Explainable intelligent maintenance solution for connected manufacturing systems – X-IMS
Explainable intelligent maintenance solution for connected manufacturing systems
/
Challenges and objectives
In the current industrialization landscape, the advent of the factory of the future, characterized by digitalization and connected intelligence, poses new requirements in terms of reliability, equipment availability, work safety and maintenance cost reduction. Faced with these challenges, the X-IMS project proposes an intelligent and explainable maintenance solution. By combining self-monitoring and decision-making capabilities, this solution addresses multidisciplinary scientific and technical challenges to develop an innovative maintenance management system. This solution provides efficient processing of heterogeneous data streams for the automated construction of robust and interpretable health indicators, ensuring continuous monitoring of multi-component systems. It includes advanced algorithms for early identification of anomalies and system-wide prognosis, enriched with explanatory capabilities for better maintenance decision-making. These algorithms allow for the consideration of prediction uncertainties, component interdependencies, and multiple action variables, thus ensuring explicit communication between decision-making processes and maintenance operators. The effectiveness and impact of the developed solution will be demonstrated through industrial application cases, thus marking a significant step forward in the landscape of the industry of the future.
The project aims to transform industrial maintenance with major innovations: the development of interpretable health indicators (HIs) for system monitoring, the integration of advanced Explainable AI (XAI) algorithms for system prognosis and health management (PHM), and the optimization of maintenance decisions. These advances are supported by an X-IMS dashboard, providing real-time visualization and facilitating instant decisions for managers. Project progress includes a literature review of XAI applied to PHM, with a structured taxonomy facilitating the choice of XAI methods meeting specific PHM needs. In view of the identified gaps regarding XAI algorithms for multimodal learning in PHM, the research focused on explaining how multimodal learning works in degradation prediction. In particular, the crucial characteristics of each data modality were identified and show how multimodal learning addresses the inadequacies of unimodal data by providing complementary information from other modalities. Algorithms to create explainable HIs at the system level were developed, providing a comprehensive overview of the health status of systems and their components under various operational conditions, advancing towards smarter and more transparent maintenance decisions.
/
/
Nguyen, D. A., Jose, S., Nguyen, T. P. K., & Medjaher, K. Explainable multimodal learning for predictive maintenance of steam generators. In PHM Society Asia-Pacific Conference. 2023, 4, 1.
Nguyen, D. A., Nguyen, T. P. K., & Medjaher, K. Enhancing Trustworthiness in AI-Based Prognostics: A Comprehensive Review of Explainable AI for PHM. In: Tran, K.P. (eds) Artificial Intelligence for Safety and Reliability Engineering: Methods, Applications, and Challenges. 2024.
Nguyen, D. A., Nguyen, T. P. K., & Medjaher, K. Bridging XAI gaps in PHM: Concise review and novel dataset for health indicator construction and prognostics of multi-component systems. In PHME 2024 Conference. 2024.
According to the practical requirements of the factory of the future, this project combines the recent advancements in different domains to develop an explainable intelligent maintenance system (X-IMS) that enables both self-monitoring and decision-making support functionalities for connected manufacturing systems. The developed solution should allow automated construction of effective and interpretable health indicators for system continuous self-monitoring. They also integrate explainable intelligent algorithms for fault detection, diagnostic, and/or prognostics at the system level. Furthermore, embedded maintenance decision optimization algorithms, that can handle prediction uncertainties, component dependencies, and impacts of multiple maintenance activities will be developed. The optimal decision process obtained by the proposed intelligent algorithms should be explicitly conveyed to managers and therefore enable them to understand, trust, and effectively deploy the developed solution in practice. The performance of the algorithms developed in this project will be verified and highlighted by real industrial applications.
Project coordination
Thi Phuong Khanh Nguyen (Ecole Nationale Ingénieurs Tarbes)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
Partnership
LGP Ecole Nationale Ingénieurs Tarbes
CRAN Université de Lorraine
Help of the ANR 231,378 euros
Beginning and duration of the scientific project:
January 2023
- 42 Months
