Defect detection method for casting turbine blades in aeroengines based on unsupervised learning
-
摘要:
为实现航空发动机涡轮叶片射线检测自动化、智能化，有效改善传统射线检测费时费力、效率低下等问题，开展基于无监督学习的涡轮叶片X-ray图像缺陷检测方法研究。基于无监督生成对抗网络，提出一种适用于航空发动机涡轮叶片X-ray图像的缺陷检测算法；构建由生成网络、判别网络和附加自编码网络组成的深度卷积生成对抗网络，设计重构损失、判别损失、编码损失及中间编码损失，并利用4种损失的加权之和构造目标函数；利用完好涡轮叶片X-ray图像进行模型训练，基于训练得到的生成网络建立航空发动机涡轮叶片X-ray图像缺陷检测模型。研究了输入图像大小、编码长度和重构损失对缺陷检测模型性能的影响。结果表明：模型在输入图片像素尺寸为128像素×128像素、编码长度为600、重构损失为
L 2的情况下检测性能最佳，area under curve（AUC）可达到0.911。该缺陷检测算法能够实现实际生产缺陷零漏检的严苛技术指标，但误检率（>62.1%）较大，作为辅助检测手段应用于实际生产可将人工检测效率提高1.6倍。Abstract:To achieve the automation and intelligence of the radiographic inspection of turbine blades in aeroengines, and to effectively address the time-consuming, labor-intensive, and inefficient problems in traditional radiographic inspection methods, a research initiative was undertaken to develop a defect detection method for X-ray images of turbine blade based on unsupervised learning. A defect inspection algorithm suitable for X-ray images of aeroengine turbine blades was proposed based on an unsupervised generative adversarial network. It consisted of a generator network, a discriminator network, and an extra encoder network. Reconstruction, discrimination, encoding, and intermediate encoding loss were designed, and the weighted sum of the four losses was used to construct the objective function. Using non-defective X-ray images for model training. A defect inspection model for X-ray images of aeroengine turbine blades was established based on the trained generator network. The effects of input image size, encoding size, and type of reconstruction loss on the performance of the defect detection model were studied. Results showed that the proposed model with an input image size of 128 pixel×128 pixel, 600 encoding size, and
L 2 reconstruction loss can achieve an area under curve (AUC) of 0.911. The defect inspection algorithm can realize strict technical indicators of zero missing rate for actual production, but the false detection rate (>62.1%) was relatively high. As an auxiliary detection method applied in actual production, it can improve the manual detection efficiency by 1.6 times. -
图 4 基于DCGAN的涡轮叶片缺陷检测流程
Figure 4. Defect detection process for turbine blades based on DCGAN
图 5 叶片缺陷检测模型整体网络结构
Figure 5. Overall network structure of blade defect detection model
图 9 4种缺陷检测模型训练阶段的总损失变化曲线
Figure 9. Total loss variation curves during training phase of four defect detection models
图 10 采用不同重构损失训练模型的损失变化曲线
Figure 10. Loss variation curves of models trained with different reconstruction losses
图 11 输入尺寸为96像素×96像素的真实测试图像与重构图像示例图
Figure 11. Sample images of real test images and reconstructed images with dimensions of 96 pixels×96 pixels
表 1 服务器及环境配置主要参数
Table 1. Primary parameters of server and environmental configurations
参数名称 数值及说明 CPU Intel Xeon（R）Silver 4114@2.20 GHz GPU NVIDIA GeForce GTX 2080Ti 操作系统 Windows 10 Python 3.6.9 PyTorch 1.5.1 TorchVision 0.6.1 CUDA 10.1 cuDNN 7.6.5 
下载: 导出CSV
表 3 4种缺陷检测模型主要参数及检测精度
Table 3. Primary parameters and detection accuracy of four defect detection models
模型 输入尺寸 编码尺寸 每轮训练
时长/s重构
损失AUC A 64×64 600 29.314 L2 0.768 B 96×96 600 33.362 L2 0.852 C 128×128 600 79.016 L2 0.911 D 160×160 600 92.671 L2 0.880
下载: 导出CSV
表 4 不同重构损失的缺陷检测模型主要参数及精度
Table 4. Primary parameters and accuracy of defect detection models with different reconstruction losses
模型 输入尺寸 编码尺寸 每轮训练
时长/s重构
损失AUC H 128×128 600 85.575 L1 0.879 C 128×128 600 79.016 L2 0.911
下载: 导出CSV
表 5 不同模型在零漏检率情况下各自的误检率
Table 5. Respective false positive rates for different models at zero missing rates
% 参数 模型 A B C D E F G H 图像输入尺寸/像素 64×64 96×96 128×128 160×160 128×128 128×128 128×128 128×128 编码长度 600 600 600 600 400 800 1000 600 重构损失 L2 L2 L2 L2 L2 L2 L2 L1 漏检率/% 0 0 0 0 0 0 0 0 最小误检率/% 69.6 62.1 68.5 72.5 64.2 81.1 70.3 75.9
下载: 导出CSV
-
[1] BALLAL D R,ZELINA J. Progress in aeroengine technology (1939—2003)[J]. Journal of Aircraft,2004,41(1): 43-50. doi: 10.2514/1.562 [2] 李洪松,刘永葆,贺星,等. 考虑耦合损伤的燃气轮机叶片材料高低周复合疲劳寿命研究[J]. 推进技术,2022,43(2): 201005. LI Hongsong,LIU Yongbao,HE Xing,et al. Combined high and low cycle fatigue life of gas turbine blade materials considering coupling damage[J]. Journal of Propulsion Technology,2022,43(2): 201005. (in ChineseLI Hongsong, LIU Yongbao, HE Xing, et al. Combined high and low cycle fatigue life of gas turbine blade materials considering coupling damage[J]. Journal of Propulsion Technology, 2022, 43(2): 201005. (in Chinese) [3] 牛建平,尹冬梅,王琳,等. 熔模铸造某燃机涡轮叶片工艺研究[J]. 特种铸造及有色合金,2019,39(11): 1226-1229. NIU Jianping,YIN Dongmei,WANG Lin,et al. Investment casting technology for the turbine blade[J]. Special Casting & Nonferrous Alloys,2019,39(11): 1226-1229. (in ChineseNIU Jianping, YIN Dongmei, WANG Lin, et al. Investment casting technology for the turbine blade[J]. Special Casting & Nonferrous Alloys, 2019, 39(11): 1226-1229. (in Chinese) [4] 徐丽,刚铁,张明波,等. 铸件缺陷无损检测方法的研究现状[J]. 铸造,2002,51(9): 535-540. XU Li,GANG Tie,ZHANG Mingbo,et al. Review of non-destructive test methods for castings defect[J]. Foundry,2002,51(9): 535-540. (in ChineseXU Li, GANG Tie, ZHANG Mingbo, et al. Review of non-destructive test methods for castings defect[J]. Foundry, 2002, 51(9): 535-540. (in Chinese) [5] VIJAYA LAKSHMI M R,MONDAL A K,JADHAV C K,et al. Overview of NDT methods applied on an aero engine turbine rotor blade[J]. Insight-Non-Destructive Testing and Condition Monitoring,2013,55(9): 482-486. doi: 10.1784/insi.2012.55.9.482 [6] 俞梦倩,吴伟,邬冠华,等. 航空发动机涡轮叶片DR检测工艺参数优化[J]. 航空动力学报,2023,38(8): 1837-1845. YU Mengqian,WU Wei,WU Guanhua,et al. Optimization of DR detection process parameters for aero-engine turbine blades[J]. Journal of Aerospace Power,2023,38(8): 1837-1845. (in ChineseYU Mengqian, WU Wei, WU Guanhua, et al. Optimization of DR detection process parameters for aero-engine turbine blades[J]. Journal of Aerospace Power, 2023, 38(8): 1837-1845. (in Chinese) [7] 何超,陈果,王雨薇. 基于深度学习的航空发动机内部损伤实时检测方法[J]. 航空动力学报,2023,38(8): 1857-1864. HE Chao,CHEN Guo,WANG Yuwei. Real-time detection method of aero-engine internal damage based on deep learning[J]. Journal of Aerospace Power,2023,38(8): 1857-1864. (in ChineseHE Chao, CHEN Guo, WANG Yuwei. Real-time detection method of aero-engine internal damage based on deep learning[J]. Journal of Aerospace Power, 2023, 38(8): 1857-1864. (in Chinese) [8] SHANG H,WU J,SUN C,et al. Global prior transformer network in intelligent borescope inspection for surface damage detection of aeroengine blade[J]. IEEE Transactions on Industrial Informatics,2022,19(8): 8865-8877. [9] 张辉,张邹铨,陈煜嵘,等. 工业铸件缺陷无损检测技术的应用进展与展望[J]. 自动化学报,2022,48(4): 935-956. ZHANG Hui,ZHANG Zouquan,CHEN Yurong,et al. Application advance and prospect of nondestructive testing technology for industrial casting defects[J]. Acta Automatica Sinica,2022,48(4): 935-956. (in ChineseZHANG Hui, ZHANG Zouquan, CHEN Yurong, et al. Application advance and prospect of nondestructive testing technology for industrial casting defects[J]. Acta Automatica Sinica, 2022, 48(4): 935-956. (in Chinese) [10] CHEN Changxing,ABDULLAH A,KOK S H,et al. Review of industry workpiece classification and defect detection using deep learning[J]. International Journal of Advanced Computer Science and Applications,2022,13(4): 329-340. [11] DHRUVA KUMAR D,FANG Cheng,ZHENG Yue,et al. Semi-supervised transfer learning-based automatic weld defect detection and visual inspection[J]. Engineering Structures,2023,292: 116580. doi: 10.1016/j.engstruct.2023.116580 [12] ROHKOHL E,KRAKEN M,SCHÖNEMANN M,et al. How to characterize a NDT method for weld inspection in battery cell manufacturing using deep learning[J]. The International Journal of Advanced Manufacturing Technology,2022,119(7): 4829-4843. [13] 王靖然,王桂棠,杨波,等. 深度学习在焊缝缺陷检测的应用研究综述[J]. 机电工程技术,2021,50(3): 65-68. WANG Jingran,WANG Guitang,YANG Bo,et al. Summary of research on application of deep learning in weld defect detection[J]. Mechanical & Electrical Engineering Technology,2021,50(3): 65-68. (in ChineseWANG Jingran, WANG Guitang, YANG Bo, et al. Summary of research on application of deep learning in weld defect detection[J]. Mechanical & Electrical Engineering Technology, 2021, 50(3): 65-68. (in Chinese) [14] 王斌,李敏,雷承霖,等. 基于深度学习的织物疵点检测研究进展[J]. 纺织学报,2023,44(1): 219-227. WANG Bin,LI Min,LEI Chenglin,et al. Research progress in fabric defect detection based on deep learning[J]. Journal of Textile Research,2023,44(1): 219-227. (in ChineseWANG Bin, LI Min, LEI Chenglin, et al. Research progress in fabric defect detection based on deep learning[J]. Journal of Textile Research, 2023, 44(1): 219-227. (in Chinese) [15] KAHRAMAN Y,DURMUŞOĞLU A. Deep learning-based fabric defect detection: a review[J]. Textile Research Journal,2023,93(5/6): 1485-1503. [16] LIU Qiang,WANG Chuan,LI Yusheng,et al. A fabric defect detection method based on deep learning[J]. IEEE Access,2022,10: 4284-4296. doi: 10.1109/ACCESS.2021.3140118 [17] 张露,朱文俊,祝双武. 织物疵点自动检测方法及应用进展[J]. 纺织科技进展,2022(2): 21-26. ZHANG Lu,ZHU Wenjun,ZHU Shuangwu. Methods and application of automatic fabric defects detection[J]. Progress in Textile Science & Technology,2022(2): 21-26. (in ChineseZHANG Lu, ZHU Wenjun, ZHU Shuangwu. Methods and application of automatic fabric defects detection[J]. Progress in Textile Science & Technology, 2022(2): 21-26. (in Chinese) [18] 刘国才,顾冬冬,刘骁,等. 用于肿瘤调强放射治疗影像分析与转换的深度学习方法[J]. 中国生物医学工程学报,2022,41(2): 224-237. LIU Guocai,GU Dongdong,LIU Xiao,et al. Deep learning methods for image analysis and synthesis for intensity modulated radiotherapy: a review[J]. Chinese Journal of Biomedical Engineering,2022,41(2): 224-237. (in ChineseLIU Guocai, GU Dongdong, LIU Xiao, et al. Deep learning methods for image analysis and synthesis for intensity modulated radiotherapy: a review[J]. Chinese Journal of Biomedical Engineering, 2022, 41(2): 224-237. (in Chinese) [19] 张颖,仇大伟,刘静. 生成对抗网络在肝脏肿瘤图像分割中的应用综述[J]. 计算机工程与应用,2022,58(16): 18-30. ZHANG Ying,QIU Dawei,LIU Jing. Review on application of generative adversarial network in liver tumor image segmentation[J]. Computer Engineering and Applications,2022,58(16): 18-30. (in ChineseZHANG Ying, QIU Dawei, LIU Jing. Review on application of generative adversarial network in liver tumor image segmentation[J]. Computer Engineering and Applications, 2022, 58(16): 18-30. (in Chinese) [20] AKCAY S,BRECKON T. Towards automatic threat detection: a survey of advances of deep learning within X-ray security imaging[J]. Pattern Recognition,2022,122: 108245. doi: 10.1016/j.patcog.2021.108245 [21] VUKADINOVIC D,OSÉS M R,ANDERSON D. Automated detection of inorganic powders in X-ray images of airport luggage[J]. Journal of Transportation Security,2023,16(1): 3. doi: 10.1007/s12198-023-00261-5 [22] 冯洋博. 基于生成对抗网络的缺陷检测方法[D]. 天津: 天津理工大学,2020. FENG Yangbo. Defect detection method based on generative adversarial network[D]. Tianjin: Tianjin University of Technology,2020. (in ChineseFENG Yangbo. Defect detection method based on generative adversarial network[D]. Tianjin: Tianjin University of Technology, 2020. (in Chinese) [23] ZHOU Qihong,MEI Jun,ZHANG Qian,et al. Semi-supervised fabric defect detection based on image reconstruction and density estimation[J]. Textile Research Journal,2021,91(9/10): 962-972. [24] ZHANG Gaowei,PAN Yue,ZHANG Limao. Semi-supervised learning with GAN for automatic defect detection from images[J]. Automation in Construction,2021,128: 103764. doi: 10.1016/j.autcon.2021.103764 [25] 殷列栋. 基于重构残差的轮胎缺陷X光图像异常检测算法研究[D]. 杭州: 浙江大学,2021. YIN Liedong. Research on anomaly detection algorithm of tire defect based on X-ray image reconstruction residual[D]. Hangzhou: Zhejiang University,2021. (in ChineseYIN Liedong. Research on anomaly detection algorithm of tire defect based on X-ray image reconstruction residual[D]. Hangzhou: Zhejiang University, 2021. (in Chinese) [26] XU Rongge,HAO Ruiyang,HUANG Biqing. Efficient surface defect detection using self-supervised learning strategy and segmentation network[J]. Advanced Engineering Informatics,2022,52: 101566. doi: 10.1016/j.aei.2022.101566 [27] WANG Donghuan,XIAO Hong,WU Dingyi. Application of unsupervised adversarial learning in radiographic testing of aeroengine turbine blades[J]. NDT & E International,2023,134: 102766. [28] KINGMA D P,BA J. Adam: a method for stochastic optimization[EB/OL]. 2014: arXiv: 1412.6980. http://arxiv.org/abs/1412.6980. -
点击查看大图
计量
- 文章访问数: 782
- HTML浏览量: 687
- PDF量: 116
- 被引次数: 0