基于双尺度相似性的发动机寿命预测

Engine Life Prediction Based on Two-Scale Similarity

针对基于欧氏距离的移动相似性对不同时间序列相似度的刻画准确度不高和预测精度不高的问题,提出一种基于曼哈顿距离和斜率双尺度的改进相似性寿命预测方法。通过传感器数据变化幅度、各项参数出厂时的数值差异和下发时的数值差异这三个标准筛选传感器数据,并基于筛选结果构建一维健康指数;基于曼哈顿距离和斜率结合固定匹配和移动匹配进行初步寿命预测;利用移动匹配的特性建立筛选标准,并筛选获得最终预测结果;采用NASA的数据集对该方法进行验证。结果表明:该方法的预测精度相比当前的相似性预测方法提升30.35%,证明该改进方法对不同时间序列之间的相似度有更加精确的刻画。

Aiming at the problem of low accuracy on portraying similarity and life prediction for moving similarity by using Euclidean distance for different time series similarities,an improved similarity life prediction method based on Manhattan distance and slope was proposed.Firstly,the sensor parameters were screened by three criterion which are the sensor parameters variation range,the value difference of each sensor parameter at engine leaving the factory,and value difference of each sensor parameter at engine removal,and based on a comprehensive analysis on the selected sensor parameters,a one-dimensional health index was set up.Then Manhattan distance and slope were used to combine with the fixed matching and moving matching,and a preliminary life prediction was performed.The moving matching was used to set up the filtering rules and the final results were achieved.Finally,the method was verified by NASA data set,and the results show that the prediction accuracy of this method is 30.35% higher than the current ones,which proves that the proposed method is more accurate on portraying the similarity among different time series.