Analysis On The Law Of Short Fatigue Crack Based On Genetic Algorithm And BP Neural Networks

With the development of science and technology, the power mechanisms and engineering equipment are in pursuit of elevated temperature, high pressure and high speed. The mechanical parts are working under cyclic loading with bad conditions. It makes the reliability of the material appears to be more prominent. The crack is important parameter to describe the fatigue damage in the study of fatigue material. Scientific researchers have done considerable study on the initiation and propagation of metal fatigue crack. The study on long cracks is close to maturity. The results from the present investigation suggest that the stage of short crack sometimes accounting for the most of total fatigue life. Short crack behavior is very different compared to long crack, but the study on the law of short crack is yet on the initial stage. There are an abundance of advantage of study on fatigue short crack, such as, improving metal Fatigue damage mechanisms, life forecast, structural safety study, all of which could provide an important theoretical basis.The evolutions of short cracks always attract the material field's concern. A large number of models and prediction equations have been proposed. The most of them confront many problems so that the applications have been limited and no quantitative equation is widely accepted so far. Explicit equations can be avoided establishing with the introduction of GA-BP neural network. The law that the data imply is directly pulled out resulting in effectively resolving these problems.It is the main object of the experiment in the paper.20# steel is carried out fatigue short crack material test for low cycle at high temperature and complex stress states. By combination of the experiment technique and simulations, the material under various stress state is experimentally researched. The leading short crack initiation and propagation behavior is tracked and recorded by experiment interrupted and replication method. MATLAB and other software is adapted to process data consist of short crack growth rate and short crack density. The reason of crack growth rate changing is obtained though the analysis of experimental data.BP neural network is considered especially suitable for dealing with discrete, noisy and incomplete data. Introduction of genetic algorithm can remedy the defects of the BP neural network, and reduce the convergence time and errors at the same time. Genetic algorithm-BP neural network is used to simulate short fatigue crack growth rate and crack density in this paper. The results indict simulated results match experimental data well and small error is achieved. Through the whole process simulation for crack initiation and propagation, it proves that this method can be used to simulate the whole process and reflect crack propagation law.GA-BP neural network is a better tool to process the data and describe the fatigue crack law of fatigue short crack evolution with higher accuracy. Comparison between the BP neural network simulation results and GA-BP neural network simulation results show that genetic algorithm improve the property of BP neural network in convergence time and prediction accuracy. It confirms that the genetic algorithm-BP network used in the evolution law research of short fatigue crack is feasible.