Generative AI: SystemX launches an R&D project to support decision-making in complex industrial settings

The “Technical Knowledge Management for Complex Systems” project is led by SystemX and brings together Air Liquide and Michelin. It involves adapting generative AI models to industrial contexts by leveraging heterogeneous technical data. The goal is to integrate generative AI into operators’ daily work to assist them in making decisions regarding the maintenance or technical design of complex systems.

SystemX, the Institute for Technological Research (IRT) dedicated to the digital engineering of future systems, announces the launch of the 42-month R&D project “Technical Knowledge Management for Complex Systems.” This is the institute’s first project in the field of generative AI applied to industry. In collaboration with Air Liquide and Michelin, and based on use cases provided by these companies, it aims to develop a scientific framework and technical foundation for evaluating generative AI models in an industrial context. This project also aims to advance these models by tailoring them to specific business domains through the use of multimodal technical data (textual, sensor-based, diagrams, images, etc.).

Given the growing complexity of industrial systems, new approaches are needed to optimize maintenance and engineering operations, reduce human and technical errors, and accelerate innovation during the design and operational phases. Generative artificial intelligence offers numerous opportunities in this field, but AI models have not yet lived up to industrial expectations.

“The rapid adoption of generative AI by industry is a major competitive challenge. While AI providers offer generative AI frameworks and toolchains addressing a wide variety of use cases, these do not meet key industrial challenges, particularly regarding data multimodality and the high standards required by business operations. It is with this objective in mind that the ‘Technical Knowledge Management of Complex Systems’ project was proposed by SystemX to its partners, leveraging the understanding and exploitation of multimodal data to support technical decision-making,” comments Sana Tmar, project lead for ‘Technical Knowledge Management of Complex Systems’ at SystemX.

The project addresses two main scientific objectives:

– To evaluate large language models (LLMs)—whether proprietary or provided by vendors—to quantify their performance and uncertainties and compare them, using representative benchmarks and a rigorous methodology;
– And the development of hybrid generative AI models that integrate explicit domain knowledge of a physical or symbolic nature, as well as other types of data such as time series, to specialize them for industrial contexts.

From a practical standpoint, the work carried out within this project aims to facilitate decision support for engineering professions by using generative AI capable of leveraging multimodal data in a specific business domain, providing them with business reports in various formats (summaries, analyses, recommendations) that draw on the required technical knowledge.

Industrial partners Air Liquide and Michelin are collaborating on two real-world, industrially viable use cases that will enable the development of innovative generative models tailored to their business challenges.

The first use case aims to develop a generative AI system capable of assisting operators in the maintenance of industrial assets. The challenge is to develop a proprietary intelligent system capable of adapting in real time to the operational conditions of different sites and the skills of operators. This system will diagnose failures by cross-referencing data from industrial sensors and incident histories, and will propose specific, tailored maintenance actions. It will facilitate the work of on-site operators and contribute to optimizing industrial performance. The ability to cross-reference multimodal data sources (data from sensors and textual data) represents a significant breakthrough in this application of generative AI.

The second use case involves the development of a proprietary design assistant based on generative AI. This system will transform the way engineers design and validate their technical specifications. For example, it will offer proactive design assistance using multimodal models (combining text, images, and external data) trained on technical corpora capable of suggesting optimizations based on engineering principles and the history of similar projects. It will ensure project compliance verification through hybrid solutions combining semantic analysis and the extraction of business rules from industry standards. Finally, it will automatically generate technical data sheets, user manuals, and specifications that comply with industry standards, providing valuable resources for business teams.