Ultrasonic Fouling Detector Powered by Machine Learning

Abstract

Guided waves can be used to monitor structural health in industrial pipelines, and e.g. allow detection of accumulated precipitation on the surface of pipe. Propagation of guided waves in a tubular structure carrying possible fouling can be separated from a clean structure due to variation in wave propagation properties at the fouled area. In addition, multiple propagation paths around the tubular structure allow locating the fouled areas. In this study, we obtained dispersion curves of a tubular structure loaded with a local fouling layer of different thickness by using numerical simulations. We combined the dispersion curve information with simulated and measured times-of-arrival of guided wave propagation to second order helicoidal paths and used a Gaussian Process machine learning approach to estimate location of fouling on a steel pipe.