Reliability Analysis Using Evolutionary Neural Network And Its Application In Engineering

Implicit performance function problem has always been a hot issues in structural reliability theory. Recently reliability analysis using artificial neural network (ANN) obtained some research achievements. However, two drawbacks still exist. First, traditional ANN employs error back propagation (BP) algorithm.BP algorithm is local convergence, which is easily lead to failure when training. Second, how to construct ANN samples has been neglected by many researchers. The two drawbacks encounter prolems when the reliability analysis is operated. To address these issues, this paper presents some studies as follows:(1) The evolutionary neural network (EANN) model is proposed.the main feature of this model is that it combines the genetic algorithm and BP algorithm while training. The training algorithm is essentially a kind of optimization algorithm. GA is good at global search but inefficient in fine-tuned local search, while traditional gradient descent algorithm on the contrast. EANN employs GA to search for a near-optimal set of initial connection weights and then use BP to perform local search from these initial weights. Compared with traditional ANN models, the EANN models are more robust and accurate.(2) Samples are very important for ANN. How to construct ANN samples has been neglected in the engineering application. The problem is definitely discussed in this paper. The design of experiment (DoE) is introduced to direct ANN samples design. Based on the summarization of the research productions, the uniform design (UD) is recommended for the ANN sample design.(3) A new reliability analysis method using EANN is proposed. The examples demonstrate the effectiveness and correctness of the proposed method. And this method is more robust and accurate than the traditional method.(4)The proposed reliability analysis method using EANN is combined with Bishop method, which creats a new slope reliability analysis method. Then, this method was successfully applied to evaluate the reliability of an open-pit slope in Guang Dong province, China. Bishop method is popular for its efficiency and accuracy. And the new slope reliability analysis method makes more values.(5) All the proposed method platform are programmed based on MATLAB. These codes have certain value for researchers and engineers.