Unsupervised Fouling Reconstruction in the Pipe Bend

Guided wave tomography allows the investigation of extended areas with a limited number of measurements, typically using transducer arrays. Most imaging methods rely heavily on the material properties such as dispersion curves, that in the case of fouling deposition are not always known. In the case of complex shaped structures, geometrical anisotropy can bring additional complexity. Here we present an unsupervised machine learning approach based on the Gaussian process to detect and characterize the fouling in a pipe bend that relies only on the difference between clean/healthy and fouled/damaged measured signals.