Application Of Improved Back-Propagation Neural Network To Short Fatigue Crack Evolution

Generally speaking, the development of fatigue is composed of cyclic deformation, the initiation and propagation of the micro-cracks and crack propagation to fracture. Recently more and more attention is paid to the research on the evolution of short fatigue crack in the field of fatigue fracture, but there is still not a perfect theory. Understanding the initiation and development of short fatigue crack which is the starting point of fatigue damage is not only help understanding the microscopic mechanism of fatigue crack evolution and revealing the physical mechanism of fatigue damage, but also help exploring the influence factors on fatigue life and providing technical support for the material design and engineering structural optimization. Ultimately the fatigue life prediction technique and assessment methods are established, which provide a theoretical basis for the evaluation of structural reliability and security. With the short fatigue crack test of20steel at high temperature low cycle under complex stress state, the following work is carried out in this paper:(1) The fatigue damage mechanism for20steel at high temperature is revealed through the experimental study on the smooth surface specimens.Research shows that short fatigue cracks initiate the ferrite-pearlite grain boundaries at high temperatures, individual cracks originate from the slip bands inside the ferrite grains. In the phase of the micro structural effect growth, the short crack propagation rate exhibits deceleration clearly. After the initiation of short fatigue cracks, the deceleration happens while the crack tips meet the pearlite banded structure. Once the dominant short crack overcomes the resistance of pearlite banded structure, the intense acceleration happens and the propagation of short fatigue crack enters the phase of short fatigue crack propagation. In this stage the dominant short crack format and develop to long crack which lead to the failure fracture. During the initiation and propagation of short fatigue crack, the density of short crack increases dramatically in the beginning, which to be maximized when the life ratio N/Nf is about0.3-0.4, and then declines slightly and tends to a steady trend. Average angles of short fatigue crack perpendicular to the direction of maximum principal stress axes.(2) By Introducing a self-adaptive strategy, improved Back-Propagation neural network is presented. The valid description of short fatigue crack behavior is also proposed. The density and the propagation rate of short fatigue cracks is simulated respectively with improved Back-Propagation neural network.A way that improves Back-Propagation neuralnetwork aiming at evolution of short fatigue crack, which combine with a self-adaptive strategy is shown in this paper. This method optimizes the weights and thresholds of the BP network, and aggregates the characteristics of the local precise search of the BP network and the global optimization of the improved Genetic Algorithm, which integrates more factors and reflects complicated relation. Comparing the results of the experiment of short fatigue crack for low cycle under complex stress at high temperature with the simulation results of improved Back-Propagation neural network, it is proved that the method is feasible and accurate.(3) The fractal theory is used in this paper and the fractal feature of fatigue damage was extracted. As the input parameter, improved Back-Propagation neural network is used in the fatigue life prediction. Comparing the experimental results with the simulation results of improved Back-Propagation neural network, it is proved that the method is feasible..