Reliability study of long-term service concrete bridges based on improved bayesian neural network

Abstract Long term use of concrete bridges faces problems such as aging and decreased structural performance, posing significant safety hazards to their use. To address this issue, a long-term reliability analysis method for concrete bridges based on an improved Bayesian network is proposed. Hierarchical and expert methods are used to analyze the influencing factors of long-term service bridges and to construct evaluation subsystems. The innovation of the research is to use a dynamic BN network to describe the relationship between the characteristics of different factors and improve the accuracy of risk correlation analysis. Meanwhile, the study considered the dynamic changes in network inference and adopted the Kalman filter model as the inference decision method to ensure the effectiveness of problem analysis. Meanwhile, the study used an improved Bayesian network to construct a reliability analysis model to analyze the degradation and reliability of bridge performance to achieve the analysis of bridge health status. In the model performance test, the research model performed the best in accuracy, with an accuracy of 96.25% in multi-scale data. In forward reasoning, the inferred results of research techniques were close to the measured values, such as a safety failure rate of 23%, which was consistent with the measured values. In the analysis of bridge performance degradation, research techniques could accurately evaluate the structural performance of bridges under different working conditions. For example, in working condition point 2, the measured value was 651 KN·m, while the research technique was 653 KN·m, which was superior to traditional techniques. In addition, the research model predicted the degradation of bridge performance. The results showed that the bridge could experience severe structural degradation in 12 years. Effective maintenance measures needed to be taken for the bridge to ensure its safety. Finally, a comprehensive comparison of multiple technologies was conducted, and it was found that the bridge detection model required 0.56 days, which was better than the manual 3.51 days, and reached 99.56% in the applicable detection scenarios. It can be concluded that this technology can effectively analyze bridge safety and meet the requirements of bridge engineering operation and management. This study will provide technical support for the condition assessment and maintenance management of bridges.