Neural network method in bridge condition assessment by B-TBU model
1. School of Highway, Chang′an University, Xi′an, Shaanxi 710064, China; 2. School of Civil Engineering, Southwest Jiaotong University, Chengdu, Sichuan 610031, China
Abstract:The present commonly-used bridge condition assessment methods were easily influenced by human subjectivity and arbitrariness, and the impact of historical evaluation data on the current state was also neglected by these methods. To solve the problem that the current true state of bridges could not be accurately reflected, the B-TBU model method was proposed based on Bayesian inference characteristics of taking prior information influence into account. The influence of historical evaluation data on the assessment of current state was considered, and the bridge state over the past twenty years was reevaluated. The evaluation results show that the state evaluation accuracy can be significantly improved with the bridge annual state evaluation accuracy increased above 90%. The BP neural network and the ELM neural network were preliminarily introduced into B-TBU model to train B-TBU model. The results show that high evaluation accuracy can be achieved via the neural network method with the annual state evaluation accuracy of about 80%.