European Commission logo
English English
CORDIS - EU research results

Safe and Trusted Human Centric Artificial Intelligence in Future Manufacturing Lines

Article Category

Article available in the following languages:

People-first approach helps build trust in manufacturing AI

To be successfully adopted in manufacturing, artificial intelligence systems need to be better understood and better trusted. The EU-funded STAR project used human-centric design to build safe and reliable technologies.

Industrial Technologies icon Industrial Technologies

By increasing automation and enhancing the intelligence of manufacturing processes, artificial intelligence (AI) can improve production quality, while reducing costs, transitioning Europe to Industry 5.0. Predictive algorithms can indicate the best time to service machinery, and identify product defects, avoiding costly production downtime. AI can also optimise value chains, by analysing big data to forecast supply, demand and inventory levels, benefiting logistics and production schedules. Yet these advances also introduce significant risks, such as the introduction of biased systems due to inadequate training data or an increased vulnerability to cyberattack. But perhaps the biggest is a lack of understanding and trust from those on the production line. “If Industry 5.0 is to fulfil its potential, it needs not only the support of these people but the benefit of their experience,” says John Soldatos, technical coordinator of the EU-funded STAR (Safe and Trusted Human Centric Artificial Intelligence in Future Manufacturing Lines) project. The STAR team collaborated with key stakeholders to develop various advanced AI technologies. In three pilots, these were evaluated and validated for both technical and social performance – in particular, their trustworthiness.

Putting AI through its paces

The Human-Robot Collaboration (cobotics) pilot took place in a Philips (website in Dutch) factory in the Netherlands. Here, STAR’s active learning systems were tested in AI-driven quality inspection, and were shown to increase process efficiency without compromising cost or workflow. “The AI consults humans when uncertain, thus avoiding errors and misclassifications while also letting the AI benefit from human knowledge, significantly improving the speed and quality of its training,” explains Babis Ipektsidis, project manager at Netcompany–Intrasoft, the project host. The AI security pilot applied explainable AI systems to product customisation in automotive air vents. It was demonstrated at the IBER-OLEFF production facilities in Portugal. Here, variations in monthly orders make it difficult to optimise the manufacturing process. “Explainable AI helped operators understand how automation can make production lines more flexible, while also adapting them to changes, such as the introduction of new parts or end products,” notes Ipektsidis. Lastly, the Safety with AI pilot tested cobotic operations at the German Research Center for Artificial Intelligence (DFKI). Simulated reality systems, based on reinforcement learning, trained robots to safely move around human co-workers while completing shop floor tasks. By defining dynamic safety zones for robots, the AI enhanced the safety of cobotic working, with no collisions observed. The team applied their novel methodology for evaluating trusted AI systems, both technically and socially, to all the pilots. “The technical performance satisfied us that these robotic solutions can be used in real-life scenarios to improve production processes,” explains Soldatos. “While human safety can be assured, employee training will be vital, especially on tasks such as reading dashboards and understanding data-driven results.”

Navigating the wider working environment

Soldatos notes that showcasing the benefits of trust-building AI systems has not been without its challenges: “The sector is moving fast, so anticipating regulation is difficult. The AI Act emerged during the project, but we managed to align our solutions with it.” The project’s ongoing collaborations help address these challenges. STAR researchers are leading activities in the ‘AI projects in manufacturing’ initiative, which facilitates collaboration and knowledge sharing. STAR also contributes assets, such as information about Active Learning AI models, to the AI4EU portal. The project has made some of its resources available through its market platform and has widely published its results, including an Open Access book about trusted AI solutions, which has already been downloaded over 40 000 times. STAR’s partners are currently working to advance the maturity of their prototypes, with the aim of launching commercial products within a few years of the project’s conclusion. While several products are currently protected by proprietary licences, partners offer open-source versions of selected results.


STAR, automation, manufacturing, artificial intelligence, human, algorithm, trust, cobot

Discover other articles in the same domain of application