To address the issues of poor completeness and low accuracy in 3D point cloud reconstruction of damaged aero-engine blades, a 3D reconstruction method based on the UCD-PatchmatchNet model was proposed. First, the U-Net network was introduced to reconstruct the multi-scale feature extraction structure of the PatchmatchNet model, thereby enhancing its feature extraction capability. Second, the convolutional block attention module (CBAM) was employed to optimize the feature extraction module, enabling the model to focus on key regions and strengthening its feature extraction ability. Meanwhile, the randomly initialized depth map module was replaced with a Delaunay triangulation method based on depth gradient sampling to generate the initial depth map, improving the accuracy of the predicted depth map. Finally, validation was conducted on both public datasets and a self-established dataset of damaged aero-engine blades. The results show that, compared to the PatchmatchNet model, the proposed UCD-PatchmatchNet model reduces the accuracy error by 12.9%, the completeness error by 4.0%, and the overall error by 9.4%. The reconstructed point clouds of damaged blades exhibit good completeness and accuracy, better meeting the requirements for damage analysis of aero-engine blades.
CAI Shu-yu, TIAN Zhuang, HE Chong, FAN Jie.
Reconstruction Method for Damaged Blades of Aircraft Engines Based on UCD-PatchmatchNet Model[J].
Science Technology and Engineering, 2026, 26(27): 11938-11952 DOI:10.12404/j.issn.1671-1815.2507497
基于传统技术路线的航空发动机叶片类零件三维重建,可进一步划分为主动式重建与被动式重建两类。在主动重建方法中,激光扫描法[2]的重建精度更高,但该方法高度依赖专用精密设备,不仅设备结构复杂,购置成本也十分高昂;结构光法[3]、Kinect技术[4]、TOF(time of flight)技术[5]这类方法对设备的要求更低,但普遍存在重建精度不足的问题,无法适配大规模复杂场景的重建需求。在被动式重建方法中,目前主要采用搭载相机设备的多(双)目立体视觉重建系统完成重建,这类方法难以处理叶片的弱纹理区域,无法很好地满足当前航空发动机叶片重建对高精度、高效率、高自动化的要求。
针对这一问题,基于U-Net网络构建新的特征提取和融合结构,增强模型对全局上下文信息的感知能力。与基于FPN的多尺度特征提取结构自上而下的单向信息流动不同,基于U-Net网络的特征提取结构主要由编码器-解码器架构与跳跃连接组成,结构如图4所示。每个编码器块包含2个卷积块ConvBnReLU(convolution-batch normalization-rectified linear unit)和1个下采样层Subsampling(第1个编码器块只包含2个卷积块),特征图通过卷积改变通道数并保持图像尺寸不变,同时为了防止ConvBnReLU滑动时跳过像素导致丢失重要信息,分别设置卷积核Kc(kernel)大小为3×3、步长Sc(stride)为1、填充Pc(padding)为1,空洞卷积Dc(dilation)为1。假设F为输入特征图,F*为输出特征图,则特征图尺寸计算公式为
DongJinlin. Research on inspection and maintenance of aircraft engine blades[J]. Internal Combustion Engine & Parts, 2025(9): 73-75.
[3]
MurakamiT C, ManoT, SaikawaS, et al. A three-dimensional single-cell-resolution whole-brain atlas using CUBIC-X expansion microscopy and tissue clearing[J]. Nature Neuroscience, 2018, 21(4): 625-637.
[4]
ParkJ J. Fusion algorithms and high-performance applications for vehicular cloud computing[J]. The Journal of Supercomputing, 2018, 74(3): 995-1000.
[5]
LiaoJ X, YangD, LiT H, et al. Fusion feature for LSH-based image retrieval in a cloud datacenter[J]. Multimedia Tools and Applications, 2016, 75(23): 15405-15427.
[6]
HansardM, LeeS, ChoiO, et al. Time-of-flight cameras: principles, methods and applications[M]. Berlin: Springer Science & Business Media, 2012.
WangWenju, TangBang, GuZehua, et al. Overview of multi-view 3D reconstruction techniques in deep learning[J]. Computer Engineering and Applications, 2025, 61(6): 22-35.
[9]
WangP, LiuL, LiuY, et al. Neus: learning neural implicit surfaces by volume rendering for multi-view reconstruction[EB/OL]. arXiv Preprint, arXiv: 2106. 10689, 2021.
[10]
XiJ, ShiY, WangY, et al. Raymvsnet: Learning ray-based 1d implicit fields for accurate multi-view stereo[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2022: 8595-8605.
[11]
ChangD, BožičA, ZhangT, et al. RC-MVSNet: unsupervised multi-view stereo with neural rendering[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 665-680.
[12]
LongX, LinC, WangP, et al. Sparseneus: fast generalizable neural surface reconstruction from sparse views[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 210-227.
[13]
ChengK, LongX, YangK, et al. Gaussianpro: 3d gaussian splatting with progressive propagation[C]// Forty-first International Conference on Machine Learning. Vienna: JMLR, 2024: 5678-5690.
[14]
XieH Z, YaoH X, SunX S, et al. Pix2Vox:context aware 3D reconstruction from single and multi-view images[C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 2690-2698.
[15]
JiM, ZhangJ, DaiQ, et al. SurfaceNet+:an end-to-end 3d neural network for very sparse multi-view stereopsis[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 43(11): 4078-4093.
[16]
JiM Q, GallJ, ZhengH T, et al. SurfaceNet: an end-to-end 3D neural network for multiview stereopsis[C]// Proceeding of the 2017 IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 2326-2334.
[17]
WangN Y, ZhangY D, LiZ W, et al. Pixel2Mesh:generating 3D mesh models from single RGB images[C]// Proceedings of the 15th European Conference on Computer Vision. Cham: Springer International Publishing, 2018: 55-71.
[18]
WenC, ZhangY D, LiZ W, et al. Pixel2Mesh++:multi view 3D mesh generation via deformation[C]// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 1042-1051.
[19]
YaoY, LuoZ, LiS, et al. Mvsnet: depth inference for unstructured multi-view stereo[C]// Proceedings of the European Conference on Computer Vision (ECCV). Cham: Springer, 2018: 767-783.
[20]
ChengS, XuZ, ZhuS, et al. Deep stereo using adaptive thin volume representation with uncertainty awareness[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 2524-2534.
[21]
YangJ, MaoW, AlvarezJ M, et al. Cost volume pyramid based depth inference for multi-view stereo[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 4877-4886.
[22]
YuZ, GaoS. Fast-mvsnet: sparse-to-dense multi-view stereo with learned propagation and gauss-newton refinement[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2020: 1949-1958.
[23]
WeiZ, ZhuQ, MinC, et al. Aa-rmvsnet: adaptive aggregation recurrent multi-view stereo network[C]// Proceedings of the IEEE/CVF International conference on Computer Vision. New York: IEEE, 2021: 6187-6196.
[24]
CaiY, LiL, WangD, et al. MFNet: multi-level fusion aware feature pyramid based multi-view stereo network for 3D reconstruction[J]. Applied Intelligence, 2023, 53(4): 4289-4301.
YanJ, WeiZ, YiH, et al. Dense hybrid recurrent multi-view stereo net with dynamic consistency checking[C]// European Conference on Computer Vision. Cham: Springer International Publishing, 2020: 674-689.
[27]
WangF, GallianiS, VogelC, et al. Patchmatchnet: learned multi-view patchmatch stereo[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2021: 14194-14203.
[28]
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017(30): 770-778.
[29]
WangX F, ZhuZ, HuangG, et al. MVSTER: epipolar transformer for efficient multi-view stereo[C]// Proceeding of the 17th European Conference on Computer Vision. Tel Aviv: Springer, 2022: 573-591.
[30]
CaoC J, RenX L, FuY W. MVSFormer: multi-view stereo by learning robust image features and temperature-based depth[EB/OL]. arXiv Preprint, arXiv: 2208. 02541, 2022.
[31]
ZhouH, CaoY, ChuW, et al. Seedformer: Patch seeds based point cloud completion with upsample transformer[C]// European conference on computer vision. Cham: Springer Nature Switzerland, 2022: 416-432.
[32]
AnL, ZhouP, ZhouM, et al. PointTr: Low-overlappoint cloud registration with transformer[J]. IEEE Sensors Journal, 2024, 24(8): 12795-12805.
[33]
苏建宇. 基于结构光三维测量的叶片表面反光图像修复方法研究[D]. 沈阳: 东北大学, 2022.
[34]
SuJianyu. Research on leaf surface reflection image restoration method based on structured light 3D measurement[D]. Shenyang: Northeastern University, 2022.
TangYuqin. Point cloud denoising and surface reconstruction in turbine blade preparation based on deep learning[D]. Mianyang: Southwest University of Science and Technology, 2023.
[37]
GroueixT, FisherM, KimV G, et al. A papier-mâché approach to learning 3d surface generation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 216-224.
XieJiake. Blade 3D reconstruction based on coordinate measuring machine and computer vision[J]. Dongfang Turbine, 2025(2): 57-63.
[40]
XiaoY, WuX, YinY, et al. PoseNeRF: in-situ 3D reconstruction method based on joint optimization of pose and neural radiation field for smooth and weakly textured aeroengine blade[J]. Sensors (Basel, Switzerland), 2025, 25(19): 6145.
[41]
RonnebergerO, FischerP, BroxT. U-net: convolutional networks for biomedical image segmentation[C]// International Conference on Medical Image Computing and Computer-assisted Intervention. Cham: Springer International Publishing, 2015: 234-241.
[42]
WooS, ParkJ, LeeJ Y, et al. Cbam: convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision (ECCV). Cham: Spring, 2018: 3-19.
[43]
AanæsH, JensenR R, VogiatzisG, et al. Large-scale data for multiple-view stereopsis[J]. International Journal of Computer Vision, 2016, 120(2): 153-168.
PDF
(38069KB)
