近门槛值区疲劳裂纹扩展行为的神经网络预测方法

Neural network based prediction of fatigue crack growth behavior in near-threshold regime

近门槛值区的疲劳裂纹扩展行为对结构损伤容限设计至关重要,然而数据获取耗时费力,发展简捷精准的预测方法一直是人们的追求。以CrNiMoV钢为对象,在掌握疲劳裂纹扩展参数之间关系基础上构建神经网络模型,通过比较神经网络预测与Zhu-Xuan理论模型结果,分析神经网络方法的预测精度和能力。研究表明,神经网络方法可以快速有效地获取近门槛值区疲劳裂纹扩展行为,低应力比R时Zhu-Xuan模型预测精度优于神经网络,高应力比R时神经网络方法预测效果较好;预测未知R下近门槛值区裂纹扩展行为时,神经网络方法误差较大。

Fatigue crack propagation behavior in the near-threshold regime is crucial for damage tolerant design of engineering structures.However,data acquisition is quite time-consuming and cost ineffective,development of simple,quick and accurate prediction methods has long been the pursuit of fatigue research community.Using CrNiMoV steel as the object,a neural network model was established based on mastering the relationship between fatigue crack propagation parameters of the steel.The prediction accuracy and ability of the neural network method were analyzed by comparing the results of the neural network prediction and the Zhu-Xuan physical model.The results show that the neural network method can quickly and effectively obtain the fatigue crack propagation behavior in the near-threshold regime.The prediction accuracy of the Zhu-Xuan model is better than that of the neural network at low stress ratio R,and the performance of the neural network method is better at high stress ratio R.The neural network method has a large error when predicting the crack propagation behavior approaching fatigue threshold under unknown stress ratios.