Generalisation of human-centred AI applications for production optimization
The growing demand for product diversity, quality and sustainability poses enormous challenges for manufacturing technology. To master this complexity, especially artificial intelligence (AI) can be utilised to generate new added value and digital services from the available product- and production-specific data. However, the lack of implementation concepts and standards as well as the frequently low interoperability of AI applications implemented in isolated solutions still hinder the widespread use of AI in German manufacturing companies. In addition, the available data is often of insufficient quality and quantity. Especially for small and medium-sized enterprises that offer a wide range of products outside of large-scale production, there is a high barrier to entry for AI applications. Productivity, flexibility and robustness of production facilities are mostly based on the process knowledge of experts.
Objectives© Oliver Petrovic
One approach lies in the holistic consideration of the key factors humans, AI and production resources in transferable, human-centred AI applications. By considering this triad as a learning overall system, broad value creation potentials can be raised. Hybrid intelligence systems are created in which the complementary strengths of experts and AI are combined. On the one hand, AI learns from humans by including them in the training loop of the models, on the other hand, process transparency is increased by processing the data in user-friendly AI assistance systems. Transferring the experience of experts into digital services thus shifts the barrier to entry for unlocking the productivity potential of AI in the direction of smaller series.
To enable such systems, it is necessary to offer intuitive interfaces to humans across the entire digital value chain - from data acquisition across the shop floor, through middleware and the data platform, to the AI application. It must be possible to interact with data and AI models via the human machine interface (HMI) in addition to pure process operation.
Data must be made usable and processed through modelling, contextualisation and labelling. Humans take on different roles in the overall system depending on the form of interaction and the expertise they bring to the table. The AI, on the other hand, must be further developed from a black box to an explainable AI so that it becomes clear to the users which correlations underlie the findings of the AI. Building on this, users must be offered communication channels that allow feedback on the AI's insights. Through these feedback channels, AI models can be learned efficiently on the basis of smaller batches and with fewer training iterations, thanks to the expertise brought in by humans.
However, the human expertise brought into the digital value chain as a generalising element is not sufficient on its own to ensure broad transferability of the solutions. A high level of interoperability of the data and AI models must additionally be sought with regard to the process, the production equipment used, the manufactured product, the middleware, data platform and the interacting operators. In this way, modular, broadly applicable and scalable systems can be developed in contrast to isolated solutions, which open up new value creation potential and business models.
Project Consortia© Oliver Petrovic
In the research project GeMeKI, the WZL of RWTH Aachen University, in cooperation with 11 other partners under the leadership of aiXbrain GmbH, is investigating the proposed solutions using the example of use cases from the manufacturing context in realistic best practices. An adhesive-based joining process, a milling process and a metal forming process are being examined, which will then be transferred into generic implementation concepts.
Within the context of GeMeKI, organisations that are not part of the consortium are involved through various channels. In addition to the direct cooperation with the ProLern research area, the consortium uses a network of alliances and associated partners (e.g. IoP, AI Center of RWTH Aachen University) for strategic consulting, for mitigation measures of risks and in the area of knowledge and transfer of results. In addition, committees for the four thematic areas of joining, cutting, forming and human-AI interaction are linked up in coordination with the consortium.