In large-scale factories and power plants, clogging in metal pipes poses a significant risk for serious accidents, making regular inspections essential. Traditionally, non-destructive inspection methods such as endoscopy and radiographic techniques have been commonly used, but the operational challenges incur significant time and effort. Inspection methods using electromagnetic waves, as proposed in previous studies, offer the advantage of rapid defect detection regardless of pipe length. However, inspection methods using electromagnetic waves often rely on experience-based threshold settings, which limits detection accuracy improvements. In this study, we propose a defect detection method that uses machine learning to analyzes the large datasets obtained from electromagnetic field simulators. Specifically, converting the propagation characteristics derived from S-parameters into statistical values significantly improved the accuracy of the predictive model. This advancement enabled the detection of small defects as tiny as 2mm, which were previously difficult to identify using conventional methods, thereby increasing the accuracy and efficiency of defect detection. Through this research, we demonstrate the effectiveness of machine learning-based defect detection, which was not thoroughly explored in previous studies.
Hiroki et al. (Tue,) studied this question.