Compound Fault Separation and Diagnosis Method Based on FSA-CNN and DAN
Jiechao Dong, Liping Yan, Yuanqing Xia · 2021 CAA Symposium on Fault Detection, Supervision, and Safety for Technical Processes (SAFEPROCESS) · 2021
With the industry becoming larger and more complicated, the technology of fault diagnosis becomes more and more important. Due to the strong coupling and non-stationarity of the compound fault, the existing fault diagnosis methods cannot accurately identify all single faults’ detailed information contained in the compound fault. This paper proposes a compound fault separation and diagnosis method based on FSACNN and DAN. Firstly, in order to highlight certain frequency segments, frequency segment attention module is added to CNN. Secondly, a compound fault feature separation framework based on DAN is proposed, which can separate compound fault to two fault components accurately. Thiredly, a signature matrix is introduced into ELM to improve the performance of the classifier. Finally, ablation experiments are designed to prove the advantage of the proposed method.