Publication: Predictive real-time pneumatic Actuator leakage during cycle test Using machine learning approach.
No Thumbnail Available
Date
2025-08
Authors
Jaafar, Muhamad Aliff Ikmal
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Traditional cycle testing of Atmospheric (ATM) doors requires pausing for pressure decay tests due to different pneumatic setups, disrupting the process and delaying overall testing time. Previous studies have investigated leakage detection methods, such as using acoustic emission and wavelet transform for pneumatic leak detection and employing support vector machines (SVM) for diagnosing pneumatic actuator faults. However, these approaches are generally limited to offline analysis and do not support real-time monitoring. Furthermore, recent years have seen little progress in advancing leakage detection methods specifically tailored to high-cycle pneumatic systems, such as ATM doors. This study addresses the gap by performing real-time leakage evaluation using transient pneumatic pressure data and the position of the piston. This helps in integrating machine learning and Internet of Things (IoT)
technologies into a new cycle test system that continuously monitors pressure levels and door positions while storing data on a server and displaying it in the cloud for
remote monitoring. Data from ATM doors with varying Pressure Loss Ratios (PLRs) were collected and used to develop predictive models. K-Nearest Neighbors (KNN) achieved 93.86% accuracy for classification, while a multi-output Artificial Neural Network (ANN) demonstrated RMSE below 0.0001 and R² above 0.85 in regression tasks. The final system enables real-time leakage assessment and remote status tracking, offering a significant advancement in predictive maintenance for pneumatic systems.