Fault Diagnosis of Rotating Machinery Based on FMEA and Zero-shot Learning

Abstract

Data-driven intelligent fault diagnosis is now a research hotspot. However, the fault data collected in actual working conditions is limited, which leads to a decline in the diagnostic ability of fault diagnosis models that rely on balanced data in actual engineering. Fortunately, the zero-shot problem for some fault classes can be solved by transferring zero-shot learning from machine vision to rotating machinery fault diagnosis. Inspired by this, we propose an attribute description method based on Failure Mode and Effects Analysis (FMEA) to solve the problem of semantic description of rotating machinery fault data, so as to be used for zero-shot fault diagnosis of rotating machinery. The framework analyzes the failure modes of rotating machinery based on FMEA, and establishes a fault attribute dictionary. After that, the attribute classifier is trained using the multi-domain features of the data. Finally, the fault data of unknown class is diagnosed based on Euclidean distance. The effectiveness of the framework is verified by public datasets. This framework provides a new perspective for zero-shot fault diagnosis of rotating machinery.