Expected attribution prior guided interpretable framework for machine fault diagnosis

Open Webpage

Jie Li, Junwei Gu, Yu Wang, Yanyang Zi, Mingquan Zhang, Haijun Zhang

Published: 2026, Last Modified: 18 Nov 2025Expert Syst. Appl. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A novel diagnostic framework EAP-IF integrates expert logic to reduce attribution uncertainty.•An expected attribution prior loss encodes expert knowledge to enhance model interpretability.•A disentangle-from-normal architecture separates fault and normal signals for better diagnosis.•The network decodes decision logic into expert-interpretable and verifiable diagnostic rules.•EAP-IF outperforms state-of-the-art methods in both interpretability and generalization.