement processes, evaluate supplier performance and improve sourcing decisions. Compared to traditional approaches, AI allows for more proactive and data-driven procurement strategies, helping manufacturers respond to supply risks and shifting market dynamics (Dubey et al., 2021[99]; Odumbo and Nimma, 2025[86]).

One core application is the automation of routine procurement tasks, such as purchase order generation, invoice processing and contract review. NLP and robotic process automation enable systems to extract and structure information from contracts, e‑mails and supplier databases, streamlining workflows and reducing administrative overhead (Guida et al., 2023[100]). AI is also used to assess supplier performance and risk. Models analyse indicators such as delivery timeliness, price stability and defect rates, often combined with external data like logistics disruptions or financial news. In this way, it can help flag underperformance or predict issues (Guida et al., 2023[100]).

In strategic sourcing, AI tools can support supplier selection by matching buyer needs with suppliers’ capabilities. These systems use pattern recognition, semantic search and performance data to identify alternative or backup suppliers, evaluate long-term fit and support diversification. During disruptions (e.g. pandemics, trade wars), such tools can be useful in helping enterprises identify viable alternatives and adjust sourcing strategies in near real time (Glory, 2023[101]; Guida et al., 2023[100]). Some systems also support ongoing relationship monitoring, identifying opportunities for co-development or early signs of disengagement based on contract compliance, communication patterns and collaboration history (Allal-Chérif, Simón-Moya and Ballester, 2021[91]).

Logistics and transportation optimisation

AI can play a strategic role in optimising the mid- and upstream logistics operations that underpin manufacturing supply chains. These operations include the co‑ordination of regional, national or international transport flows between suppliers, production sites and warehouses (Nweje and Taiwo, 2025[98]). Ensuring smooth and efficient operations at this scale is critical for maintaining service levels, reducing operational costs and navigating the growing complexity of globalised production networks. AI systems can be used to analyse large volumes of operational, geographic and market data to enhance fleet management, route design and delivery planning (Odumbo and Nimma, 2025[86]).

A central application is transport route optimisation across regional or international corridors. Here, ML and heuristic algorithms process structured and unstructured data (e.g. traffic, fuel prices, delivery windows and route availability) to determine the most efficient paths for multi-vehicle, multi-stop deliveries (Islam et al., 2024[94]; Nweje and Taiwo, 2025[98]). This helps manufacturers and logistics providers to lower fuel consumption, increase punctuality and mitigate delays caused by disruptions or capacity constraints (Odumbo and Nimma, 2025[86]).

Emerging applications of AI in supply chain optimisation

While core applications of AI in supply chain management focus on planning, sourcing and logistics co‑ordination, recent advancements are extending into domains characterised by environmental impact and strategic sustainability goals. One important area is the use of AI to enable more sustainable and circular supply chains.

AI for sustainable supply chain optimisation

AI is increasingly recognised as an enabler of more environmentally and socially sustainable supply chains (Pal, 2023[102]). Across manufacturing networks, AI models can be used to support greener sourcing, reduce emissions, minimise waste and enable circular economy practices (Naz et al., 2022[103]). These applications go beyond mere efficiency gains: they also support strategic sustainability goals by embedding environmental criteria into planning, procurement and operational decision making.

One use-case is the application of AI to optimise transportation and production processes with sustainability metrics in mind (Feng, K-H and Zhu, 2022[104]). For example, AI-powered route optimisation can reduce fuel consumption and greenhouse gas emissions by identifying the most efficient delivery paths. Meanwhile, intelligent scheduling can align production batches to reduce energy peaks and idle (Odumbo and Nimma, 2025[86]). Furthermore, AI could facilitate greener procurement and supplier management. Algorithms can incorporate carbon footprint data, environmental certifications or compliance histories into supplier evaluation, supporting more sustainable sourcing decisions (Guida et al., 2023[100]).

A notable example is Elio.Earth (2025[105]), an Austrian company that leverages AI to enhance sustainability and traceability in manufacturing. Elio developed a proprietary large language model that could extract and interpret information from unstructured documents and operational datasets. Combined with AI agents, the company’s solution enables the retrieval of emissions-related data across manufacturing sites and transportation pathways. By tracking energy use and emissions at the production level and mapping transport impacts across different supplier sources, Elio’s tools allow manufacturers to distinguish between products by sustainability profile, supporting greener sourcing and compliance with environmental targets.

Insights from interviews

Current AI use-cases

AI adoption in the manufacturing sector remains at an early and evolving stage in the European Union. Manufacturing is regarded as one of the most conservative sectors in AI adoption, progressing at a significantly slower pace than sectors such as fintech or healthcare. The sector largely remains in an exploratory phase, evaluating where AI integration might add value rather than pursuing widespread deployment. Suppliers noted that organisations offering services are more proactive, while traditional product-based manufacturers often lag due to a lack of implementation plans or internal digital capabilities.

Despite slow AI uptake in the sector, interviewees noted that the immediate advantages of AI in manufacturing may be considerable. Seemingly small marginal improvements, such as a 1-2% increase in defect detection accuracy through AI, can deliver high returns in high-volume settings. For example, one automation firm noted that achieving a 3% improvement in quality control could nearly double operational profits.

One interviewee pointed out that although the European Union missed the first wave of digitalisation, it remains a global leader in high-value industrial manufacturing. They emphasised that its strength lies in its deep engineering expertise; long-standing production know-how; and a vast installed base of machinery across sectors like mobility, health and advanced manufacturing. The main constraint is a highly fragmented landscape of SMEs and largely brownfield infrastructure, which limits digital connectivity and slows AI deployment. Nevertheless, the interviewee considered the availability of GenAI and large language models as a powerful tool for manufacturers to extract value from previously unusable unstructured data. This, in turn, could unlock a unique opportunity to digitise legacy knowledge and expertise (see “Outlook”).

As one interviewee described, the manufacturing sector can be broadly divided into three categories of enterprises in their approach to AI. First, early movers are typically large, innovation-driven manufacturers that possess strong R&D capacity. They are actively shaping initiatives like Manufacturing X, a government-backed effort to establish an EU standard for industrial data ecosystems. Second, fast followers include medium-sized, often family-owned equipment suppliers (e.g. those producing robotic grippers or suction units) that adapt quickly to the standards set by early movers but tend to lack in-house AI expertise. Finally, the lagging majority consists largely of SMEs that either have yet to recognise the potential of AI or lack the skilled workforce to act on it. The interviewee emphasised that while manufacturing is supported by a handful of major players, it depends heavily on a long tail of niche enterprises delivering highly specialised machinery. “This is why digitalisation in manufacturing is so complex,” they noted, “and why EU and national-level investment is critical. These smaller enterprises are essential but cannot build this infrastructure alone.”

Several interviewees – whose use-cases are described in the previous section – reported both internal AI solutions and those offered to other manufacturing enterprises. These applications mostly focus on quality control and assurance, PdM and robot control, while supply chain optimisation remains less commonly addressed. One large automation company, for example, employs AI-enhanced vision systems and robotic guidance for tasks like picking, sorting and depalletisation. AI is also used in battery quality inspection, analysing the mechanical and electrical resistance of welds. Such systems were reported to be transforming production processes in manufacturing operations. Their high costs limit broader accessibility but demonstrate how AI is transforming production in capital-intensive settings.

Another major industrial technology provider both uses and markets a broad portfolio of AI solutions, spanning PdM, robotics, vision systems, anomaly detection and engineering design, as well as tailored vertical applications. These solutions are underpinned by a combination of traditional ML and deep‑learning techniques. The company also applies AI for tasks like design space exploration, reducing simulation time from 48 to 24 hours. Their customers tend to be other large industrial operators that value reliability, scale and regulatory compliance.

A large machinery manufacturer has also implemented AI in PdM, quality control and supply chain management. The firm emphasised that AI adoption did not lead to workforce reductions but instead enabled a restructuring of teams around higher-value tasks such as data analytics. Its AI systems enhance accuracy of defect detection and accelerate resolution of issues, while democratising technical expertise across the organisation. As the CEO described, “AI allows us to replace the retiring mechanical expert with a data analyst, one expert’s knowledge, refined over decades, can now be embedded in a system and scaled across our entire workforce.”

These large enterprises typically serve industrial clients with global reach and complex systems. While they benefit from rich datasets and internal IT capabilities, they continue to face high costs, workforce retraining needs and regulatory complexity, particularly in relation to the EU AI Act and GDPR frameworks. Notably, they have emerged as key advocates for building EU-wide data ecosystems, arguing that data sharing across value chains will reduce duplication, enable better co‑ordination and unlock productivity gains.

Similarly, medium-size manufacturers offer AI solutions for quality control, design optimisation and automation. One global food equipment supplier provides AI for robotic automation, PdM and camera-based quality control. A software development firm with a focus on manufacturing offers GenAI platforms for ERP integration, chatbots and on-site assistant tools, helping mid-sized manufacturers digitise unstructured data and automate internal processes. About half of their clients have dedicated IT teams, while the other half depend on outsourced AI services. These enterprises are motivated by efficiency gains but tend to adopt AI only after seeing concrete proof-of-concept results.

Medium-sized manufacturers reported that their customers – operating in sectors like food processing, machinery and specialised industrial goods – are increasingly pressured to modernise, both due to competition and rising labour costs. In this context, they see AI as a tool to reduce costs, face labour shortages, and increase sustainability and resilience.

For instance, a PdM provider works with SMEs using sensor data, synthetic data generation and visual inspection systems. Its clients often lack digital infrastructure but are increasingly eager to adopt AI as competitive pressures mount. The company assists in everything from sensor placement to cloud set-up, offering clients real-time dashboards and AI agents capable of autonomous machine intervention. Clients report improvements of up to 40% in operational efficiency through AI-driven monitoring and defect detection.

A European battery refurbishment company has gradually introduced AI to support maintenance diagnostics and ERP integration. It aims to analyse hundreds of data points per battery cell, enabling better preventive action and customer insights. However, slow decision-making cycles and regulatory complexity (e.g. GDPR) were reported as hindering faster AI adoption among enterprises of a similar size.

Customers in this segment typically seek tangible, low-risk gains, often prioritising tools that can enhance uptime, reduce waste or offset labour constraints. These enterprises are most constrained by costs, lack of internal AI expertise and data fragmentation. SMEs frequently cite data scarcity as a key barrier.

Outlook

While AI adoption often targets small-scale improvements, most benefits will be in the long term. Several interviewees, particularly from large enterprises, stressed that AI adoption in manufacturing often targets incremental improvements, such as in quality control, maintenance or process optimisation. However, they also noted that the most significant long-term competitive advantages are expected to stem from scalable and strategically transformative applications, including GenAI, digital twins and advanced supply chain analytics. Thus, EU public authorities and industry associations must continue to monitor examples of such high-potential applications of AI in manufacturing, improve understanding on drivers for their adoption and build on the resulting knowledge to generate demonstration effects that help promote broader uptake.

Crucially, GenAI offers a way to transform AI from a project-based effort into a scalable product-based capability, which has long been a bottleneck in industrial settings. One interviewee noted that traditional industrial AI projects often struggle with replication due to inconsistent machine data and varying operational environments. GenAI helps overcome this by leveraging pre-trained models that reduce the need for repetitive customisation. When tuned with domain-specific data, GenAI enables much faster deployment across different factory settings, lowering integration costs and dramatically shortening return on investment cycles. This was considered especially valuable in sectors like the German machine tool industry, where profit margins are tight and project-based AI rollouts have proven economically unviable. Another interviewee echoed this position, noting the first wave of AI was characterised by custom model development, long timelines, high cost and the need for specialised talent. On the other hand, the second wave – characterised by generalised, open-source models – allows for faster, cheaper implementation. They noted that many enterprises overestimate the difficulty and cost of AI today, not realising the availability of modular, easy-to-use solutions.

One of the most transformative impacts of AI lies upstream at the design stage, where decisions determine up to 80% of a product’s sustainability footprint and production cost. AI-powered design tools are already enabling engineers to simulate, test and optimise thousands of design variants digitally before any physical resources are committed. This approach is seen as critical to improving resource efficiency, supporting circular economy goals, and reducing production errors downstream.

GenAI is a pivotal enabler in this shift. By applying large language and multi-modal models to computer-aided design files, design simulations and engineering documentation, GenAI helps engineers rapidly explore design space, accelerate simulation cycles and avoid redundant mistakes. Over time, these tools can drastically shorten time to market while improving product performance.

Another interviewee emphasised how GenAI allows enterprises to unlock the value of unstructured data, such as historical service reports, quality inspection images and handwritten engineering notes, that were previously inaccessible to traditional analytics. When tuned with company-specific data, these models can replicate the decision-making logic of experienced workers, including those approaching retirement. This can help preserve institutional knowledge and make it accessible across the organisation, addressing labour shortages and skill gaps. For the interviewee, GenAI is both an automation tool and a mechanism for democratising expertise, enabling less experienced workers to perform advanced tasks with AI support. This shift is seen as vital in light of Europe’s ageing workforce and the declining attractiveness of factory work. AI-guided copilots and assistant tools help new hires operate machinery, interpret diagnostics and troubleshoot issues with little prior experience, improving safety, productivity and workforce retention.

Another company reinforced this point by citing the case of an electricity transmission firm that used GenAI to significantly ease the workload of its field engineers. The engineers turned unstructured data, such as photos, manuals, reports and documents, into searchable, structured formats. In so doing, they could quickly find information with simple voice or text commands through a copilot-enabled assistant. This helped streamline field operations and improve overall efficiency. As a result, the team working on these tasks was reduced from 60 to 35 people, allowing the company to reassign staff to other areas. The company also used GenAI to automate chat features on its website, making them faster and more useful than previous systems. These AI-powered chats were connected to the company’s ERP system through application procedure interfaces (APIs), allowing them to pull relevant data instantly and answer queries more effectively.

Beyond operations, GenAI is expected to revolutionise how manufacturers approach data sharing and ecosystem collaboration. Traditionally, industrial “AI has scaled like a project, not a product”, due to heterogeneous machine data, fragmented systems and siloed operations. GenAI could overcome this by reducing the need for case-by-case data labelling and model training. With pre-trained models adapted to industrial domains, deployment can occur faster and at lower cost. This opens the door to AI at scale even for small- and mid-sized manufacturers. However, realising this potential depends on the development of industrial-grade models, with semantics and modalities tailored to machine data, not just general language.

Overall, several interviewees, especially large enterprises, considered GenAI as a pathway to scale AI adoption by turning legacy knowledge into actionable insight, reducing implementation time and democratising access to expert decision making. In their view, if properly deployed within supportive policy and infrastructure frameworks, GenAI could transform AI adoption in manufacturing from incremental to exponential, securing EU manufacturing competitiveness in the next industrial era.

Key barriers and challenges

Enterprises interviewed in this study highlighted several challenges and barriers that limit further adoption of AI in the manufacturing sector.

Data availability and quality

• Enterprises face systemic challenges in accessing clean, complete and usable internal data. Manufacturing enterprises often lack centralised data repositories. Operational data (e.g. temperature, vibration or machine status) are scattered across different systems, stored in incompatible formats or embedded in free-text maintenance logs. Many SMEs still rely on pen-and-paper documentation or outdated Excel spreadsheets. Moreover, many SMEs do not rely on data analysis to manage their business (BPI France, 2025[36]). Under such circumstances, building a reliable data pipeline for training AI models can be difficult. Even in large enterprises, ERP systems and production monitoring tools are not always integrated. This leads to data silos between departments like maintenance, quality and procurement.

• The high cost of data labelling and preprocessing limits the use of supervised learning. AI applications that rely on supervised learning (e.g. defect detection or PdM) require extensive human effort to label data. For example, identifying defective products from image data often requires expert technicians to classify thousands of images manually. Additionally, raw sensor data must be cleaned, normalised and transformed to remove noise and align time-series data from different sources.

• External data are fragmented and difficult to access due to low interoperability and confidentiality concerns. Customer data (e.g. preferences, usage feedback) are often scattered across different e‑commerce systems, Customer Relationship Management software or external platforms, complicating integration. Supplier data (e.g. delivery times, quality metrics and sourcing risks) are typically treated as proprietary and not shared in machine-readable formats.

• Data-sharing reluctance, internally and externally, constrains AI integration. Internally, departments may treat data as a performance asset and hesitate to share them, particularly when key performance indicators are linked to their control. Externally, suppliers and customers often resist sharing operational data due to concerns about intellectual property, commercial leverage or reputational risk. This undermines multi-tier AI use-cases such as predictive supply chain co‑ordination or joint quality improvement. Some enterprises expressed a need for clearer legal guidance and trust-building mechanisms to unlock cross-organisational data value.

• New entrants lack historical data and struggle to build AI models from scratch. PdM models, for instance, depend on long-term operational histories, including breakdowns, failure causes and associated process parameters. Without this, enterprises cannot train robust models. This creates a competitive advantage for well-established enterprises, reinforcing barriers to entry and limiting innovation diffusion.

• Synthetic data help bridge gaps but are computationally intensive and must be carefully validated. Several enterprises are generating synthetic sensor readings or images to simulate rare failure events or edge cases in production. However, this requires significant compute capacity and expert input to ensure fidelity. Unless continuously fine-tuned, models trained on synthetic data risk failing to perform in real-world conditions.

Infrastructure and compute capacity

• Many manufacturing sites lack foundational digital infrastructure for AI deployment. The basic precondition for AI adoption – i.e. real-time machine connectivity – is often missing. Older machines do not support sensor integration or communication protocols and retrofitting them is costly and labour-intensive. According to interviewees, digital twins remain largely aspirational, with most enterprises limited to partial simulations of isolated equipment rather than full production line replicas.

• Sensor reliability and deployment challenges undermine data integrity. Enterprises report frequent issues with sensor calibration, drift and signal interference, especially in high-heat, high-vibration environments. For instance, installing vision-based systems in dusty or reflective settings introduces noise that degrades image quality. Moreover, sensor placement requires careful adaptation to the characteristics of each process step; a set-up that performs well in one assembly line may not apply in another.

• Limited access to cloud and high-performance computing (HPC) resources impedes training and scaling. AI model training – especially for deep learning or reinforcement learning – requires graphic processing units, large memory capacity and efficient parallelisation. Many enterprises lack the internal IT infrastructure and cannot afford third-party HPC resources. Others face legal or contractual limitations around cloud storage, particularly for proprietary or safety-critical data.

Financial barriers to AI adoption

• AI deployment often entails significant upfront investment in equipment, infrastructure and expertise. For instance, implementing an AI-based quality inspection system may require costly infrared and thermal cameras, specialised lighting setups, extensive data acquisition and ongoing monitoring. These capital-intensive requirements can be prohibitive, particularly for smaller manufacturers.

• Funding instruments are misaligned with industrial timelines. Public funding programmes may take 6-12 months from application to disbursement. Interviewees stressed that this timeline is not aligned with product cycles and investment planning of SMEs. Moreover, application processes require detailed technical proposals, cost breakdowns and co-financing commitments. These can exceed the capacities of smaller enterprises that lack a dedicated grant-writing team.

• Excessive administrative requirements can discourage participation in public funding programmes. One AI provider declined to reapply for EU funding after securing initial support, citing hundreds of pages of reporting obligations and minimal flexibility in implementation. Several others mentioned they had missed deadlines or eligibility windows due to unclear processes or limited internal capacity.

• Recurring costs of commercial models create lock-in and threaten long-term affordability. Subscriptions to cloud-based AI services (e.g. computer vision APIs, foundational models) incur ongoing fees that scale with usage. Enterprises are concerned that these costs will escalate as adoption deepens. Some hesitate to build core business functions around proprietary AI models.

• Investors often underestimate the maturity gap in industrial AI. Many investors aim for quick returns, but manufacturing AI projects can sometimes require two to three years to reach reliable and scalable performance. This leads to underfunding or premature termination of promising initiatives. Firms noted that investor expectations are often shaped by consumer tech models, which do not map well onto industrial contexts.

• Off-the-shelf AI systems rarely transfer across factories without heavy customisation. Differences in machine configurations, operator behaviour and product tolerances mean that AI models developed for one site cannot simply be copied to another. Each deployment often requires retraining, new data ingestion pipelines and adapted feedback interfaces. SMEs find it especially difficult to justify repeated investments when scaling AI across production sites.

Regulatory complexity, bureaucracy and fragmentation of the EU market

• Legal requirements are sometimes perceived as overly complex and ambiguous, posing challenges for implementation. Several interviewees noted potential difficulties to accurately define the scope of “high-risk AI” under the EU AI Act, particularly for emerging or borderline use-cases in manufacturing. GDPR compliance, especially in cross-border data flows, adds significant documentation and auditing requirements. In this regard, enterprises without in-house legal expertise, particularly smaller ones, often feel at a disadvantage.

• Legal and administrative fragmentation hampers cross-border scale-up. Expanding AI-enabled manufacturing operations across the European Union means navigating different tax regimes, business registration rules and sectoral regulations. For SMEs, this introduces costly legal compliance burdens. Enterprises often delay or abandon EU-wide rollouts due to the complexity of aligning operations across jurisdictions.

• Language and documentation barriers create inefficiencies. AI documentation, manuals and compliance paperwork must be translated for each national market. Real-time interfaces, dashboards and user prompts also require multilingual support. For enterprises without dedicated localisation teams, this increases both the cost and time required for AI deployment.

• Support mechanisms vary significantly by Member State. Interviewees stressed that some countries offer strong public support for AI adoption (e.g. digital innovation hubs, tax incentives, and collaborative R&D networks), but others provide little beyond general digitalisation strategies. As a result, enterprises in “support-rich” countries are better positioned to adopt AI, exacerbating regional inequalities within the single market.

• Industrial policy co‑ordination lags global competitors. Compared to more centralised approaches in other countries, the decentralised industrial policy in the European Union fragments funding streams, research agendas and vendor ecosystems. Enterprises noted that European companies often compete against well-coordinated foreign competitors backed by substantial government support.

• Lack of EU-level standards limits interoperability. Efforts like Open Platform Communications Unified Architecture (OPC UA) are seen as steps in the right direction, but many enterprises still struggle with incompatible machine interfaces, data schemas and reporting standards across borders. Without stronger enforcement or incentives for common standards, AI adoption remains piecemeal and context specific.

Workforce and organisational barriers

• The EU AI talent pool is unevenly distributed, creating regional disparities. Interviewees reported that countries like Germany (and neighbouring Switzerland in the Schengen Area), and the Netherlands, attract and retain top AI talent due to higher salaries, stronger research ecosystems and greater industrial demand. In contrast, enterprises in southern and eastern Europe face difficulties attracting skilled professionals and experience frequent talent outflow to the United States or the United Kingdom. This brain drain undermines local AI ecosystems and weakens long-term competitiveness.

• The retirement of experienced staff risks eroding critical domain knowledge, which remains tacit and is rarely digitised. Enterprises raised concerns about the retirement of long-serving operators and maintenance personnel who hold deep, experience-based knowledge of production systems. Without systematic efforts to capture and digitise this expertise, AI models cannot be trained effectively or adjusted for real-world variance. This loss is especially pronounced in small enterprises without formal documentation practices.

• Concerns over job security reduce employee buy-in for AI projects. Many interviewees noted that AI initiatives encounter passive resistance from employees who worry about job loss. Even when enterprises position AI as a tool to support – not replace – workers, the fear of automation often leads to reduced co‑operation during implementation. For example, operators may be less willing to share insights or validate data if they believe this could render their roles obsolete.

• Managerial scepticism and inertia delay adoption. Senior decision makers, especially in traditional industrial enterprises, often rely on heuristics and personal experience rather than data-driven approaches. When AI suggestions contradict their intuition, these insights are dismissed. Without internal champions or exposure to successful pilots, AI remains relegated to innovation units without broader buy-in.

• Generational divides limit collaboration across teams. According to interviewees, younger employees are generally more open to experimenting with AI tools and data visualisation platforms. Older employees bring indispensable process knowledge but are often less comfortable with digital interfaces. Enterprises described challenges in creating “mixed” teams that bridge this divide effectively, especially when younger staff lack credibility in the eyes of senior technicians.

• Trust in AI-generated decisions is still evolving. Some operators and managers struggle to accept AI outputs if they cannot fully understand the model’s logic. For example, when an AI system suggests reducing the speed of a machine to prevent defects, human supervisors may override it unless clear explanations are provided. Enterprises stressed the importance of building trust through transparency, performance benchmarking and operator-in-the-loop systems.

• Enterprises are sometimes sceptical of academic research, viewing it as misaligned with the practical demands of manufacturing. Interviewees emphasised that much academic research in AI lacks the operational robustness needed for deployment in manufacturing environments. Pilot projects often fail to move beyond proof-of-concept due to issues with scalability, maintainability or misalignment with business processes. This disconnect – exacerbated by legacy failures of earlier AI projects in industrial settings – contributes to scepticism and risk aversion.

Key recommendations to enhance AI uptake in manufacturing in the European Union

The analysis of AI uptake in the manufacturing sector, together with the literature review for key use-cases and industry interviews, suggest the European Union should consider policy actions to help increase AI adoption in manufacturing enterprises. These are described in this section.

Strengthen data sharing and security

Data availability and quality are a major challenge limiting the scope for AI adoption in the manufacturing sector. Enterprises find it difficult and expensive to access clean, complete and usable internal data and are reluctant to share data often considered as proprietary. Strengthening and expanding on existing data-sharing mechanisms while addressing data security concerns and supporting data access management, would help EU manufacturing enterprises overcome some of the data challenges. In parallel, EU manufacturing enterprises should continue to be supported in their digitalisation efforts, including those of smaller size or in more traditional industries that might face additional difficulties in leveraging data to manage and expand their activities.

Foster data sharing within and across sectors across the European Union: the European Union already provides manufacturing enterprises with data-sharing mechanisms that, according to interviewees, have successfully enhanced data sharing. For example, the Manufacturing X data space helps create a data ecosystem for manufacturing enterprises, where companies share data according to predefined rules and standards. Catena‑X is a similar mechanism dedicated to data sharing among enterprises in the automotive sector. It seems to have been particularly successful in incentivising supply chain data sharing among SMEs. These data-sharing mechanisms could also help the industry with better operational and machine performance data, PdM and sensor-generated data, quality control and inspection data, computer-aided design files, engineering documentation and simulation results.

The European Union should therefore leverage data-sharing mechanisms and consider extending and tailoring them, alongside the predefined rules and standards, to specific industries. It should prioritise those with the greatest potential for scaled AI adoption and resulting productivity gains. Limited knowledge of and understanding about data-sharing mechanisms among manufacturing enterprises, including how secure data are, might be limiting the potential for these solutions. The European Union should therefore consider increasing the awareness and incentivise enterprises to join these initiatives. To that end, it should demonstrate benefits, while mitigating safety concerns in dedicated industry fora. Identifying successful adoption examples could help showcase the value added of AI adoption to reluctant enterprises in respective industries.

Ensure secure and simple data access management: addressing enterprises’ concerns about sharing data that are often considered proprietary will require ensuring secure and GDPR-compliant data storage and protection. It will be equally important to develop a simplified but trustworthy procedure for enterprises and relevant stakeholders to access data (including anonymised sensitive data), while ensuring that enterprises are equipped with the necessary skills to access and manage data (see below).

Enterprises and industry associations play a key role, including through industry consortia to tackle data-sharing challenges and identify best practices. Dedicated training programmes to ensure that manufacturing enterprises have the required skills (see below) and developing standards and certification for AI use in manufacturing could further incentivise manufacturing enterprises to adopt AI.

Enhance infrastructure and AI capabilities

Manufacturing enterprises have identified infrastructure and compute capacity as a challenge. Many manufacturing sites still lack foundational digital infrastructure for AI deployment, including real-time machine connectivity. Legacy machinery and equipment, challenges with sensor reliability and practical deployment and the costs of upgrading manufacturing sites to meet AI needs mean that digital twins remain largely aspirational. Moreover, training and scaling AI are also limited by access to cloud and HPC resources, including both in terms of costs and contractual limitations around cloud storage for proprietary data.

Address the specific AI infrastructure needs of the EU manufacturing sector: as part of renewed efforts to help develop dynamic AI ecosystems by bringing together compute, data and talent (e.g. AI Factories), the European Union should help address the advanced simulation data, storage and processing needs of manufacturing enterprises. Support should be provided to establish dedicated AI compute and development centres with HPC capabilities focused on simulation data for manufacturing enterprises. This would further enhance EU AI manufacturing capabilities and help advance AI-driven innovation and optimisation in industrial applications.

Strengthen domestic AI ecosystems: support for further enhancing industry-academia research collaboration and strategic institutional partnerships would nurture the EU manufacturing AI ecosystem. Pilot projects often fail to move beyond proof-of-concept due to issues with scalability, maintainability or misalignment with business processes. Bridging the gap between academic AI research and practical enterprise applications would accelerate AI deployment in manufacturing. The European Union should enhance its support for research institutions, cross-border collaboration and industry partnerships. To that end, it should strengthen programmes for innovative AI solutions that align with real-time industrial needs for boosting manufacturing productivity. Enterprises note that the bureaucratic processes and effort required to obtain public funds entail a significant cost incommensurate with the financial support provided. The European Union should consider whether there is scope for streamlining the application processes for start-ups and small enterprises to access public support, while continuing to ensure good due diligence.

Help enterprises navigate policies and regulatory frameworks

Regulatory complexity, bureaucracy and fragmentation of the EU market entail challenges to the implementation of AI in manufacturing. Legal requirements are sometimes perceived as overly complex and ambiguous. This includes excessive administrative requirements to participate in public support programmes, whose requirements and extent of support also vary significantly across Member States.

Despite efforts like OPC UA providing steps in the right direction, EU-level standards are needed to enhance interoperability and avoid piecemeal and context-specific AI adoption. Moreover, legal and administrative fragmentation hampers cross-border scale-up as enterprises need to navigate different tax regimes, business registration rules and sectoral regulations. These are compounded by costs associated with language and documentation challenges associated with both compliance and the practical implementation of AI in manufacturing (e.g. AI documentation, manuals, monitoring dashboards).

Develop practical guidelines and standards: helping enterprises understand the legal framework for AI adoption and providing practical support through, for example, practical guidelines and standards, would help manufacturing enterprises navigate implementation concerns. This also includes clarifying the applicability of the AI Act and related regulation with clear, hands-on support through specific manufacturing case studies and training.

Enterprises often feel pressured to adopt AI without a clear, industry-specific implementation plan, leading to isolated and suboptimal AI deployment. Identifying and highlighting industry best practices alongside the development of harmonised standards for safe and responsible AI adoption for enterprises in manufacturing would also be important.

Reduce cumulative regulatory burden and fragmentation: there is scope to streamline and simplify compliance across regulations. This includes, for example, providing pathways for compliance across the GDPR and AI regulations, with consolidated guidance documents and process alignment. The EU AI Board could play an important co‑ordinating role, ensuring consistent AI regulation enforcement and minimising regulatory fragmentation. Additionally, policymakers should consider the impact of other regulations, including for example whether there is scope to facilitate greater industrial AI collaboration, while ensuring market competition.

Help ensure a global level-playing field: enterprises note concerns with unfair commercial practices and lax intellectual property enforcement by other jurisdictions. The European Union should continue working with like-minded partners towards a global level-playing field that helps prevent unfair practices.

Strengthen co‑ordination between AI and industrial strategies and policies: revision of the EU Coordinated Plan on AI would benefit from better co‑ordination with the EU industrial strategy and relevant industrial policies. Close collaboration between responsible directors general would help ensure that AI policies are well co‑ordinated with, and complement, industrial policy efforts at the sectoral and Member State level.

Further develop skills and talent and promote trust in AI

Workforce and organisational barriers can significantly hinder AI adoption by manufacturing enterprises. Skills required for successful development of AI in manufacturing include a combination of AI-related skills that are not abundantly available and in high demand, and technical industry-specific skills that can also fall short. In addition to concerns with talent concentration in high-paying countries and brain drain, enterprises note that retirement of experienced staff erodes critical domain knowledge, which remains tacit and is rarely digitised, notably among smaller enterprises.

Employees in the manufacturing sector have two important types of concerns that might act as a barrier to AI adoption: i) job security and fear of automation that can result in passive resistance to AI adoption; and ii) concerns about AI-generated decisions, with operators and managers struggling to accept AI outputs. With regards to senior decision makers, managerial scepticism and inertia tend to delay adoption. Earlier failures of AI projects in industrial settings can exacerbate scepticism and risk aversion.

Strengthen AI workforce development: AI talent is an essential building block for AI adoption in manufacturing enterprises. EU-wide AI education programmes should be expanded and connect with the industry, labour unions and academia. Together, they should work to understand how best to design both scientific and vocational education and training curricula that can provide a well-balanced combination of skills required by manufacturing enterprises considering AI deployment. The European Union should continue supporting reskilling and upskilling programmes for both technical and non-technical staff to address skill gaps for AI adoption in manufacturing sectors and help enterprises manage the transition towards AI solutions. In a context of scarce global supply of AI talent and concerns of brain drain in the EU, policymakers might consider enhancing dedicated programmes to attract (and retain) AI talent. For example, they could support enterprises willing to create highly skilled job opportunities within the manufacturing sector.

Build trust and awareness in AI adoption: Resistance to change engrained in corporate culture acts as an important barrier to AI adoption, notably in more traditional manufacturing industries. To address this challenge, the European Union needs to support initiatives that build trust on AI among all employees, including both operators and managers. More specifically, the European Union should promote communication campaigns and success-story showcases to build trust in AI use more broadly. At the firm-level, support for AI mentorship and training programmes, as well as within-firm knowledge exchange, could help equip enterprises with a workforce that can make the most of both AI advancements and available manufacturing expertise. Policymakers should also work closely with both the industry and labour unions on two fronts. First, they should develop a clear strategy to minimise the impacts of AI adoption on the manufacturing workforce. Second, they should communicate on existing programmes to help workers upskill and reskill. In so doing, they would address concerns about automation and job losses.

References

[79] Addionics (2024), “Homepage”, https://addionics.com/ (accessed on 24 April 2025).

[12] Agrawal, A. et al. (2023), “Digital twin: Where do humans fit in?”, Automation in Construction, Vol. 148, p. 104749, https://doi.org/10.1007/s00146-020-01049-0.

[61] Ahmed, I., G. Jeon and F. Piccialli (2022), “From artificial intelligence to explainable artificial intelligence in industry 4.0: A survey on what, how, and where”, IEEE Transactions on Industrial Informatics, Vol. 18/8, pp. 5031-5042, https://doi.org/10.1109/TII.2022.3146552.

[89] Aktepe, A., E. Yanık and S. Ersöz (2021), “Demand forecasting application with regression and artificial intelligence methods in a construction machinery company”, Journal of Intelligent Manufacturing, Vol. 32/6, pp. 1587-1604, https://doi.org/10.1007/s10845-021-01737-8.

[96] Albayrak Ünal, Ö., B. Erkayman and B. Usanmaz (2023), “Applications of artificial intelligence in inventory management: A systematic review of the literature”, Archives of Computational Methods in Engineering, Vol. 30/4, pp. 2605-2625, https://doi.org/10.1007/s11831-022-09879-5.

[91] Allal-Chérif, O., V. Simón-Moya and A. Ballester (2021), “Intelligent purchasing: How artificial intelligence can redefine the purchasing function”, Journal of Business Research, Vol. 124, pp. 69-76, https://doi.org/10.1016/j.jbusres.2020.11.050.

[47] Banjeree, D., K. Ashok and K. Sharma (2024), “AI enhanced predictive maintenance for manufacturing system”, International Journal of Research and Review Techniques, Vol. 3/1, pp. 143-146, https://www.researchgate.net/publication/383022732_AI_Enhanced_Predictive_Maintenance_for_Manufacturing_System.

[23] Benhanifia, A. et al. (2025), “Systematic review of predictive maintenance practices in the manufacturing sector”, Intelligent Systems with Applications, Vol. 26, p. 200501, https://doi.org/10.1016/j.iswa.2025.200501.

[36] BPI France (2025), “L’IA dans les PME et ETI françaises : une révolution tranquille”, BPI France, https://presse.bpifrance.fr/lia-dans-les-pme-et-eti-francaises-une-revolution-tranquille (accessed on 28 March  2025).

[63] Brito, L. et al. (2022), “An explainable artificial intelligence approach for unsupervised fault detection and diagnosis in rotating machinery”, Mechanical Systems and Signal Processing, Vol. 163, p. 108105, https://doi.org/10.1016/j.ymssp.2021.108105.

[67] Çallı, M., E. Albak and F. Öztürk (2022), “Prediction and optimization of the design and process parameters of a hybrid DED product using artificial intelligence”, Applied Sciences, Vol. 12/10, p. 5027, https://doi.org/10.3390/app12105027.

[31] Calvino, F. et al. (2024), “A sectoral taxonomy of AI intensity”, OECD Artificial Intelligence Papers, No. 30, OECD Publishing, Paris, https://doi.org/10.1787/1f6377b5-en.

[34] Calvino, F. and L. Fontanelli (2023), “A portrait of AI adopters across countries: Firm characteristics, assets’ complementarities and productivity”, OECD Science, Technology and Industry Working Papers, No. 2023/02, OECD Publishing, Paris, https://doi.org/10.1787/0fb79bb9-en.

[72] Cardellicchio, A. et al. (2024), “Automatic quality control of aluminium parts welds based on 3D data and artificial intelligence”, Journal of Intelligent Manufacturing, Vol. 35/4, pp. 1629-1648, https://doi.org/10.1007/s10845-023-02124-1.

[80] Comau (n.d.), “Homepage”, https://www.comau.com/en/ (accessed on 1 April 2025).

[41] Cottongins (2025), “Cost analysis: Predictive maintenance vs. reactive repairs”, https://cottongins.org/blog/cost-analysis-predictive-maintenance-vs-reactive-repairs (accessed on 9 July 2025).

[17] Culot, G., M. Podrecca and G. Nassimbeni (2024), “Artificial intelligence in supply chain management: A systematic literature review of empirical studies and research directions”, Computers in Industry, Vol. 162, p. 104132, https://doi.org/10.1016/j.compind.2024.104132.

[5] De Backer, K., I. Desnoyers-James and L. Moussiegt (2015), “’Manufacturing or Services – That is (not) the Question’: The Role of Manufacturing and Services in OECD Economies”, OECD Science, Technology and Industry Policy Papers, No. 19, OECD Publishing, Paris, https://doi.org/10.1787/5js64ks09dmn-en.

[99] Dubey, R. et al. (2021), “Facilitating artificial intelligence powered supply chain analytics through alliance management during the pandemic crises in the B2B context”, Industrial Marketing Management, Vol. 96, pp. 135-146, https://doi.org/10.1016/j.indmarman.2021.05.003.

[105] Elio.Earth (2025), “Homepage”, https://www.elio.earth/ (accessed on 18 April  2025).

[29] ERT (2025), Harnessing Generative AI for Europe’s Industrial Future, The European Round Table for Industry (ERT).

[18] Espina-Romero, L. et al. (2024), “Challenges and opportunities in the implementation of AI in manufacturing: A bibliometric analysis”, Sci, Vol. 6/4, p. 60, https://doi.org/10.3390/sci6040060.

[19] Espinosa-Jaramillo, M. (2024), “Digital Twins in Supply Chain Operations Bridging the Physical and Digital Worlds using AI.”, Journal of Electrical Systems, Vol. 20/10s, pp. 1764-1774, https://doi.org/10.52783/jes.5434.

[2] Eurostat (2025), “Artificial intelligence by NACE Rev. 2 activity”, in Science, technology, digital society, Eurostat, https://doi.org/10.2908/ISOC_EB_AIN2.

[3] Eurostat (2025), “Enterprises by detailed NACE Rev. 2 activity and special aggregates”, in Industry, trade and services, Eurostat, https://doi.org/10.2908/SBS_OVW_ACT.

[33] Eurostat (2025), “Statistics explained: Use of artificial intelligence in enterprises”, https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises (accessed on 13 March 2025).

[87] Eyo-Udo, N. (2024), “Leveraging artificial intelligence for enhanced supply chain optimization”, Open Access Research Journal of Multidisciplinary Studies, Vol. 7/2, pp. 001-015, https://doi.org/10.53022/oarjms.2024.7.2.0044.

[104] Feng, Y., L. K-H and Q. Zhu (2022), “Green supply chain innovation: Emergence, adoption, and challenges”, International Journal of Production Economics, Vol. 248, p. 108497, https://doi.org/10.1016/j.ijpe.2022.108497.

[52] Fernandes, M., J. Corchado and G. Marreiros (2022), “Machine learning techniques applied to mechanical fault diagnosis and fault prognosis in the context of real industrial manufacturing use-cases: A systematic literature review”, Applied Intelligence, Vol. 52/12, pp. 14246-14280, https://doi.org/10.1007/s10489-022-03344-3.

[30] Filippucci, F., P. Gal and M. Schief (2024), “Miracle or myth? Assessing the macroeconomic productivity gains from artificial intelligence”, OECD Artificial Intelligence Papers, No. 29, OECD Publishing, Paris, https://doi.org/10.1787/b524a072-en.

[42] Fordal, J. et al. (2023), “Application of sensor data based predictive maintenance and artificial neural networks to enable Industry 4.0.”, Advances in Manufacturing, Vol. 11/2, pp. 248-263, https://doi.org/10.1007/s40436-022-00433-x.

[37] Frankfurter Allgemeine (2025), “Title”, [English translation], https://www.faz.net/agenturmeldungen/dpa/studie-kuenstliche-intelligenz-ermoeglicht-milliardengewinne-110390455.html (accessed on 25 April 2025).

[77] Fu, Y. et al. (2022), “Machine learning algorithms for defect detection in metal laser-based additive manufacturing: A review”, Journal of Manufacturing Processes, Vol. 75, pp. 693-710, https://doi.org/10.1016/j.jmapro.2021.12.061.

[20] Gao, R. et al. (2024), “Artificial Intelligence in manufacturing: State of the art, perspectives, and future directions”, CIRP Annals, Vol. 73/2, pp. 723-749, https://doi.org/10.1016/j.cirp.2024.04.101.

[101] Glory, E. (2023), “Supplier relationship management in the era of Industry 4.0”, European Journal of Supply Chain Management, Vol. 1/1, pp. 41-51, https://forthworthjournals.org/journals/index.php/EJSCM/article/view/40.

[57] Goldmann, C. et al. (2023), “Interpreting learning models in manufacturing processes: Towards explainable AI methods to improve trust in classifier predictions”, Journal of Industrial Information Integration, Vol. 33, p. 100439, https://doi.org/10.1016/j.jii.2023.100439.

[85] Grover, N. (2025), “AI-enabled supply chain optimization”, International Journal of Advanced Research in Science, Communication and Technology, pp. 28-44, https://doi.org/10.48175/ijarsct-23103.

[100] Guida, M. et al. (2023), “The role of artificial intelligence in the procurement process: State of the art and research agenda”, Journal of Purchasing and Supply Management, Vol. 29/2, p. 100823, https://doi.org/10.1016/j.pursup.2023.100823.

[13] Hoffmann, R. and C. Reich (2023), “A systematic literature review on artificial intelligence and explainable artificial intelligence for visual quality assurance in manufacturing”, Electronics, Vol. 12/22, p. 4572, https://doi.org/10.3390/electronics12224572.

[59] Huang, Z. et al. (2023), “Hybrid learning-based digital twin for manufacturing process: Modeling framework and implementation”, Robotics and Computer-Integrated Manufacturing, Vol. 82, p. 102545, https://doi.org/10.1016/j.rcim.2023.102545.

[65] Hütten, N. et al. (2024), “Deep learning for automated visual inspection in manufacturing and maintenance: A survey of open-access papers”, Applied System Innovation, Vol. 7/1, p. 11, https://doi.org/10.3390/asi7010011.

[94] Islam, M. et al. (2024), “Role of artificial intelligence and machine learning in optimizing inventory management across global industrial manufacturing & supply chain: A multi-country review”, International Journal of Management Information Systems and Data Science, Vol. 1/2, pp. 1-14, https://doi.org/10.62304/ijmisds.v1i2.105.

[38] Jakubowski, J. et al. (2024), “Artificial intelligence approaches for predictive maintenance in the steel industry: A survey”, arXiv, Vol. 2405.12785, https://doi.org/10.48550/arXiv.2405.12785.

[83] Javaid, M., A. Haleem and R. Singh (2023), “A study on ChatGPT for Industry 4.0: Background, potentials, challenges and eventualities”, Journal of Economy and Technology, Vol. 1, pp. 127-143, https://doi.org/10.1016/j.ject.2023.08.001.

[95] Jones, J. (2025), “The role of artificial intelligence in enhancing demand forecasting and supply chain decision-making”, Preprints, https://doi.org/10.20944/preprints202503.0502.v1.

[45] Keleko, A. et al. (2022), “Artificial intelligence and real-time predictive maintenance in industry 4.0: A bibliometric analysis”, AI and Ethics, Vol. 2/4, pp. 553-577, https://doi.org/10.1007/s43681-021-00132-6.

[24] Keramati Feyz Abadi, M. et al. (2025), “Leveraging AI for energy-efficient manufacturing systems: Review and future prospectives”, Journal of Manufacturing Systems, Vol. 78, pp. 153-177, https://doi.org/10.1016/j.jmsy.2024.11.017.

[93] Khadem, M., A. Khadem and S. Khadem (2023), “Application of artificial intelligence in supply chain: Revolutionizing efficiency and optimization”, International Journal of Industrial Engineering, Vol. 5/1, pp. 29-38, https://www.researchgate.net/publication/374163747_Application_of_Artificial_Intelligence_in_Supply_Chain_Revolutionizing_Efficiency_and_Optimization.

[68] Khinvasara, T., S. Ness and A. Shankar (2024), “Leveraging AI for enhanced quality assurance in medical device manufacturing”, Asian Journal of Research in Computer Science, Vol. 17/6, pp. 13-35, https://doi.org/10.9734/ajrcos/2024/v17i6454.

[25] Khurram, M. et al. (2025), “Artificial Intelligence in manufacturing industry worker safety: A new paradigm for hazard prevention and mitigation”, Processes, Vol. 13/5, p. 1312, https://doi.org/10.3390/pr13051312.

[81] Kim, S. et al. (2022), “Recent advances of artificial intelligence in manufacturing industrial sectors: A review”, International Journal of Precision Engineering and Manufacturing, Vol. 23, pp. 111-129, https://doi.org/10.1007/s12541-021-00600-3.

[70] Kuric, I. et al. (2022), “Approach to automated visual inspection of objects based on artificial intelligence”, Applied Sciences, Vol. 12/2, p. 864, https://doi.org/10.3390/app12020864.

[66] Lee, K. et al. (2021), “Deploying an artificial intelligence-based defect finder for manufacturing quality management”, AI Magazine, Vol. 42/2, pp. 5-18, https://doi.org/10.1609/aimag.v42i2.15094.

[71] Liu, F. et al. (2024), “Automated industrial crack inspection system based on edge-edge collaboration of multiple cameras and programmable logic controller”, IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), pp. 1-4, https://doi.org/10.1109/BMSB62888.2024.10608346.

[21] Liu, Y., S. Pan and E. Ballot (2024), “Unveiling the potential of digital twins in logistics and supply chain management: Services, capabilities, and research opportunities”, Digital Engineering, Vol. 3, p. 100025, https://doi.org/10.1016/j.dte.2024.100025.

[50] López, D. et al. (2023), “Fusing anomaly detection with false positive mitigation methodology for predictive maintenance under multivariate time series”, Information Fusion, Vol. 100, p. 101957, https://doi.org/10.1016/j.inffus.2023.101957.

[60] Mata, O. et al. (2025), “Digital twin designs with generative AI: Crafting a comprehensive framework for manufacturing systems”, Journal of Intelligent Manufacturing, pp. 1-24, https://doi.org/10.1007/s10845-025-02583-8.

[58] Mikołajewska, E. et al. (2025), “Generative AI in AI-based digital twins for fault diagnosis for predictive maintenance in industry 4.0/5.0”, Applied Sciences, Vol. 15/6, p. 3166, https://doi.org/10.3390/app15063166.

[74] Moba Group (2025), “Homepage”, https://moba.net/ (accessed on 2 April 2025).

[44] MPO (2024), “Czechia as a technological leader. Government approved the national strategy for artificial Intelligence of the Czech Republic 2030”, Ministry of Industry and Trade, Czech Republic, https://www.mpo.gov.cz/en/guidepost/for-the-media/press-releases/czechia-as-a-technological-leader--government-approved-the-national-strategy-for-artificial-intelligence-of-the-czech-republic-2030--282278/ (accessed on 5 November 2024).

[14] Mypati, O. et al. (2023), “A critical review on applications of artificial intelligence in manufacturing”, Artificial Intelligence Review, Vol. 56/S1, pp. 661-768, https://doi.org/10.1007/s10462-023-10535-y.

[54] Nath, A., S. Udmale and S. Singh (2021), “Role of artificial intelligence in rotor fault diagnosis: A comprehensive review”, Artificial Intelligence Review, Vol. 54, pp. 2609-2668, https://doi.org/10.1007/s10462-020-09910-w.

[103] Naz, F. et al. (2022), “Reviewing the applications of artificial intelligence in sustainable supply chains: Exploring research propositions for future directions”, Business Strategy and the Environment, Vol. 31/5, pp. 2400-2423, https://doi.org/10.1002/bse.3034.

[7] Nolan, A. (2021), “Artificial intelligence, its diffusion and uses in manufacturing”, OECD Going Digital Toolkit Notes, No. 12, OECD Publishing, Paris, https://doi.org/10.1787/249e2003-en.

[98] Nweje, U. and M. Taiwo (2025), “Leveraging artificial intelligence for predictive supply chain management, focus on how AI-driven tools are revolutionizing demand forecasting and inventory optimization”, International Journal of Science and Research Archive, Vol. 14/1, pp. 230-250, https://doi.org/10.30574/ijsra.2025.14.1.0027.

[86] Odumbo, O. and S. Nimma (2025), “Leveraging artificial intelligence to maximize efficiency in supply chain process optimization”, International Journal of Research Publication and Review, Vol. 6/1, pp. 3035-3050, https://doi.org/10.55248/gengpi.6.0125.0508.

[1] OECD (2025), Annual value added and its components by economic activity, (database), https://data-explorer.oecd.org (accessed on 28 March 2025).

[6] OECD (2025), “Mapping the semiconductor value chain: Working towards identifying dependencies and vulnerabilities”, OECD Science, Technology and Industry Policy Papers, No. 182, OECD Publishing, Paris, https://doi.org/10.1787/4154cdbf-en.

[8] OECD (2022), “OECD Framework for the Classification of AI systems”, OECD Digital Economy Papers, No. 323, OECD Publishing, Paris, https://doi.org/10.1787/cb6d9eca-en.

[32] OECD.AI (2025), AI talent concentration by country and industryAI talent concentration by country and industry, data from LinkedIn Economic Graph, last updated 2025-04-07, (database), https://oecd.ai/en/data?selectedArea=ai-jobs-and-skills&selectedVisualization=ai-talent-concentration-by-country-and-industry&visualizationFiltersHash=eyJmaWx0ZXJzIjp7IkluZHVzdHJ5IjoiTWFudWZhY3R1cmluZyJ9fQ%3D%3D (accessed on 10 April 2025).

[90] Olorunyomi, S. et al. (2024), “Leveraging artificial intelligence for enhanced supply chain optimization: A comprehensive review of current practices and future potentials”, International Journal of Management & Entrepreneurship Research, Vol. 6/3, pp. 707-721, https://doi.org/10.51594/ijmer.v6i3.882.

[22] Ouahabi, N. et al. (2024), “Leveraging digital twin into dynamic production scheduling: A review”, Robotics and Computer-Integrated Manufacturing, Vol. 89, p. 102778, https://doi.org/10.1016/j.rcim.2024.102778.

[26] Ouahabi, N. et al. (2025), “Dynamic production scheduling and maintenance planning under opportunistic grouping”, Computers & Industrial Engineering, Vol. 199, p. 110646, https://doi.org/10.1016/j.cie.2024.110646.

[102] Pal, S. (2023), “Integrating AI in sustainable supply chain management: A new paradigm for enhanced transparency and sustainability”, International Journal for Research in Applied Science and Engineering Technology, Vol. 11/4, pp. 2979-2984, https://doi.org/10.22214/ijraset.2023.54139.

[69] Pfrommer, J. et al. (2018), “Optimisation of manufacturing process parameters using deep neural networks as surrogate models”, Procedia CiRP, Vol. 72, pp. 426-431, https://doi.org/10.1016/j.procir.2018.03.046.

[97] Phantasma Labs (2025), “Homepage”, https://www.phantasma.global/ (accessed on 27 March 2025).

[35] Phantasma Labs (2025), “Revolutionize production planning with AI-driven optimization”, https://www.phantasma.global/ (accessed on 5 January 2025).

[15] Plathottam, S. et al. (2023), “A review of artificial intelligence applications in manufacturing operations”, Journal of Advanced Manufacturing and Processing, Vol. 5/3, https://doi.org/10.1002/amp2.10159.

[92] Quoc Khoa, B. et al. (2024), “Impact of artificial intelligence’s part in supply chain planning and decision making optimization”, International Journal of Multidisciplinary Research and Growth Evaluation, Vol. 5/6, pp. 837-856.

[10] Radanliev, P. et al. (2020), “Artificial intelligence in cyber physical systems”, AI & SOCIETY, Vol. 36/3, pp. 783-796, https://doi.org/10.1007/s00146-020-01049-0.

[76] Rathi, N. and N. Rathi (2020), “An application of ANN for modeling and optimisation of process parameters of manufacturing process: A review”, International Journal of Engineering Applied Sciences and Technology, Vol. 4, pp. 127-134, https://doi.org/10.33564/IJEAST.2020.v04i12.017.

[46] Raza, F. (2023), “AI for predictive maintenance in industrial systems”, Cosmic Bulletin of Business Management, Vol. 2/1, pp. 167-183, https://doi.org/10.13140/RG.2.2.27313.35688.

[88] Riad, M., M. Naimi and C. Okar (2024), “Enhancing supply chain resilience through artificial intelligence: Developing a comprehensive conceptual framework for AI implementation and supply chain optimization”, Logistics, Vol. 8/4, p. 111, https://doi.org/10.3390/logistics8040111.

[40] Rojek, I. et al. (2023), “An artificial intelligence approach for improving maintenance”, Applied Sciences, Vol. 13/8, p. 4971, https://doi.org/10.3390/app13084971.

[4] Rossi, C. (2022), “A comparative investment analysis of batch versus continuous pharmaceutical manufacturing technologies”, Journal of Pharmaceutical Innovation, Vol. 17/4, pp. 1373-1391, https://doi.org/10.1007/s12247-021-09612-y.

[64] Rožanec, J. et al. (2022), “Towards a comprehensive visual quality inspection for industry 4.0”, IFAC-PapersOnLine, Vol. 55/10, pp. 690-695, https://doi.org/10.1016/j.ifacol.2022.09.486.

[62] Scaife, A. (2024), “Improve predictive maintenance through the application of artificial intelligence: A systematic review”, Results in Engineering, p. 101645, https://doi.org/10.1016/j.rineng.2023.101645.

[82] Senoner, J., T. Netland and S. Feuerriegel (2022), “Using explainable artificial intelligence to improve process quality: Evidence from semiconductor manufacturing”, Management Science, Vol. 68/8, pp. 5704-5723, https://doi.org/10.1287/mnsc.2021.4190.

[49] Sensorminds (2024), “Homepage”, https://sensorminds.com/ (accessed on 28 March 2025).

[28] Shafiee, S. (2025), “Generative AI in manufacturing: A literature review of recent applications and future prospects”, Procedia CIRP, Vol. 132, pp. 1-6, https://doi.org/10.1016/j.procir.2025.01.001.

[73] Shafi, I. et al. (2023), “Deep learning-based real time defect detection for optimization of aircraft manufacturing and control performance”, Drones, Vol. 7/1, p. 31, https://doi.org/10.3390/drones7010031.

[11] Shrivastav, M. (2022), “Barriers related to AI implementation in supply chain management”, Journal of Global Information Management, Vol. 30/8, pp. 1-19, https://doi.org/10.4018/JGIM.296725.

[9] Sierla, S. et al. (2018), “Automatic assembly planning based on digital product descriptions”, Computers in Industry, Vol. 97, pp. 34-46, https://doi.org/10.1016/j.compind.2018.01.013.

[53] Singh and Desai (2023), “Automated surface defect detection framework using machine vision and convolutional neural networks”, Journal of Intelligent Manufacturing, Vol. 34/4, pp. 1995-2011, https://doi.org/10.1007/s10845-021-01878-w.

[84] Singh, S. (2024), “Leveraging AI and machine learning in six-sigma documentation for pharmaceutical quality assurance”, Chinese Journal of Applied Physiology, Vol. e20240005, https://doi.org/10.62958/j.cjap.2024.005.

[51] Singh, V. et al. (2023), “Artificial intelligence application in fault diagnostics of rotating industrial machines: A state-of-the-art review”, Journal of Intelligent Manufacturing, Vol. 34/3, pp. 931-960, https://doi.org/10.1007/s10845-021-01861-5.

[16] Soori, M., B. Arezoo and R. Dastres (2023), “Digital twin for smart manufacturing, A review”, Sustainable Manufacturing and Service Economics, Vol. 2, p. 100017, https://doi.org/10.1016/j.smse.2023.100017.

[43] Surucu, O., S. Gadsden and J. Yawney (2023), “Condition monitoring using machine learning: A review of theory, applications, and recent advances”, Expert Systems with Applications, Vol. 221, p. 119738, https://doi.org/10.1016/j.eswa.2023.119738.

[55] Taşcı, B., A. Omar and A. Serkan (2023), “Remaining useful lifetime prediction for predictive maintenance in manufacturing”, Computers & Industrial Engineering, Vol. 184, p. 109566, https://doi.org/10.1016/j.cie.2023.109566.

[39] Ucar, A., M. Karakose and N. Kırımça (2024), “Artificial intelligence for predictive maintenance applications: Key components, trustworthiness and future trends”, Applied Sciences, Vol. 14/2, p. 898, https://doi.org/10.3390/app14020898.

[27] Urrea, C. and J. Kern (2025), “Recent advances and challenges in industrial robotics: A systematic review of technological trends and emerging applications”, Processes, Vol. 13/3, p. 832, https://doi.org/10.3390/pr13030832.

[56] Wang, Y., Y. Zhao and S. Addepalli (2020), “Remaining useful life prediction using deep learning approaches: A review”, Procedia manufacturing, Vol. 49, pp. 81-88, https://doi.org/10.1016/j.promfg.2020.06.015.

[48] Wiese, T. (2024), “Predictive maintenance using artificial intelligence in critical infrastructure: A decision-making framework”, International Journal of Engineering, Business and Management (IJEBM), Vol. 8/4, pp. 1-4, https://doi.org/10.22161/ijebm.8.4.

[75] Yang, J. et al. (2020), “Using deep learning to detect defects in manufacturing: A comprehensive survey and current challenges”, Materials, Vol. 13/24, p. 5755, https://doi.org/10.3390/ma13245755.

[78] Zimmerling, C. et al. (2022), “Optimisation of manufacturing process parameters for variable component geometries using reinforcement learning”, Materials & Design, Vol. 214/110, p. 423, https://doi.org/10.1016/j.matdes.2022.110423.