Bayesian algorithm optimized deep model for multi-signal fault identification of hydraulic pumps

Hydraulic pump (HP) supplies the power for a hydraulic transmission system. It is critical for the fault diagnosis of HP to guarantee the healthy operation of systems. The rapid progress of artificial intelligence and measuring technique empowers data-driven methods. Diagnosis methods based on machine learning are limited in time-consuming parameter adjustment. Moreover, current methods based on one measured signal achieve fault diagnosis only by using limited feature information. To address the two problems, this research proposes an intelligent fault diagnosis method for HP. First, original signals from multiple sources are obtained using multiple sensors. Due to the effective solution of frequency aliasing and the good robustness to the background noise of the signals, an intelligent method combining an improved deep model and synchrosqueezing wavelet transform is further established. The introduction of Bayesian algorithm completes the adaptive selection of main parameters and contributes to reduce the computing time. The optimal model is selected by updating hyperparameter groups and evaluating the performance. The optimized diagnostic model presents good convergence and stability. The diagnostic accuracies achieve 96.69 %, 97.58 %, and 99.89 % for three signals. The proposed diagnostic method presents the effectiveness in mining fault characteristic information and intelligent identification of common fault types.