Stress prediction of bridges using ANSYS soft and general regression neural network. (June 2022)

• Record Type: Journal Article
• Authors: Wu, Ying; Zhang, Lihua; Liu, Hongbing; Lu, Pengzhen
• Is Part Of: Structures. Volume 40(2022)
• Page Start: 812
• Page End: 823
• Publication Date: 2022-06
• Subjects: Bridge engineering -- Dynamic load test data -- Intelligence algorithm -- Regression analysis -- Static behavior prediction; Structural engineering -- Periodicals; 624.1
• DOI: 10.1016/j.istruc.2022.04.068
• Languages: English
• ISSNs: 2352-0124

Abstract

Abstract: Bridge managers are often eager to identify the carrying capacities and service performance characteristics of bridge structures. In view of the massive number of bridges managed and maintained in China, providing a quick evaluation of bridge performance bridges has become an urgent challenge for bridge engineering. Aiming at many adverse problems of bridge load tests, such as their high costs, long time consumption, low efficiency, impact on traffic, and damage to bridge structures, a new method for bridge static behavior prediction is proposed herein, based on dynamic load test data. The method adopts a generalized regression neural network algorithm with fast convergence and a nonlinear approximation, determines the sensitive parameters based on a full-bridge sensitivity analysis, forms training samples using a uniform design sampling method, and establishes intelligent agent model. The initial finite element analysis model of the bridge structure is updated using the proposed proxy model and dynamic load test data, and the static behavior of the existing bridge structure is accurately predicted based on the updated finite element model. To verify the correctness of the proposed method, a practical bridge is considered as the research object, and the static response of the bridge structure is predicted. The predicted results are consistent with the static load test results for the bridge, verifying the correctness of the proposed method. The results show that the maximum error between the theoretical calculation result and measured value is only 4.34% and the minimum error is only 3.52%. The strain prediction results are consistent with the measured results, and the modified theoretical analysis model provides high accuracy for strain prediction results under different working conditions. The application in practical engineering shows that the prediction results under different load conditions are adaptable and highly accurate.