Predictive Maintenance of an Hydraulic System Using Spiking Neural Networks

Abstract Predictive maintenance is a vital component of modern machinery management, allowing operators to anticipate and prevent equipment failures. In this study, we introduce a novel approach using spiking neural networks (SNNs) for predictive maintenance of hydraulic systems. While prior research has used traditional feature engineering and machine learning or artificial neural networks for predictive maintenance of hydraulic systems, this study utilizes SNNs instead. SNNs are event-driven artificial neural networks that can process real-time data with high accuracy, making them an ideal tool for predictive maintenance. The proposed approach leverages historical sensor data to accurately predict system failures in hydraulic systems. The public data set used was experimentally obtained with a hydraulic test rig with 17 sensors distributed throughout the system. The system had four independent modes of failure with varying degrees of severity and a fifth parameter indicating whether the system had reached stable conditions or not. The pre-processed data was used to train the SNN, which was then tested for its ability to predict system failures. The results showed that the SNN was able to accurately predict system cooling condition failures with a high degree of accuracy(99.46%), precision(99.73%), Recall(98.95%), F1-Score(98.95%), demonstrating its potential for use in predictive maintenance applications. In conclusion, this study highlights the significance of SNNs as a valuable tool for improving the reliability and efficiency of hydraulic systems. With further development, SNNs are poised to become the gold standard in predictive maintenance.