基于神经网络响应面的疲劳裂纹扩展寿命的可靠性分析

RELIABILITY ANALYSIS FOR FATIGUE CRACK GROWTH LIFE BASED ON NEURAL NETWORK RESPONSE SURFACE

当失效形式的极限状态方程中随机变量个数较多或非线性较高时 ,其形式很复杂 ,因此传统的计算失效概率的方法不再适用。针对疲劳裂纹扩展寿命失效概率计算的复杂性 ,提出基于神经网络响应面的可靠性分析方法。首先建立神经网络响应面模拟疲劳裂纹扩展寿命的极限状态方程 ,然后使用遗传算法 (GA)计算可靠性指标。数值试验表明 ,本方法可以快速、精确地模拟疲劳裂纹扩展寿命的极限状态函数 ,进而计算出失效概率和可靠性指标。同其他模拟技术相比 ,在精度相同的情况下 。

On condition that the limit state function is highly complex due to many random variables or nonlinearity, the conventional methods for calculation of failure probability are inappropriate. In response to the complexity of calculation for failure probability regarding fatigue crack growth life, a method for reliability analysis based on neural network response surface was presented. First of all, a neural network response surface was established to simulate the limit state function of fatigue crack growth life, then the gene arithmetic(GA)was used to calculate the reliability index. Numeric experiments indicate that limit state function for fatigue crack growth life can be quickly, exactly simulated by using this method, then the failure probability and the reliability index will be easily worked out. Compared with other simulation technique, neural network response surface method can greatly reduce the simulation time at the same precision level.