Deng, WeiKun (University of Technologie Tarbes Occitanie Pyrénées), Nguyen, Thi Phuong Khanh (University of Technologie Tarbes Occitanie Pyrénées), Medjaher, Kamal (University of Technology Tarbes Occitanie Pyrénées (UTTOP)), Gogu, Christian (Institut Supérieur de l'Aéronautique et de l'Espace), Morio, Jérôme (ONERA)

E2E Gated-Mamba for Cross-Scenarios Prognostics

Scheduled for presentation during the Invited Session "Advanced Prognostics and Health Management of Mechatronic Systems: Methods and Applications" (WeAT1), Wednesday, July 16, 2025, 10:20−10:40, Room 105

Joint 10th IFAC Symposium on Mechatronic Systems and 14th Symposium on Robotics, July 15-18, 2025, Paris, France

This information is tentative and subject to change. Compiled on July 16, 2025

Abstract

Data-driven Prognostics and Health Management (PHM) models face challenges in generalizing across diverse industrial scenarios, particularly in meeting the“4Cs” requirements: cross-device, cross-prediction targets, cross-sequence lengths, and cross-input physics quantities. Existing models often require retraining for each new application, limiting scalability and flexibility. Addressing these issues, we propose Gated-Mamba, a novel PHM backbone based on a Selective State Space Model (SSM) with a unique gated unit design, enabling end-to-end (E2E) processing across various PHM tasks without retraining. By integrating dynamic input-gated mechanisms and a unified encoding design, Gated-Mamba efficiently handles variable-length sequences and diverse physical inputs, while state-space updates ensure the retention of relevant degradation information within a single architecture. Validations on mixed datasets, including bearings, batteries, and tool wear, show that Gated-Mamba outperforms state-of-the-art models in prognostic accuracy and computational efficiency. This offers a scalable and efficient solution for predictive maintenance and enhanced reliability in complex mechatronic systems.