Shear Capacity And Sensitivity Analysis Of RC Deep Flexural Members Based On BP-ANN Model

Members are referred to as reinforced concrete deep flexural members when the beams’ span-depth ratio is greater than 5, the major failure modes is shear failure. As the influence factors members controling shear strength of the members are various and complex, the academia has not achieved the unified and mutual recognition about the failure mechanism and calculation methods. As a result, the problem of einforced concrete deep flexural members’ shear strength remains to be further research.In recent years, the research of artificial neural network is on the rise. Based on the data collected in past, its precision can achieve satisfactory degree through the self learning process, and don’t need an accurate failure criterion and the bearing capacity of the conduction theory. With the development of science and technology, this kind of method that is simple and easy to calculate and can keep learning to achieve a higher precision is more and more received by the researchers.Based on 271 sets of the bending shear test data collected by our research group, an artificial neural network model calculating the shear strength is established by extracting the main influencing factors. Based on the established model, we did the single factor analysis and orthogonal test analysis to get the change law and degree of the influence factors against to the shear bearing capacity.This article also introduced the foreign popular method of sensitivity analysis. The global sensitivity analysis method was introduced into civil engineering field for the first time, and applied to several main computing model involved in this paper to the sensitivity of the independent variables for each model. Based on calculating the sensitivity of variables, it can be found the main influence factors of the model. Liminated small sensitivity’s variables to optimize the calculation model of the artificial neural network. Through Contrasting we found that the before and after optimization model don’t have too big difference on the accuracy of the calculation results and discreteness, but the optimized model’s calculation and the complexity is reduced greatly. The main works accomplished in this paper were showed as follow:(1) The research status of deep flexural members and the artificial neural network concept were introduced, and analyzed several main influence factors of reinforced concrete deep beam’s shear performance. List the calculation formula of deep flexural members’ shear strength from the world’s leading countries respectively.(2) the paper introduced the working principle of artificial neural network model and itsmodel selection method. Based on the collected 271 sets, we used the Matlab software to training for good neural network, and it has been established. Through the better BP- ANN model we did the single factor analysis and orthogonal test design analysis.(3) Through a variety of global sensitivity analysis method, analysed the influence factors sensitivity for each model and the experiment. Based on sensitivity calculation results,we optimized the established BP-ANN model, and the results proved that the optimized model and old model has the same performance, but the complexity is decreased obviously.Through the comparison between selected BP- ANN model、the four standard and the test results, the feasibility and precision of BP- ANN models meet the requirements, BP-ANN models can be used for analysis members’ shear performance. At the same time, we verified the sensitivity’s operability in shear performance influence factor analysis and model optimization.