Condition Monitoring of Hydraulic Systems Using Multi-output Classification Convolutional Neural Networks

Abstract

Hydraulic systems are crucial in aviation and construction applications, efficiently transmitting heavy loads with minimal effort. Condition monitoring techniques are essential for optimizing the quality and performance of hydraulic systems. It serves as a tool for decision-making in maintenance activities, enabling more effective and informed operations. In modern industrial settings, monitoring machinery has become critical for enhancing the cost-efficiency of hydraulic systems. This paper introduces a deep learning (DL) framework utilizing convolutional neural network (CNN) for condition monitoring of hydraulic systems with raw data for health prediction of multiple components such as valves, cooler, accumulators and pump. This method uses CNN to analyze raw time-series sensor data with varying sampling frequencies (100, 10, and 1 Hz) in which the input matrix of the CNN model is a 3-D matrix and predicts the functional state of the hydraulic system. Demonstrating the effectiveness of CNNs in handling multi-output classification tasks by achieving high accuracy across all targets, and evaluated with accuracy, recall, precision, and F1-score and achieved an overall accuracy of 98.5%. This study illustrates the effectiveness of DL techniques in condition monitoring of hydraulic systems for multi-target classification tasks. © 2025 Elsevier B.V., All rights reserved.