The change of residual stress is complex in the fatigue cycle. The method of evaluating the fatigue state through physical models couldn’t express the possibility of residual stress evolution. The data-driven model doesn’t conform to the physical evolution law in detail and the most model need large-scale data. In this paper, a probabilistic model based on physical model and Gaussian Process Flow is proposed to describe the fatigue life degradation curve of alloy components in practical production. It could get different evolution paths instead of statistical parameters by Gaussian Process Flow. Different evolution paths are calibrated by Kalman filter, and the model is optimized combined with the physical model. Through a fatigue test for nickel alloy, the proposed model could better describe the evolution relationship between fatigue stress and fatigue cycle under small samples.

疲劳循环中残余应力的变化是复杂的。通过物理模型评估疲劳状态的方法无法表达残余应力演化的可能性。数据驱动的模型在细节上不符合物理演化规律，大多数模型都需要大规模的数据。本文提出了一种基于物理模型和高斯过程流的概率模型来描述合金部件在实际生产中的疲劳寿命退化曲线。通过高斯过程流可以得到不同的演化路径而不是统计参数。通过卡尔曼滤波器对不同的演化路径进行标定，并结合物理模型对模型进行优化。通过镍合金的疲劳试验，