文献类型：期刊文献

英文题名：A BASELINE-FREE MULTIMODAL PIPE DAMAGE IDENTIFICATION METHOD BY MACHINE LEARNING

作者：Wang, Mingyuan[1];Gu, Yiqing[1];Li, Yaokai[1];Jia, Jiuhong[1];Tu, Shan-Tung[1]

机构：[1]East China Univ Sci & Technol, Key Lab Pressure Syst & Safety, Minist Educ, Shanghai, Peoples R China

年份：2025

卷号：32

期号：4

起止页码：1

外文期刊名：METROLOGY AND MEASUREMENT SYSTEMS

收录：;EI(收录号：20260419947382);WOS:【SCI-EXPANDED(收录号:WOS:001696287200005)】；

基金：This work is supported by the National Natural Science Foundation of China [Grant no. 52575175] .

语种：英文

外文关键词：regional resonance pair; multiple damage; line structured light; multimodal; vibration modal analysis

摘要：Structural Health Monitoring (SHM) of pipe infrastructures is of paramount importance to prevent catastrophic failures induced by defects such as corrosion. Conventional damage identification methodologies are frequently faced with challenges, including baseline dependency, limitations inherent in single-sensor data, and considerable economic expenditure. This paper presents a novel, baseline-free, multi-modal damage identification methodology developed for Level 3 assessment of multiple damages, encompassing their detection, localisation, and quantification. Initially, Level 1 damage identification is accomplished through observation of the Regional Resonance Pair (RRP) phenomenon. Subsequently, potential damage regions are predicted by a Multilayer Perceptron (MLP) model that uses vibration modal frequencies, generating a Macro-F1 score of 0.8131 on the test set; this prediction is then integrated with a high-precision local point cloud, acquired via Line Structured Light (LSL) technology, to achieve precise Level 2 damage location, with a reported error as low as 1.78%. Following localisation, Level 3 quantification of the damage is performed using point cloud registration, fusion, and voxelisation techniques, enabling accurate prediction of damage volume with a quantification error of merely 2.47%.