Research On Object Detection Algorithm For Disassembly Station Of Aero-engine Based On Vision

After the aircraft engine is put into use,it needs to be disassembled regularly for maintenance.Because the surface of the engine is full of pipelines,it can not be directly stressed during disassembly.Therefore,auxiliary disassembly stations must be installed on both sides.The manual visual alignment method is used to align the disassembly station with the disassembly frame on the transport vehicle for disassembly.This kind of manual disassembly method has low efficiency,high labor intensity and low degree of automation.With the rapid development of computer vision technology,object detection based on vision is widely used.In view of the above situation,this thesis takes the engine disassembly station as the target to conduct in-depth research on detection and positioning algorithm,proposes and designs the target recognition network algorithm based on deep learning,and completes the research on the positioning method of the disassembly station through binocular stereo vision,and verifies the algorithm with the experimental detection results.The main work of this thesis is as follows:Target detection of engine disassembly position based on deep learning.Considering the real-time and the requirement of the accuracy of the disassembly station,yolov3 network model is chosen as the basic network of the detection algorithm.The experimental results show that the original network model can quickly identify the disassembly position,but the recognition accuracy is low.At the same time,there are still some problems,such as missed detection,false detection and poor positioning accuracy of the boundary box to the target in the image.Aiming at the problems of the original network algorithm and combining with the characteristics of the target,an improved target detection algorithm based on yolov3 is proposed.By adjusting the prediction scale of the network and making full use of the context information and shallow details,the recognition ability of the algorithm to the disassembly position is enhanced;Secondly,the dependence of traditional K-means clustering algorithm is analyzed,and K-means + + algorithm is proposed to optimize the clustering.The experimental results show that the recognition accuracy of the improved yolov3 network algorithm is higher than that of the original algorithm,and the situation of missing and false detection is significantly improved.A fusion algorithm based on full connected network(FCN)and detection network is designed to achieve a higher precision positioning of the disassembly station boundary box.In this method,FCN is used to segment the object in the image at the pixel level.The mask of the segmentation result is mapped to the object detection output,and the classification result and the boundary box coordinate information of the object are obtained.The experimental results show that this method can not only predict the edge of the target more accurately,but also make the contour of the target more continuous and complete.The average intersection union ratio of the improved network model is higher than that of the original network model,and it can effectively solve the problem of poor positioning accuracy of the boundary box.According to the disassembly station detection method based on the fusion of target detection network and image segmentation network results,the disassembly station target detection system and side fixed axis system are designed,and the relevant experiments and analysis are completed.The experimental results show that the proposed method has high recognition accuracy for the disassembly position and high positioning accuracy for the target bounding box.The positioning accuracy of the side fixed axis system is good,and it can preliminarily judge the position state of the engine and the transport vehicle.