基于非线性数据融合的设备多阶段寿命预测

为解决设备监测数据具有维数高、非线性且退化过程中存在多阶段的问题，提出了一种基于非线性数据融合和多阶段退化的设备寿命预测模型.首先，利用神经网络理论中的自编码器对表征设备退化的多维参数进行了融合，构建出设备的退化指示量；然后，利用CUSUM算法提取出设备退化过程中的分段点；最后，构建了多阶段维纳退化模型，从而实现对设备寿命的预测.利用航空发动机状态监测数据对所提模型进行了验证，剩余寿命预测的平均误差为0.254 5，低于传统的基于线性数据融合方法和基于单阶段维纳过程退化模型的寿命预测方法.结果证明，基于非线性数据融合的多阶段退化模型具有很好的鲁棒性，对设备的寿命预测更加精准.

Multistage Lifetime Estimation for Equipment Based on Nonlinear Data Fusion

To solve the problem that the measure data is high in dimensionality, nonlinear and multistage in the process of degradation. A lifetime estimation model based on nonlinear data fusion and multistage degradation is proposed. Firstly, the degradation indicator of equipment was constructed by fuse data fused by AutoEncoder technology based on neural network theory. Then the breakpoint of the multi-stage degradation process of the equipment was extracted by CUSUM breakpoint analysis method. At last, the multi-stage Wiener degradation model is constructed to predict the lifetime. The proposed method is validated by aero-engine condition monitoring data. The results show that compared with the PCA method and the single-stage Wiener degradation model, the average error of remaining useful lifetime prediction is 0.254 5, lower than the traditional methods. The verification results show that the multi-stage lifetime estimation model based on nonlinear data fusion has good robustness and more accurate.