基于超统计的多阶段航空发动机剩余寿命预测

Remaining useful life prediction of multi-stage aero-engine based on super statistics

针对传统航空发动机剩余寿命(RUL)预测模型无法客观描述多阶段性能衰退过程及对于RUL预测精度不高的问题,提出了一个新的多阶段航空发动机RUL预测模型,包括超统计理论、突变点检测、无迹卡尔曼滤波(UKF)与非线性预测4部分内容。提出了基于超统计理论的多阶段分割滤波(BS-MSF)算法。首先,该算法采用超统计理论进行突变点检测,将航空发动机的健康数据划分为多个退化阶段;其次,应用UKF对融合的时变参数进行滤波处理;最后,通过非线性拟合对发动机RUL进行预测,实验采用美国NASA发布的航空发动机数据进行数据分析和验证。结果表明:所提算法在发动机性能退化中的预测具有更好的适应性和更小的拟合误差,能更准确地预测发动机的RUL,预测精度比单阶段方法提高5.5%。

Traditional aero-engine Remaining Useful Life(RUL) model cannot objectively describe the multi-stage degeneration process, and the accuracy of RUL prediction is low. To solve this problem, a new multi-stage RUL prediction model for RUL prediction is proposed, including super statistics theory, mutation point detection, Unscented Kalman Filter(UKF) and nonlinear prediction. In the paper, a Multi-stage Segmentation Filtering based on Super statistics(BS-MSF) theory algorithm is proposed. In this algorithm, first, super statistics theory is used to conduct mutation point detection and divide the health data of aero-engine into multiple degradation phases. Then, UKF is used to filter the fused time-varying parameters. Finally, the real RUL of the aero-engine is estimated by nonlinear fitting. nonlinear fitting, and the aero-engine data was released by National Aeronautics and Space Administration. Simulation results show that the presented method has better adaptability in predicting engine performance degradation, smaller fitting error, and more accurate prediction of RUL. The prediction accuracy is 5.5% higher than that of single-stage method.