Remote anomaly detection of pipeworks using One-Class SVM and wireless microphone

Abstract

A remote anomaly detection experiment of pipeworks using one-class support vector machine (One-Class SVM) and a wireless microphone device was carried out, as an example of an abnormal noise detection system using machine learning. In the experiment, seven kinds of acoustic signals were measured remotely using the wireless microphone device, under the condition that water flowed in the aluminum alloy pipe and burst waves were given from the attached piezo element. High-pass filters with different cutoff frequencies were applied to measured acoustic signals, and features such as wave crest factor and peak frequency were extracted from the time and frequency domains of the time-divided waveform after filtering. An anomaly detection model was constructed using two features that were reduced by principal component analysis (PCA) after standardized as training data, and the existence of burst waves in test data was diagnosed. As a result, the accuracy of anomaly detection was sufficient to diagnose the presence of burst waves in most cases, which shows this technique is very promising for remote anomaly detection.