基于改进Faster R-CNN的航空发动机制件表面缺陷检测算法  

Aeroengine Parts Surface Defects Detection Algorithm Based on Improved Faster R-CNN

作　　者：唐嘉鸿 黄颀 田春岐[2] TANG Jiahong;HUANG Qi;TIAN Chunqi(AECC Shanghai Commercial Aircraft Engine Manufacturing Co.,Ltd.,Shanghai 201306,China;College of Electronics and Information Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]中国航发上海商用航空发动机制造有限责任公司,上海201306 [2]同济大学电子与信息工程学院,上海201804

出　　处：《机床与液压》2022年第23期93-98,共6页Machine Tool & Hydraulics

基　　金：上海市工业互联网资助项目(2018-GYHLW-02043);上海市人工智能产学研专项资助项目(PKX2020-R13);上海市信息化发展专项资金项目(201901010)。

摘　　要：针对现有航空发动机制件缺陷检测所存在的检测准确率低、速度慢等问题,提出一种基于改进Faster R-CNN算法的缺陷检测方法。该算法使用深度残差网络提取缺陷特征,采用含有内容感知重组的特征金字塔模型融合各层次特征图,并根据检测框尺度选取相应层次的特征图进行检测和识别,在RCNN部分使用分层采样实现挖掘难例,增强模型对难例样本的学习。实验结果表明:所提算法具有较高的检测准确率,而且能够有效提升检测速度。Aiming at the problems of low detection accuracy and inefficient detection in defect detection of aeroengine parts, an algorithm based on improved Faster R-CNN was proposed.In this algorithm, deep residual network was used to extract features of defects.The pyramid model with content-aware reassembly of features(CARAFE) was used to integrate the feature maps of each level, and the corresponding level feature maps were selected for detection and recognition according to the scale of anchor boxes.In the part of RCNN,stratified sampling was used to mine hard samples to enhance its learning.The experimental results show that the proposed algorithm possesses higher detection accuracy and the detection speed is improved effectively.

关 键 词：制件缺陷检测 Faster R-CNN算法 内容感知重组 分层采样 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术] TG115[金属学及工艺—物理冶金]