Cognitive-assisted agile manufacturing for a workforce supported by trustworthy artificial intelligence

Background and Problem

In knowledge-intensive manufacturing processes, people are at the centre. Training new workers in these processes is time-consuming and costly for companies. Sectors such as the Italian textile industry also suffer from the shortage of skilled workers caused by demographic change. A further challenge for the manufacturing industry is the constant competition from high-quality products.

COALA addresses these challenges by developing a voice-controlled digital assistant for the manufacturing sector. The COALA solution combines AI components with a voice-controlled assistant that serves as an interface to graphical and textual content directly related to specific tasks. It supports both workers who use analytics tools and new workers undergoing on-the-job training. Alongside the technology, a training and instruction concept was developed that focuses on building and expanding workers' competences in human–AI collaboration. The aim was for the COALA solution to provide tangible support for the way in which workers carry out their work. It was also intended to enable companies to maintain or improve the quality of their production processes and products.

Research questions

• For which work processes and tasks are voice assistance systems suitable as information tools?

• How can the voice assistant be used as a learning medium in the work process of textile manufacturing?

• How can an exemplary learning process be designed using the smallest possible learning units and a voice assistant as the transfer medium?

Approach

At the start of the project, questionnaires were developed and distributed to employees. These served to assess employees' attitudes towards voice assistance systems and to record their experience of structured teaching-learning arrangements. In addition, work process observations and supplementary expert interviews were carried out in the participating textile use cases. These insights made it possible to identify exemplary tasks and processes, place them in a teaching-learning context and develop micro-learning units from them. The learning units were designed to convey learning content of varying intensity – from simple information to complex, video-supported explanations – through a dialogue between workers and the voice assistant.

Participant observation of the voice assistant in use in textile manufacturing and in quality control in household appliance manufacturing yielded insights into usage behaviour and the transfer of information. In addition, tests were carried out with two groups of vocational school students. Each group was offered learning content in conventional form through subject-systematic classroom instruction. They were also given the opportunity to explore the phenomena and content of skilled work in an action-oriented way on real machines on the production floor. Final tests showed a significant increase in learning outcomes when more time was spent on the tasks. The micro-learning units were also received positively as a didactic tool.

Results

The COALA solution was implemented on the basis of Mycroft, a privacy-friendly open-source assistant. It integrated prescriptive quality analytics into an AI system supporting the on-the-job training of new workers and combined this with a novel explanation platform – the WHY Engine. COALA also addressed AI ethics during the design, introduction and use of the new solutions. Key components for the introduction of the solution were a new didactic concept to inform workers about the opportunities, challenges and risks of human–AI collaboration, together with the accompanying change management process. Using three use cases (textiles, white goods, liquid packaging), the results were evaluated by the responsible EU officials in common manufacturing processes of considerable economic relevance.

Dissemination and Transfer Activities

• Presentation and paper at the 17th Annual Conference of the Ingenieurpädagogische Wissenschaftsgesellschaft, 2023

• The underlying didactic concept for the voice assistance system (COALA) is listed as a key innovation in the European Union's Innovation Radar.

• EU officials encouraged a follow-up project to ensure long-term continuation (WASABI).