基于小样本量的航空发动机材料高低周复合疲劳P-S-N曲线优化方法

Optimization method of combined high and low cycle fatigue P-S-N curve for aeroengine materials with small size sample

航空发动机材料的高低周复合疲劳P-S-N曲线是评估转子部件寿命的重要依据。然而,P-S-N曲线的确定对实验时间和材料成本均有较高的要求。基于物理信息机器学习方法,提出了基于小样本量的高低周复合疲劳P-S-N曲线的优化处理方法。该方法将等效寿命原理及寿命分布一致性判据通过损失函数引入极限学习机(ELM)模型,并考虑了上层输入变量和下层ELM模型参数的双层优化策略。随后通过应用疲劳实验数据,对比分析该方法与数据驱动机器学习方法以及传统P-S-N曲线拟合方法进行。结果表明:该方法不仅有效地解决了拟合过程中应力水平与高低周复合疲劳寿命标准差不呈线性关系的问题,且拟合得到的概率疲劳寿命分布与母体真值更为接近,具有较高的准确性。

The P-S-N curve of high and low cycle fatigue for aero-engine materials are essential to evaluate the service life of rotor components.However,it requires extensive experimental time and high material cost.Based on the physical-informed machine learning(PIML)method,a novel optimization method was proposed for combined high and low-cycle fatigue P-S-N curves with a small size sample,in which the equivalent principle of fatigue life and the consistency criteria of life distributions were introduced into the extreme learning machine(ELM)through its loss function.In addition,bi-level optimization was employed with the upper level of model input variables and the lower level of the ELM model parameters.Subsequently,the proposed PIML method was compared with a data-driven machine learning method and traditional P-S-N curve fitting methods through the fatigue test data.The results show that the method not only effectively solves the problem of nonlinearity between the stress level and the standard deviation of fatigue life,but also presents the highest accuracy of the predicted probabilistic lives.