Hammering Acoustic Analysis Using Machine Learning Techniques for Piping Inspection
In Japan, the deterioration of industrial plants built during the period of high economic growth in the middle of the 20th century has recently become a social concern. Corrosion under insulation (CUI) of piping in such plants is a pressing problem. X-ray and ultrasound inspections are conventional methods for detecting CUI; however, these methods are time-consuming and expensive. Therefore, rapid and low-cost screening techniques for CUI are required. We develop a hammering-type inspection robot system that moves inside the piping and records hammering sounds. Furthermore, we propose an acoustic analysis method to identify anomalous parts from the hammering sound using machine learning techniques. Using three testing pipes, we can successfully identify anomalous parts through acoustic analysis using a deep neural network as a supervised learning method. However, in practical piping inspections, the detection of anomalies without training data is required for further applications. Therefore, we investigate unsupervised learning anomaly detection using an autoencoder and a variational autoencoder and report the results.