Classification of PCCP-damage acoustic-emission signals using the ACW-GA-PSO-BPNN method

Breakage of prestressed steel wires and cracking in concrete structures are common failure modes in prestressed concrete cylinder pipes (PCCPs). This study employed prototype pipe wire breakage tests and small beam three-point bending experiments. Acoustic emission (AE) technology was used to systematically collect signals from cutting-induced wire breakage, corrosion-induced wire breakage, and concrete cracking. A three-class classification model was constructed using a back propagation neural network (BPNN), maintaining high accuracy across repeated tests and achieving a peak accuracy of 97.5 %. Key parameters identified through the mean impact value (MIV) method were statistically analyzed, and the results showed strong agreement with the mechanisms of signal generation. The model directly utilizes AE parameters, enabling end-to-end modeling. To enhance learning stability and generalization, a hybrid genetic-algorithm and particle-swarm-optimization (GA-PSO) procedure was introduced for hyperparameter tuning, which improved both accuracy and convergence efficiency. In addition, an adaptive class-weight (ACW) strategy effectively mitigated class imbalance. This research provides a feasible solution for intelligent identification of PCCP damage modes, supports timely risk warning and targeted repair guidance, and can contribute to improved safety in large water-conveyance projects. • An Adaptive Class Weight-GA-PSO-BP model was developed to classify PCCP damage. • An accuracy of 97.5 % was achieved in identifying wire breakage and cracking signals. • Model convergence was enhanced by a hybrid algorithm without sacrificing accuracy. • Class imbalance was addressed by an Adaptive Class Weight strategy. • Key parameters were identified via Mean Impact Value analysis.