Data and prior knowledge driven FMEA severity assessment via deep attention networks
Über dieses Buch
Traditional Failure Mode and Effects Analysis (FMEA) suffers from subjectivity and static assumptions in severity evaluation, especially when applied to complex, data-rich industrial systems. To overcome these limitations, this paper proposes a new dynamic FMEA severity quantification hybrid framework that integrates real-time equipment status monitoring data with domain prior knowledge through deep learning. A deep attention network is proposed to capture long-term dependencies in multivariate sensor streams, while adaptively processing key health status features through attention mechanisms. Engineering prior knowledge is utilized to establish fault severity degrees, integrating real-time equipment condition monitoring data to quantify FMEA severity, thereby enabling a dynamic quantitative representation of FMEA severity. Experimental verification on real industrial datasets shows that the proposed method can achieve dynamic FMEA severity assessment of equipment. This work bridges the gap between predictive health monitoring and equipment risk assessment, providing a practical solution for reliability centered maintenance in data rich industrial environments.
