In recent years, AI adoption has made significant strides. The demand for AI in manufacturing is rapidly growing, driven by the need for increased efficiency, reduced operational costs, and enhanced product quality. However, challenges to adoption still persist. Here, Nandini Chakravorti, Associate Director for Digital Engineering at the Manufacturing Technology Centre, explains how these issues can be successfully navigated.
Market analysis indicates a strong interest in AI solutions that can optimise production processes, predict maintenance needs and improve supply chain management. Expenditure on AI technologies could increase to between £27.2bn and £35.6bn by 2025, at annual growth rates of roughly 10% and 16% respectively.
The potential impact of AI in the manufacturing industry is vast, ranging from revolutionising inventory management with real-time demand forecasting, to enabling personalised product customisation through adaptive production and supporting industrial sustainability.
The increased availability of large datasets for training, enhanced investment in AI technologies and availability of infrastructure has propelled AI from the work of academic research and science fiction into tangible, real-world applications. While public attention often centres on generative AI models such as ChatGPT, the manufacturing industry has been using AI in various forms for decades.
For example, AI has been used to minimise machine downtime, increase supply chain efficiency and improve workforce safety. Recently, projects have made significant strides in demystifying and integrating the industrial metaverse within manufacturing, creating opportunities for innovation, growth and efficiency.
AI technologies such as image recognition, generative AI and natural language processing can be used to enhance user experience. AI also enhances the possibilities for personalised and intuitive human-machine interfaces, thus increasing the potential of adopting AI in manufacturing to enhance productivity.
By leveraging AI across various functions, manufacturers can achieve remarkable agility, responsiveness and customer satisfaction, positioning themselves for a competitive edge in the rapidly evolving marketplace. For example, AI and digital technologies can support servitisation as a business strategy in the future by harnessing data to improve the value-added services and solutions that are provided alongside, or as a service instead of the product itself.
However, despite significant time and research invested in AI across the industry, uptake has remained slow, and there is still a long way to go before its full potential is unlocked. This article explores the barriers holding the industry back and discusses ways to mitigate these to capitalise on the benefits of AI in manufacturing.
Barriers to adoption
A recent survey of UK businesses revealed that only around 15% of manufacturing respondents have currently adopted AI processes. Of these adopters, only 13% intend to actively use it . So, what’s holding the manufacturing industry back, and what needs to be done for its potential to be unlocked on an industry-wide scale? Feedback from various end-users and vendors points to four primary concerns: a lack of trust in AI systems, challenges in navigating regulations, unavailability of skills and uncertainties regarding AI’s practical benefits in manufacturing.
Lack of trust
AI systems face an ongoing challenge; they lack transparency in their decision-making processes, which makes it difficult to validate their outputs. AI systems can introduce significant biases and ethical concerns into operations. In fact, a recent report from MIT Technology Review found that AI use-case development is hampered by inadequate data quality (57%), weak data integration (54%) and weak governance (47%), impacting the scaling of AI solutions within industrial environments.
AI models offer limited transparency to decision making, making it hard to verify the validity of outcomes. Unclear rights and access to reliable, transparent, unbiased and relevant data from across the lifecycle of AI development, integration and use, linked with a lack of confidence or trust in the accuracy and reliability of AI methods (especially in safety-critical applications) is a significant barrier to industrial adoption. Security concerns further compound these trust issues, especially regarding access to sensitive data and vulnerability to cyber-attacks.
Uncertainties about AI’s practical benefits in manufacturing and the skills challenge
Addressing the lack of understanding around the benefits of AI adoption (including ROI when considering it as a business investment) and overcoming organisational barriers such as having a clear digital and data strategy, requires a strong emphasis on addressing the skills gap.
For successful industrial deployment, a combination of expertise in data science and a deeper understanding of manufacturing processes is needed. To cultivate a cross-functional workforce capable of supporting AI in manufacturing, a combination of specialised training and knowledge in both fields is essential. This approach will empower the workforce to drive industrial scale AI projects and foster innovation within the manufacturing ecosystem. The UK’s RTO and academia network have a positive role to play in ensuring the future workforce is equipped to use these powerful technologies to achieve the desired impact.
Complexities in navigating regulations
Numerous complex frameworks for regulating AI, together with a lack of awareness of AI policies and standards, make it difficult for manufacturing stakeholders to know which policies and standards are right for their operations. Consequently, it is harder to guarantee regulatory approval for AI-based solutions.
There is also a lack of practical guidance. Clearer, simpler recommendations for integrating AI into manufacturing processes, while ensuring trustworthiness and compliance with industry standards, are much needed. Without such guidance, the full benefits of AI in manufacturing cannot be successfully untapped, hindering widespread adoption and optimisation across the sector.
For AI to be successfully adopted within the industry, there’s a clear need for principles and practices that can facilitate the development and deployment of ethical, secure, transparent and robust AI systems. This is why the Trustworthy AI Framework is in development. As AI technology rapidly advances, these principles aim to inspire confidence among users, developers and stakeholders, preventing misuse and promoting safe AI adoption.
Trustworthy AI in practice
As manufacturers increasingly rely on AI to maintain quality standards, implementing trustworthy AI not only enhances operational efficiency, but also the quality of the final products. What’s more, a transparent AI system allows manufacturers to understand the rationale behind specific decisions. This is especially critical when product quality is synonymous with safety and regulatory compliance. It also instils confidence along the whole supply chain, and companies can trust the quality of AI-evaluated products from their direct partners and beyond.
Similarly, in the process of robotics and automation, Trustworthy AI ensures that autonomous robots, integrated with AI, perform their tasks reliably and safely. These robots handle repetitive tasks such as picking, placing, welding, painting, assembly and inspection, freeing up workers’ time for more productive activities. Integrating robots with machine learning allows them to self-learn and improve over time. Machine vision enhances their capabilities, enabling them to move and navigate in complex environments while ensuring safety when working alongside human colleagues. Trustworthy AI practices are pivotal in this context to overcome barriers of trust, safety and reliability.
How can we begin implementing trust in AI within the industry?
Developing trustworthy AI models is a complex process requiring a variety of specialised roles throughout the entirety of the AI development lifecycle. In line with the guidance provided by the UK government’s Centre for Data Ethics and Innovation (CDEI) and the Department for Science, Innovation and Technology (DSIT), Trustworthy AI revolves around five core principles: Fairness, Appropriate Transparency and Explainability, Safety, Security and Robustness. These principles guide the design, development, deployment, and use of AI systems, ensuring that AI technologies adhere to governance guidelines, prioritise data security, enhance human and asset safety, and promote model transparency.
Manufacturing firms and bodies can begin encouraging AI adoption through implementing AI assurance techniques. These are commonly used in industries like finance, to build confidence in products, systems and organisations. Applying these techniques to AI systems fosters trust and demonstrates their reliability and trustworthiness.
- Impact assessment: Used to anticipate the effect of a system on environmental, equality, human rights, data protection or other outcomes. Includes risk assessments and risk management strategies.
- Bias audit: Determines if there is unfair bias in the input data or the outcome of a decision or classification made by the system.
- Compliance audit: A review of a company’s adherence to internal policies and procedures, or external regulations or legal requirements, including specialised types of compliance audits.
- Conformity assessment: Assures that a product, service or system being supplied meets the expectations specified or claimed, before entering the market.
- Certification: The process where an independent body attests that standards of quality or performance have been met.
- Impact evaluation: Similar to impact assessments but conducted after a system has been implemented retrospectively.
- Performance testing: Used to assess the performance of a system with respect to predetermined quantitative requirements or benchmarks to ensure it meets the performance requirements of users. This can identify bottlenecks and improvements to AI performance.
- Formal verification: These prove the correctness of an AI system’s behaviour and ensure they behave as they should and provide the correct outcomes.
A framework for the future
While we have knowledge about the principles and techniques of Trustworthy AI, practical explanations of when it’s necessary and how to achieve it are in short supply. To bridge these gaps, the Manufacturing Technology Centre (MTC) has begun developing its own Trustworthy AI framework. It builds upon existing standards and provides a clear checklist for establishing Trustworthy AI in manufacturing systems. It also offers helpful guidance in implementing assurance techniques to ensure trust across the whole AI development lifecycle. An example of the framework can be seen below.
The integration of AI and industrial digital technologies has immense potential in fostering continuous improvement in manufacturing. A comprehensive framework for implementing AI in manufacturing, such as the one outlined above, combined with cross-sector team development and skills training, can effectively address the barriers of trust, uncertainties around AI’s practical benefits, and complexities in navigating regulations. With these pillars in place, manufacturers can begin to unlock AI’s full potential and drive innovation in the industry.
To download the full version of the MTC’s Trustworthy AI framework, visit www.the-mtc.org/mtc-whitepapers.
