Diagnostics Based on Acoustic Distributed Sensor Data and Machine Learning

Abstract

Accurate real-time diagnostics of high-tech systems are becoming more and more important. Therefore, the potential of distributed acoustic sensors in combination with machine learning for contactless diagnostics of machine performance has been investigated. Hereto, frequency response data of a brass plate has been gathered through experiments and a finite element model. In order to investigate the possibility of identifying the locations and weight of the masses, Support Vector Machines and Random Forest algorithms have been trained with experimental and numerical data. The Random Forest algorithm shows promising performance with short computational time, easy application, 95% accuracy and relatively easy understandability.