A robust deep residual learning approach for early fault detection, diagnosis, and condition monitoring of hydraulic systems

Abstract

Hydraulic systems are widely used in industrial applications due to their reliability and high power density, but they are also prone to faults that can affect performance and safety. Accurate and timely fault detection is therefore essential for maintaining system efficiency. This study presents a deep residual learning-based model, referred to as DRL-HFDM, for intelligent fault detection in hydraulic systems. The proposed approach utilizes multi-sensor data, including pressure, temperature, and vibration signals, to capture system behavior under different operating conditions.The dataset used in this study was preprocessed and analyzed using both bivariate and multivariate techniques to understand the relationship between sensor parameters and fault conditions. The model was then evaluated against several advanced architectures, including Transformer-based, CNN – GRU, SE-ResNet, attention-based, and CNN – BiLSTM models. The experimental results show that the proposed DRL-HFDM model achieves an accuracy of 97.1%, precision of 96.7%, recall of 96.4%, and F1-score of 96.5%, outperforming all compared models.In addition to improved predictive performance, the model demonstrates stable training behavior and efficient handling of multi-sensor inputs. The results indicate that the proposed approach is effective in identifying fault conditions and can be applied in real-time monitoring systems. This makes it a suitable solution for industrial applications requiring reliable and accurate fault detection.