Evaluation of local wall thinning severity in a metal pipe using guided microwaves

This study explores the applicability of guided microwave testing to assess the severity of partial circumferential pipe wall thinning (PWT). Numerical simulations revealed that the resonant frequency of reflected TM01 mode microwaves exhibited a negative correlation with the circumferential angle, depth, and length of PWT. Notably, high-order modes activated in the defect region produced frequency outliers, with corresponding frequencies decreasing as PWT depth or length increased. Experiments conducted using a 15 m long pipe yielded a frequency spectrum of the processed signals with a similar waveform to that of the simulation results, suggesting the feasibility of using simulation data to train the evaluation model. A one-dimensional convolutional neural network (1D CNN), built to automatically extract information from the frequency spectrum, achieved remarkable accuracy exceeding 96.7% in classifying the partial circumferential PWT severity in both simulated and experimental scenarios. Comparative analysis revealed that both 1D CNN and k-Nearest Neighbor classifiers, utilising full waveform information, outperformed traditional back-propagation neural networks relying on handcrafted frequency points. These findings highlight the critical importance of waveform information in PWT evaluation and present a novel approach for PWT severity assessment using guided microwave testing for long pipe inspection.