Study on Safety State Pattern Recognition of Reinforced Concrete Members with Machine Learning Method

The pattern recognition of structural safety states is of great significance for the performance evaluation and maintenance of existing structures and post-disaster damaged buildings. Under the general framework of machine learning, this study adopts the BP neural network as the implementation of the structural safety classification model and proposes a method of safety state pattern recognition for reinforced concrete flexural members. Based on the MATLAB platform, the method takes crack characteristic parameters extracted from the surface of concrete members as inputs and the safety state classifications defined in the FEMA-356 code as outputs, thereby establishing a mapping relationship between crack features and the corresponding safety states of the members. The experimental results show that the constructed neural network model achieved a recognition accuracy of 90.9% for safety state classification, demonstrating good pattern recognition capability. In addition, this study established a damage database for cantilevered reinforced concrete flexural members and further validated the generality and applicability of the neural network method for safety state recognition across different structural members. The results indicate a recognition accuracy of 91.7%, enabling the safety state recognition of members with different sizes, strengths, and reinforcement configurations. The proposed method enriches the technical means of structural health monitoring.