Development of a new blockage localization method in the piping system using nonlinear transformation of transverse vibration signals and Explainable AI (XAI)

Abstract

Blockages or foreign objects within the piping system can obstruct fluid flow and may lead to structural failure, thereby highlighting the need for early detection technologies. Although vibration-based approaches have been extensively studied to identify the locations of blockages, accurately determining their positions across different pipe geometries remains challenging. This difficulty is further compounded by uncertainties in parameter estimation when interpreting the correspondence between simulation models and real systems. To address these issues, this study presents a novel method that nonlinearly transforms the vibration signals of an experimental model with reference mode to a simulation model and leverages explainable AI (XAI) to identify blockage locations within pipe models. To implement the proposed approach, transverse vibration signals are obtained from both simulation and experimental models and analyzed in terms of their frequency response functions (FRFs). A nonlinear transformation is applied by scaling and shifting the eigenfrequencies of the experimental model relative to those of the simulation model, targeting specific eigenfrequencies. The transformed experimental and simulation signals are evaluated for similarity using the Pearson correlation function and subsequently transformed into the Markov Transition Field (MTF) data. The MTF data are then classified and identified using the CNN-based Grad-CAM algorithm. To validate the proposed method, several case studies are conducted on I-type and T-type pipes.