融合多传感器数据的发动机剩余寿命预测方法

针对基于单一传感器数据的剩余寿命预测方法存在数据利用率低和预测精度不高的问题，论文提出了一种融合多传感器数据的发动机剩余寿命预测方法。首先将多个传感器数据融合成一个复合健康指标来表征发动机的退化性能，采用线性维纳过程对复合健康指标进行退化建模，通过极大似然估计方法确定模型参数，进而得到发动机的预测寿命。为了确定融合系数，提出了一种利用真实寿命与预测寿命的预测均方误差最小化的方法。融合系数确定后，基于训练发动机历史寿命数据，确定出模型参数的离线估计值；然后利用Bayesian公式，同时结合发动机的实时监测数据与参数的先验分布对模型参数进行实时更新，接着在首达时间的意义下推导出剩余寿命的概率分布，进而实现了发动机的剩余寿命在线预测。最后，选择商用模块化航空推进系统仿真数据集进行数值仿真实验，结果表明：相较于基于单一传感器的方法，论文所提方法能够提高剩余寿命预测的准确性，其剩余寿命预测的相对均方误差降低了2%左右。

Remaining useful life prediction method for engine combining multi-sensors data

For remaining useful life prediction method based on single sensor data, there is a problem of low data utilization and low prediction accuracy. This paper proposes a method for predicting the remaining useful life of an engine that combines multi-sensors data. Firstly, multiple sensors data are fused into a composite health index to characterize the degraded performance of the engine. The linear Wiener process is used to model the composite health index. The model parameters are determined by the maximum likelihood estimation method, and then the predicted life of the engine is obtained. In order to determine the fusion coefficient, a method for minimizing the predicted mean square error between the actual life and the predicted life is proposed. After the fusion coefficient is determined, based on the historical life data of the training engine, the offline estimation value of the model parameters is determined. Then, using the Bayesian formula, combined with the real-time monitoring data of the engine and the prior distribution of the parameters, the model parameters are updated in real time, and the probability distribution of the remaining useful life is derived based on the first hitting time, thereby realizing the online prediction of the remaining useful life of the engine. Finally, the commercial modular aero-propulsion system simulation data set is selected for numerical simulation experiments. The results show that compared with the single sensor-based method, the proposed method can improve the accuracy of the remaining useful life prediction, and the relative mean square error of remaining useful life prediction is reduced by about 2%.