An adaptive deep learning model towards fault diagnosis of hydraulic piston pump using pressure signal

Abstract

Hydraulic pump is a critical component of a hydraulic transmission system. Its normal and stable operation will directly affect the reliability of the whole system. Owing to the complexity and concealment of fault information, it is crucial to explore an effective method for fault diagnosis of hydraulic piston pump. The obvious change of vibration signal can be observed when different fault types occur, and the implied characteristics of pressure signals are also varying. In this research, an adaptive convolutional neural network (CNN) is constructed based on the ontology representation of pump. First, time–frequency characteristics of pressure signals are obtained via continuous wavelet transform. Second, a deep CNN is established by setting initial hyperparameters. Third, Bayesian optimization is employed to achieve automatic learning of main important hyperparameters to construct an adaptive CNN. The proposed method has better performance for fault diagnosis of hydraulic piston pump compared with other modern intelligent methods.