Commercial aircraft engine post-repairing performance prediction based on fusion of multisource data
-
摘要:
针对民航发动机修后排气温度裕度预测过程中的多源异构数据融合问题,提出了卷积自编码器与极端梯度提升模型结合的方法。利用所提出的条件熵增长因子规整发动机修前多元传感器参数序列中的参数排序,采用卷积自编码器提取规整后的参数序列和维修工作范围的数据特征,并将其与发动机使用时间信息组成合成特征以训练极端梯度提升模型,从而预测发动机修后性能并评估各影响因素的重要程度。经发动机机队维修案例验证,所提方法预测精度高于单维参数序列预测方法,对发动机修后排气温度的平均相对预测误差不高于8.3%。
Abstract:To solve the problem of multisource heterogeneous data fusion in commercial aircraft engine post-repairing exhaust gas temperature margin prediction, a combined method of convolutional auto-encoder and extreme gradient boost model was proposed. This method uses the proposed cross entropy increasing factor to regularize the parameter order in the multi-dimensional engine sensor parameter series observed before repairing, and then uses convolutional auto-encoder to extract features from the regularized parameter series and engine workscope data. With the combined feature composed of the extracted features and the features representing engine using time, extreme gradient boost model is trained in order to predict engine post-repairing performance and estimate the importance of influential factors. The experiment performed on the prediction of the post-repairing performance of an engine fleet proved that the proposed method achieves higher prediction precision than prediction methods supported by one-dimentional parameter series and can predict engine post-repairing exhaust gas temperature margin with an average relative error no higher than 8.3%.
-
表 1 SAE-1D-XGBoost、CAE-1D-XGBoost、CAE-XGBoost对参数序列的重构误差
Table 1. Reconstruction error of parameter series by SAE-1D-XGBoost, CAE-1D-XGBoost and CAE-XGBoost
方法 特征值个数 MSE均值/10-3 ARE均值/10-2 MSE方差/10-6 ARE方差/10-6 SAE-1D-XGBoost 5 5.288 5.56 12.59 277.9 SAE-1D-XGBoost 10 4.757 5.29 13.73 347.0 SAE-1D-XGBoost 15 5.333 5.84 10.28 284.5 SAE-1D-XGBoost 20 4.573 5.28 10.15 285.0 CAE-1D-XGBoost 10 4.554 5.44 3.075 136.6 CAE-1D-XGBoost 20 9.173 8.20 12.30 327.4 CAE-XGBoost 25 12.60 7.39 0.1259 5.194 表 2 SAE-1D-XGBoost、CAE-1D-XGBoost、CAE-XGBoost对发动机的修后TEGTM预测误差
Table 2. Prediction error of engine post-repairing TEGTM by SAE-1D-XGBoost, CAE-1D-XGBoost and CAE-XGBoost
方法 特征值个数 MSE均值/10-2 ARE均值 SAE-1D-XGBoost 5 1.919 0.1145 SAE-1D-XGBoost 10 1.893 0.1083 SAE-1D-XGBoost 15 2.091 0.1198 SAE-1D-XGBoost 20 1.903 0.1135 CAE-1D-XGBoost 10 1.268 0.0860 CAE-1D-XGBoost 20 1.547 0.1053 CAE-XGBoost 25(均值) 1.183 0.0889 CAE-XGBoost 25(规整后) 0.8367 0.0822 表 3 不同方法中影响因素对发动机修后TEGTM的重要性占比
Table 3. Percentage of influential element importance to engine post-repairing TEGTM in different methods
方法 使用时间占比/% 送修性能占比/% 维修工作范围占比/% SAE-1D-XGBoost 47.54 12.04 40.42 CAE-1D-XGBoost 34.33 46.43 19.24 CAE-XGBoost 39.35 32.11 28.54 -
[1] MAKIS V, JARDINE A K S.Optimal replacement in the proportional hazards model[J]. INFOR:Information Systems and Operational Research, 1992, 30(1):172-183. doi: 10.1080/03155986.1992.11732183 [2] TIAN Z, LIN D, WU B.Condition based maintenance optimization considering multiple objectives[J]. Journal of Intelligent Manufacturing, 2012, 23(2):333-340. doi: 10.1007/s10845-009-0358-7 [3] GHASEMI A, YACOUT S, OUALI M S.Parameter estimation methods for condition-based maintenance with indirect observations[J]. IEEE Transactions on Reliability, 2010, 59(2):426-439. doi: 10.1109/TR.2010.2048736 [4] KAISER K A, GEBRAEEL N Z.Predictive maintenance management using sensor-based degradation models[J]. IEEE Transactions on Systems, Man, and Cybernetics-Part A:Systems and Humans, 2008, 39(4):840-849. doi: 10.1109-TSMCA.2009.2016429/ [5] KRAFT J, SETHI V, SINGH R.Optimization of aero gas turbine maintenance using advanced simulation and diagnostic methods[J]. Journal of Engineering for Gas Turbines and Power, 2014, 136(11):111601. doi: 10.1115/1.4027356 [6] HANACHI H, LIU J, BANERJEE A, et al.A physics-based modeling approach for performance monitoring in gas turbine engines[J]. IEEE Transactions on Reliability, 2015, 64(1):197-205. doi: 10.1109/TR.2014.2368872 [7] STALDER J P.Gas turbine compressor washing state of the art-Field experiences[C]//ASME 1998 International Gas Turbine and Aeroengine Congress and Exhibition.New York: ASME, 1998: V004T11A010. [8] BREMER C.Automated repair and overhaul of aero-engine and industrial gas turbine components[C]//ASME Turbo Expo 2005.New York: ASME, 2005: 841-846. [9] LU F, JIANG C, HUANG J, et al.A novel data hierarchical fusion method for gas turbine engine performance fault diagnosis[J]. Energies, 2016, 9(10):828. doi: 10.3390/en9100828 [10] ZHANG H J, ZUO H F, LIANG J.Maintenance level decision of aero-engine based on VPRS theory[J]. Transactions of Nanjing University of Aeronautics and Astronautics, 2005, 22(4):281-284. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=njhkhtdxxb-e200504003 [11] 张莹松, 吴灿, 唐海龙, 等.基于故障检测的某涡扇发动机维修决策方法[J].航空动力学报, 2017, 32(1):82-88. http://d.old.wanfangdata.com.cn/Periodical/hkdlxb201701012ZHANG Y S, WU C, TANG H L, et al.Maintenance decision method of a turbofan engine based on fault detection[J]. Journal of Propulsion and Power, 2017, 32(1):82-88(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/hkdlxb201701012 [12] CHEN T Q, GUESTRIN C.XGBoost: A scalable tree boosting system[C]//ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York: ACM, 2016: 785-794. [13] FRIEDMAN J H.Greedy function approximation:A gradient boosting machine[J]. Annals of Statistics, 2001, 29(5):1189-1232. [14] LI H, LIN Z, SHEN X, et al.A convolutional neural network cascade for face detection[C]//IEEE Conference on Computer Vision and Pattern Recognition.Piscataway, NJ: IEEE Press, 2015: 5325-5334. [15] 练绪宝, 林鸿飞, 徐博, 等.融合项目标签信息面向排序的社会化推荐算法[J].计算机科学与探索, 2017, 11(3):373-381. http://d.old.wanfangdata.com.cn/Periodical/jsjkxyts201703005LIAN X B, LIN H F, XU B, et al.Rank-oriented social recommendation algorithm with item tag information[J]. Journal of Frontiers of Computer Science and Technology, 2017, 11(3):373-381(in Chinese). http://d.old.wanfangdata.com.cn/Periodical/jsjkxyts201703005