基于关键点检测的航空发动机螺栓安装缺陷自动化检测方法

In view of the problems of complex background, small target and inconspicuous fine features of aero-engine bolts, an automated detection algorithm of aero-engine bolt installation detection based on key points detection is proposed. First, a cascaded convolutional neural network based on the Faster RCNN and the improved CPN (AD-CPN) is proposed to achieve the detection of bolts and 2D key points which can determine whether the bolt has fallen off or missed. To further detect the 3D installation detection of the bolt, the Euclidean distance selection strategy is introduced to match and screen the key points to obtain the detected point pairs. Finally, the 3D coordinates of the key points are calculated by using binocular stereo vision technology. In this way, it can judge whether the bolt is wrongly installed. Compared with CPN, the mAP,AP₅₀, and AP₇₅ of AD-CPN are improved by 2. 9%, 3. 3%, and 4%, respectively. In addition, the relative average error of bolt measurement length is approximately 3. 0% . It can be seen that the algorithm could enhance the accuracy of detection, and ensure the safe operation of aero-engines, which has great practical significance.