稀疏数据驱动的涡轮叶片表面裂纹长度提取方法  

A sparse data-driven method for extracting surface crack length of turbine blade

作　　者：李茂月[1,2] 雷金超 张成龙[1,2] 刘泽隆 Li Maoyue;Lei Jinchao;Zhang Chenglong;Liu Zelong(Key Laboratory of Advanced Manufacturing and Intelligent Technology,Ministry of Education,Harbin University of Science and Technology,Harbin 150080,China;School of Mechanical and Power Engineering,Harbin University of Science and Technology,Harbin 150080,China)

机构地区：[1]哈尔滨理工大学先进制造智能化技术教育部重点实验室,哈尔滨150080 [2]哈尔滨理工大学机械动力工程学院,哈尔滨150080

出　　处：《仪器仪表学报》2025年第1期157-169,共13页Chinese Journal of Scientific Instrument

基　　金：黑龙江省自然科学基金(LH2022E085);国家自然科学基金(51975169)项目资助。

摘　　要：航发涡轮叶片裂纹的长度测量是裂纹危险等级评判的基础和修复的前提。针对涡轮叶片表面裂纹形状不规则、目标小、数据集样本稀少和裂纹成像角度失真等问题,提出一种稀疏数据驱动的涡轮叶片表面裂纹长度提取方法。首先,为提升Unet模型在处理稀疏数据时的精度,采用GeLu函数与Vgg16网络结合的方法提取裂纹特征,将输出作为Unet网络解码部分的输入,保证模型匹配的前提下,在随机初始化权重中引入预训练权重,并在跳跃连接层中引入高效金字塔压缩注意力模块,增强模型在复杂背景下对裂纹特征的聚焦能力。然后,为了得到裂纹的单位像素特征曲线,在精分割后提出使用八邻域骨架化保留裂纹的主干特征结构。最后,深入分析了相机成像原理,讨论了叶片弦线角和相机自身参数对裂纹长度的测量影响,采用张正友标定法求解相机内部参数,建立了像素尺寸与实际尺寸转换模型。实验结果表明,与X光测量相比,该方法在测量距离为100~300 mm时,得到的裂纹长度最大误差为6.8%,证明该方法在测量涡轮叶片表面裂纹长度中对X光检测技术具有可替代性;与原算法相比,改进的算法在针对稀疏数据检测时精度显著提高,平均交并比提升了7.14%。所提出的涡轮叶片裂纹长度提取方法,为叶片质量评估及后续修复提供了理论基础和数据支持。The measurement of crack length is the basis of crack risk grade evaluation and the precondition of crack repair.Aiming at the problems of irregular shape,small target,sparse data set and distortion of crack imaging angle,a sparse data driven method was proposed to extract the surface crack length of turbine blade.Firstly,in order to improve the precision of Unet model in sparse data processing,the method of combining GeLu function with Vgg16 network is used to extract crack features,and the output is used as the input of Unet network decoding,under the premise of ensuring the model matching,the pre-training weight is introduced into the random initialization weight,and the efficient pyramid compression attention module is introduced into the jump connection layer,it enhances the ability of the model to focus on the crack characteristics under complex background.Then,in order to get the unit pixel characteristic curve of the crack,after the fine segmentation,a skeleton structure with eight neighborhood is proposed to preserve the crack backbone characteristic structure.Finally,the camera imaging principle is analyzed in depth,and the influence of blade chord angle and camera parameters on the measurement of crack length is discussed,the conversion model between pixel size and actual size is established.The experimental results show that the maximum error of crack length is 6.8%when the measuring distance is 100~300 mm,compared with x-ray measurement,it is proved that the method is substitutable for X-ray detection technique in measuring the surface crack length of turbine blade,and the improved algorithm is more accurate than the original algorithm in detecting sparse data,the average cross-over ratio increased by 7.14%.The proposed method provides theoretical basis and data support for blade quality evaluation and subsequent repair.The measurement of crack length is fundamental to the evaluation of crack risk and a prerequisite for crack repair.Aiming at the problems of irregular shapes,small target,sparse data

关 键 词：稀疏数据 深度学习 裂纹 涡轮叶片 相机标定 

分 类 号：TH741[机械工程—光学工程]