Since clogging of piping affects plant operation, there is a demand for nondestructive inspection to determine the internal condition of piping, but no method has been established for this purpose. In previous studies, the clogging condition was predicted by a CNN using the frequency characteristics obtained from an AE sensor. However, overfitting due to the small amount of training data and the lack of a clear rationale for evaluating the prediction results were pointed out as issues. Therefore, in this study, the vibration characteristics obtained by finite element analysis were converted to the frequency domain using FFT. Then, we attempted to construct an interpretable surrogate model by learning and predicting features extracted in multiple bands, such as peak values from frequency characteristics, using XGBoost..As a result, the prediction accuracy for clogging presence was about 90%, suggesting that this method is applicable to the diagnosis of the internal condition of piping. We would like to discuss the effectiveness of this method.
Nakahara et al. (Wed,) studied this question.