Non-destructive evaluation of pipes by microwave techniques and artificial neural networks

Abstract

Near-field imaging based on an electromagnetic sensor has been widely used for nondestructive detection. An approach to detect the near-surface defects in pipeline coatings and dielectric pipelines is proposed. Based on the characteristics of resonant frequency shifts, a novel method using artificial neural network (ANN) is established to quantitatively evaluate circular-section shape defects in pipes, such as air bubbles in pipeline coating layers or qualitative characterize non-circular section-shape defects. The proposed method has three important modules: a new resonator for data acquisition, a signal-processing algorithm for data preprocessing, and an ANN for quantitative imaging. In the designed sensor, we extend the tip of the sensing ring and introduce an appending in the ring gap for high sensitivity. Simulations show that the sensor can detect a defect with a radius as small as 0.7 mm. The raw resonant frequency shifts obtained by the sensor scanning at an angle interval around the specimen first are preprocessed by curve fitting, sampling, and adaptive data interpolation or truncation. Then, using an ANN, the relationships among resonant frequency shifts, external radius of the specimen, and defect size are modeled for imaging of circular-section shape defects. Preliminary simulations and measurements illustrate the efficacy of the method. Consequently, a contactless, high-resolution, near-field imaging measurement based on sensor scanning for inspecting pipe structures is obtained.