基于Copula熵传感器选择的发动机相似性 寿命预测方法

Engine Similarity Life Prediction Based on Copula Entropy for Sensor Selection

中文关键词:  预测与健康管理  寿命预测  健康因子  航空发动机  Copula熵  非线性  相似性

英文关键词:prognostics and health management  life prediction  health factors  aeroengine  Copula entropy  nonlinear  similarity

中文摘要:

      针对常见的特征选取方法难以度量传感器与性能退化之间的非线性关系，不能精准地筛选传感器构建健康因子，导致 寿命预测误差偏高的问题，提出一种基于Copula熵传感器选择的发动机相似性寿命预测方法。基于K-Means聚类得到消除工况 影响的传感器退化特征；利用Copula熵非线性度量指标筛选出与初始健康因子密切相关的（最优）传感器，重新构建健康因子；建 立每组失效发动机的指数退化模型，借助这些模型预测服役发动机当前运行周期的健康因子，基于健康因子的真实值与预测值定 义发动机之间的相似距离，在失效发动机库中搜索服役发动机的相似设备并预测剩余寿命；基于C-MAPSS数据集验证了该方法 的可行性与有效性。结果表明：在50%、70%、90%运行周期下，基于Copula熵传感器选择的发动机相似性寿命预测方法的预测误 差分别降低了39.25%、41.69%、50.53%，有效提高了寿命预测的精度，具有较大的工程应用价值。

英文摘要:

      Aiming at the problem that the common feature selection methods are difficult to measure the nonlinear relationship between sensors and performance degradation, and cannot accurately screen sensors to construct health factors, resulting in high life predic? tion errors, a similarity engine life prediction method based on Copula entropy for sensor selection was proposed. Firstly, sensor degradation characteristics with the influence of operating conditions eliminated were obtained based on K-Means clustering. Secondly, the nonlinear index-Copula entropy was used to select the optimal sensor closely related to the original health factor and reconstruct the health factor. Then, exponential degradation models were established for each group of failed engines, and the health factors of the current operating cycle of the in-service engines were predicted according to these models. The similarity distance between the two engines was defined based on the actual and predicted value of the health factors. Search the equipment similar to the in-service engine in the database of failed engines and predict the remaining useful life of the engine . Finally, the feasibility and validity of the method were verified based on the C- MAPSS dataset. The results show that the prediction errors of the proposed method are reduced by 39.25%, 41.69%, and 50.53% respectively at 50%, 70% and 90% of the operating cycle, effectively improving the accuracy of life prediction.