Part 1 (80min):
Symbolic / Declarative AI for Industry 4.0
Semantic digital twins and how they can be captured using OWL 2 and SHACL, for asset management modelling and linking of such models to industrial data.
Challenges with semantic artefacts for Industry 4.0 in coping, e.g., with simulation and physical equations.
Holistic modelling approach where a semantic artefact of a plant accounts for the domain knowledge of manufacturing, factory static characteristics as well as dynamics but goes beyond and accounts for the domain knowledge of analytics/ML as well as computational infrastructure (cloud computing infrastructure) -- everything interlinked as a holistic digital twin of the plant.
Challenges of standardisation and user accessibility of both the modelling and the semantic models (e.g., their discoverability with complex queries, modelling patterns)
Part 2 (80min): Hand On for Semantic Data Integration and and Open Manufacturing Core Model
Hand on with a real industrial scenario and standardized semantic models
Part 3 (80min): Neuro-Symbolic AI For Industry 4.0
How to bridge symbolic representation and learning methods.
Part 4 (80min): Large Language Models and Industry 4.0 + Discussions + Conclusion
Large Language Models and Industry 4.0
Discussions
Conclusion
Neuro-Symbolic AI is a prominent field that has attracted a significant attention in the last few years in both academia and industry. It has roots in semantic technology and machine learning (ML) and aims at fusing them to benefit from both worlds. Indeed, semantics allows to capture domain knowledge and integrate data, ML allows to perform data analysis and predict; while their fusion allows to guide learning via symbolic knowledge and to extract structures from data via deep learning.
Such fusion has a great potential for Industry 4.0 that aims at smart fully automated factories, moreover, it has already been used in such context while the success is still limited due to a number of factors. For example, ontologies can capture industrial assets, specifications of robots, topology of factories, rules can capture compatibility between skills of workers and manufacturing tasks; they can serve as unified schemata for extremely diverse manufacturing data and allow to convert it into manufacturing knowledge graphs. Deep learning helps predicting, for example, quality of discrete manufacturing operations. Neuro-symbolic methods allow to map relatively static factory specifications consisting of ontologies and relatively small knowledge graphs and highly dynamic and timestamped production data consisting of sensor data and log files (and represented as knowledge graphs or/and numeric vectors) into uniform numeric vector representations. This opens the doors for (deep) learning and various forms of industrial analytics. .
In this tutorial on Neuro-Symbolic AI for Industry 4.0 -- NeSyAI4 --- we will discuss the state of the art technology that ranges from semantic-based industrial modelling to industrial analytics such as quality prediction in discrete manufacturing, exemplify this with Bosch manufacturing scenarios, and also discuss limitations of existing technology in, e.g., in scalability and uncertainty handling, and important research directions to address this limitations.
The main motivation behind our tutorial is to help, on the one hand, academics to better understand industrial needs, on the other hand, industrial practitioners to familiarise themselves with the state of the art research achievements and systems. To this end, the tutorial will bring together research material, elaborate industrial examples from Bosch manufacturing, system aspects and demonstrations, practical exercises, and discussion sessions to help researchers and industrial practitioners to be on the same page and to progress further in bringing Neuro-Symbolic AI and Industry 4.0 to the next level.
Diego Rincon, Bosch Center for AI, Germany
Irlan Grangel Gonzalez, Bosch Center for AI, Germany
Mohamed H. Gad-Elrab, Bosch Center for AI, Germany
Yuqicheng Zhu, Bosch Center for AI, Germany
Baifan Zhou, Oslo Metropolitan University / University of Oslo, Norway
Steffen Staab, University of Stuttgart, Germany
Evgeny Kharlamov, Bosch Center for AI / University of Oslo, Germany