基于迁移深度降噪自动编码器的飞机关键机械部件故障诊断方法
No.91860124 No.XQ201901 1 1 1 1 1. 710072 İ塣 1 2 乤 佡 1 İ 塣 artificial neural network ANN Support vector machine SVM 2 3-5 6 豸 棬 instance-based transfer learning feature-representation transfer learning parameter transfer learning relational-knowledge transfer learning 7 1 2 Deep transfer denoising autoencoder, DTDAE 1 Denoising autoencoder, DAE 8 1 layer / Hidden/Coding layer / Output/Decoding layer DAE 1 Fig. 1 Architecture of autoencoder X X1, X2, , Xi, , Xn n Xi x1, x2, , xj, , xm m 2 0, N X X I 1 h 11e fh W X b 2 fe sigmoid W1 b1 X 22 dfX W h b 3 fd sigmoid W2 b2 Back propagation algorithm 9 1 D A E 2 2 1 1 1 2 1 1 2 2 ll n l ij li ii i nn j J n W XX - 4 L2 2 1 2 Deep transfer denoising autoencoder, DTDAE Deep denoising autoencoder, DDAE 2 DDAE 3 Softmax 1 D D A E 1 -1 1 1 1 1 log 2 ll n kk k L l ij l i j nn J y yn GW 5 L Encoder Decoder 2 Fig. 2 Architecture of deep denoising autoencoder. DDAE 裬 DDAE DDAE-A DDAE-A DDAE-B llUW llab,l 1, 2, , L-1 6 Ul al Wl bl DDAE-B DDAE-A l 1 D D A E -B 1 -1 1 1 1 1 log 2 ll n kk k L l ij li n j n J y yn GU 7 3 3.1 10 3 12kHz 0hp, 1797r/min 300 400 1 11 4 10kHz 0hp, 1600r/min 300 400 2 3 Fig. 3 Test rig of source domain data. 1 Table. 1 Description of source domain data 1797r/min 0hp 300 1 300 2 300 3 300 4 4 Fig. 4 Test rig of target domain data. 2 Table. 2 Description of target domain data 1600r/min 0hp 300 1 300 2 300 3 300 4 3.2 DTDAE DDAE-A DDAE-B DDAE-A DDAE-A WL bL WL-1 bL-1 W1 b1 W2 b2 DDAE-B DDAE-A 400-200-100-50-4 0.0001 0.01 20 DDAE-B 400-200-100-50-4 0.0001 30 DDAE SVM ANN DDAE 400-200-100-50-4 0.0001 0.01 20 SVM 40 8 ANN 400-800-4 0. 1 0.95, 500 3 5 10 3 5 4 3 Table. 3 Results comparison of different DDAE SVM ANN 15 5 82.43 79.16 69.74 63.86 30 10 83.38 81.16 71.11 65.37 60 20 85.53 83.74 73.65 69.69 100 33 87.96 85.88 76.50 73.25 150 50 90.73 87.02 79.83 75.67 4 6仯 4 6 0 5 10 20 33 50 1 60 70 80 90 100 D D A E B P N N S V M 5 Fig. 5 Diagnosis results comparison of different . 4 Table. 4 Standard deviation comparison of different DDAE SVM ANN 15 5 0.91 3.03 4.32 8.54 30 10 1.13 4.14 5.05 8.15 60 20 1.25 5.72 4.53 6.96 100 33 1.09 4.18 3.79 6.58 150 50 1.41 2.19 3.64 5.62 0 5 10 20 33 50 1 0 2 4 6 8 10 6 Fig. 6 Standard deviation comparison of different 4 0 510203350 10 2 4 6 8 10 DDAEBPNN SVM 1 2 1 ; ; . J . . 2019, 55727-34. Jiang Hongkai; Shao Haidong; Li Xingqiu. Deep Learning Theory with Application in Intelligent Fault Diagnosis of Aircraft J . Journal of Mechanical Engineering, 2019, 55 727-34. 2 S. Ma, F.L. Chu, Ensemble deep learning-based fault diagnosis of rotor bearing systems J . Comput. Ind. 2019, 105143C152. 3 F. Jia, Y.G. Lei, J. Lin, X. Zhou, and N. Lu, Deep neural networks A promising tool for fault characteristics mining and intelligent diagnosis of rotating machinery with massive data J . Mechan. Syst. Signal Process. 2016, 72-73303C315. 4 H.D. Shao, H.K. Jiang, H.W. Zhao, et al., A novel deep auto encoder feature learning for rotating machinery fault diagnosis J . Mechan. Syst. Signal Process. 2017, 95187-204. 5 R. Zhao, R.Q. Yan, Z.H. Chen, et al., Deep learning and its applications to machine health monitoring J . Mechan. Syst. Signal Process. 2019, 115 213-237. 6 L. Guo, Y. Lei, S. Xing, T. Yan, and N. Li. Deep convolutional transfer learning network A new for intelligent fault diagnosis of machines with unlabeled data J . IEEE Trans. Ind. Electron., 2018, 6697316C7325. 7 S. J. Pan and Q. Yang. A survey on transfer learning J . IEEE Trans. Knowl. Data Eng., 2010, 2210 1345C1359. 8 P. Grgel, A. Simsek, Face recognition via Deep Stacked Denoising Sparse Autoencoders DSDSA J . Appl. Math. Comput., 2019, 355 325-342. 9 Z. Meng, X.Y. Zhan, J. Li, Z.Z. Pan, An enhancement denoising autoencoder for rolling bearing fault diagnosis J . Measurement. 2018, 130 448C454. 10 Case Western Reserve University Bearing Data Center. Online. Available httpcsegroups.case.edu/bearingdatacenter/hom e. 11 H.D. Shao, H.K. Jiang, F.A. Wang, H.W. Zhao. An enhancement deep feature fusion for rotating machinery fault diagnosis J . Knowl.-Based Syst. 2017, 119 200C220. Deep Transfer Denoising Autoencoder for Aircraft Key Mechanical Component Fault Diagnosis LI Xing-qiu1, JIANG Hong-kai1, WANG Rui-xin1 WU Zheng-hong 1 1. School of Aeronautics, Northwestern Polytechnical University, Xian 710072, China Abstract Fault diagnosis for key mechanical components of aircraft is of great significance to improve its safety and reliability. In engineering practice, there still exist the following problems. Firstly, it is difficult to obtain labeled fault samples. Secondly, existing models usually require training and testing data have same distribution. Therefore, this paper proposes a deep transfer denoising autoencoder for aircraft key mechanical component fault diagnosis with few labeled samples. The includes two parts 1 a deep denoising autoencoder is established to adaptively extract effective fault features from raw vibration signals; 2 parameter transfer learning is used to construct a deep transfer denoising autoencoder to tackle fault diagnosis problems with few labeled samples. Finally, experiment data are used to verify the effectiveness of the proposed , and the results show that the proposed can effectively complete the fault diagnosis task for key mechanical components of aircraft with few labeled data. Key words deep transfer denoising autoencoder; few labeled data; key mechanical components of aircraft; fault diagnosis 1993, , 绰 02988493671; E-mail lixingqiu . 1972, ,绰 02988493671; E-mail jianghk . 1995, ,绰 02988493671; E-mail wangruixin . 1996, ,绰 02988493671; E-mail 18270353479 .