Facilitating Energy Monitoring and Fault Diagnosis of Pneumatic Cylinders with Exergy and Machine Learning

Abstract

Pneumatic systems are widely used in industrial production sectors. Increasing penetrations of Intelligent Manufacturing and Green Manufacturing are highlighting the drawbacks of pneumatic technology in terms of particularly low energy efficiency and low-level fault diagnosis intelligence. Here we propose that a combined energy-based maintenance and fault diagnostic approach for pneumatic systems could be a game-changer for pneumatics. In this study, a pneumatic cylinder with internal and external leakages is examined and a typical pneumatic experimental system is built. Exergy is adopted for evaluating the available energy of compressed air. Data-driven machine learning models, SAE + SoftMax neural network model and SAE + SVM model, are developed for fault detection and diagnosis. By comparing different machine learning methods with various pressure, flowrate, and exergy data, it is found that the diagnostic accuracy when using pressure and flowrate data is highly dependent on operating conditions, while the diagnostic accuracy when using exergy data is always high regardless of operating conditions. This indicates the promise of developing an exergy-based maintenance paradigm in pneumatic systems. Besides, with exergy and machine learning, more downstream faults can be detected and diagnosed with fewer upstream sensors. This study is the first attempt to develop an exergy-based maintenance paradigm in pneumatic systems. We hope it could inspire the following investigations in other energy domains.