An Intelligent Method for Identifying and Analyzing PCCP Wire-Break Signals Based on Prototype Testing and an Ensemble Deep Learning Model

Wire breakage is the predominant cause of failure in prestressed concrete cylinder pipe (PCCP) within water diversion and transfer projects. This paper employs deep learning models to extract signal features and identify wire breakage types. A prototype test of wire breakage was conducted on a buried PCCP (DN3400) segment with a length of 5 m. During the tests, varied internal water pressures were applied, and wire cutting and electrochemical corrosion were used to simulate wire breakage signals. Distributed fiber optic sensors facilitated real-time monitoring of signals. Meanwhile, the signals caused by knocking and concrete core cracking were also monitored. Two-dimensional characterization of wire-broken signals was achieved through synchrosqueezing wavelet transform, with multiscale features extracted using several single deep higher recognition accuracy for wire-broken signals compared to single models, achieving 98.4% accuracy. However, differentiation of wire-broken signals under different internal pressures proves challenging due to limited signal variability. In this research, the wire breakage identification model is established using an intelligent deep ensemble model, offering new methods for early warning analysis in long-term PCCP operation.