Deep Learning and Wavelet Transform Synergy for Hydraulic System Fault Diagnosis

Abstract

Effective detection ofInternal leakage internal leakage in two-stage hydraulic cylinders is essential for operational efficiency and minimizing downtime of hydraulic systems. Traditional methods often fail to detect early leakage, prompting the need for advanced techniques. This paper presents an approach for internal leakageInternal leakage detection in two-stage hydraulic cylinders by integrating wavelet transform and a convolutionalConvolution neural network (CNN)-based advanced deep learning technique. Pressure signals are acquired from the hydraulic cylinder operating under three conditions (no leakage, moderate leakage, and severe leakage). The raw signals are initially pre-processed using the Discrete Wavelet Transform (DWT) with the Daubechies 4 (db4) wavelet and level 5 decomposition to capture transient features. Following this, numerous statistical features like mean, standard deviation, entropy etc. from decomposed signals to characterize the pressure signals. SubsequentlyEfficient net, Efficient Net, a state-of-the-art CNN architecture, is trained with extracted features to classify the leakage conditions automatically. This proposed method demonstrated significant improvements in leakage detection accuracy compared to traditional approaches, showcasing the potential of combining wavelet-based preprocessing with deep learning model for condition monitoringCondition monitoring of mechanical systems. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.