金属疲劳裂纹扩展速率的贝叶斯正则化BP神经网络预测

人工神经网络是进行预报裂纹扩展率的一个重要方法。文章针对不同金属的疲劳裂纹扩展速率分别建立贝叶斯正则化BP(Back Propagation)神经网络,将各材料在不同应力比R下的疲劳裂纹扩展速率试验数据分为两部分,一部分用来进行训练网络,另一部分用来测试训练好的网络,检验其泛化能力。将从文献中获取的4种不同金属材料的疲劳试验数据作为算例,来检验网络的性能。计算结果表明贝叶斯正则化BP神经网络不仅对训练样本有很好的拟合能力,而且对于未训练过的测试样本也有较好的预测能力,即有较强的泛化能力。同时,指出了建立网络时减少门槛值附近的试验样本点,可以提高网络的预测能力。研究结果表明,该方法可以方便地获得不同应力比R下的疲劳裂纹扩展速率,从而达到减少试验次数,充分利用已有数据的目的。并且可以进一步应用于其他金属的疲劳裂纹扩展速率的预报。

Prediction of fatigue crack growth rate of metal based on Bayesian regularized BP neural network

Artificial neural network is an important method for predicting the fatigue crack growth rate. In this paper, the Bayesian regularized BP neural network is established to predict the fatigue crack growth rate of metal.The experimental data of each material at different stress ratio R are divided into two parts. One is used for training neural network, the other is used for testing the network. Experimental data of four different types of materials taken from literature were used in the analyses. The results show that the neural network has strong fitting and generalization capability. And the generalization capability of neural network is improved by reducing the training data near the threshold.So the neural network can be used for predicting the crack growth rate of different stress ratios R based on the existing data. Furthermore, it will provide a reliable and useful predictor for fatigue crack growth rate of different metals.