基于改进神经网络的疲劳裂纹扩展速率预测

Prediction of the fatigue crack growth rate based on improved neural network

裂纹的萌生与扩展是一个复杂的非线性动力学过程,裂纹扩展速率具有非线性动力学系统的混沌现象和自组织特征。将BP神经网络和L-M贝叶斯正则化算法相结合,可使BP神经网络在推广能力、收敛速度和逼近精度上能够获得很大的提高。文章利用16MnR钢CT试样实测数据进行网络训练,训练好的神经网络可以对该材料的疲劳裂纹扩展速率进行较为精确的预测。

The initiation and propagation of fatigue cracks are a nonlinear dynamic process, and the fatigue crack growth rate is characterized as chaos and self-organization. The Levenberg-Marquart （L- M） Bayesian regularization algorithm is combined with the back-propagation（BP） neural network to make the BP network achieve better generalization, faster speed of convergence and higher learning accuracy. The network is trained by using the experimental data of 16MnR steel CT specimens, and then the fatigue crack growth rate of other CT specimens is forecasted accurately by the trained neural network.