Contribution
Air Leakage Detection for Pneumatic Drives using MEMS Microphones and Machine Learning
* Presenting author
Abstract:
Some defects in industrial machines and pipe systems can be detected by experts based on acoustic events. The previous success in speech recognition motivates the ongoing development of machine learning methods for industrial acoustic measurements.This contribution introduces a concept to detect air leakages at piston rod seals of pneumatic drives at low cost and with small dimensions. It describes how these leakages can be detected with machine learning. For this approach, MEMS microphones are installed close to the seals, and their leakages are predicted by extracting interpretable features and using a supervised regression method. The use of MEMS microphones offers economic and integrative advantages.The first focus of this investigation lies on the interpretability of the results, which motivates the use of a supervised “Feature Extraction, Selection, and Regression” (FESR) approach. Extracting physically meaningful features followed by feature selection provides tools for interpretability. The second focus lies on the robustness of the method. Due to individual damages, quantitatively identical leakages may appear differently in the measurement data. This should enable broad applicability in the field.