Damage Identification of Prestressed Concrete Components Based on Machine Learning Optimization Algorithm and Piezoelectric Wave Measurement

Abstract

Prestressed concrete components have high crack resistance and stiffness, but they may suffer damage and lead to major accidents under adverse environments and extreme loads. The study uses machine learning algorithms to construct an intelligent concrete damage recognition model aimed at accurately assessing its health status. The piezoelectric wave measurement method is used to collect small wave signals from concrete. The improved backpropagation network is used to identify concrete damage characteristics in the signals, and the support vector machine is taken to correct the identification results. According to the results, the mean square error, coefficient of determination, and F1 score of the damage location recognition model constructed by integrating two classification algorithms were 7.962×10^-4, 0.9756, and 0.9836, respectively. For damage identification results, the mean square error, coefficient of determination, and F1 score of the research model were 6.548×10^-2, 0.9531, and 0.9925, respectively. In the environment with introduced noise, the recognition accuracy of the research model was 93.7%. The results indicate that the research method has higher accuracy and robustness in damage identification compared with other models, which can be applied for concrete damage detection in large buildings or long-term high load buildings.