Defect Detection Algorithm Of Fiber Optic Coil Winding Based On Deep Learning

Fiber optic gyroscope(FOG)is an important device in the field of inertial navigation.Due to the simplicity of the device and the ring-shaped optical path,it has the advantages of high reliability,small size,low cost,strong anti-interference capability and high accuracy,and is widely used in aerospace and other fields.Fiber optic coil is the core component of fiber optic gyroscope,which undertakes the role of beam propagation in the optical path.Its winding quality is crucial to the accuracy of fiber optic gyroscope.In the process of fiber optic coil,defects such as stack and gap can be produced due to tension control,motor synchronization and mechanical vibration.The paper therefore focuses on the detection of defects in fiber optic coil winding.The following are the main work contents:(1)The working principle of each part of the fiber optic coil winding system equipment platform is introduced in detail,and according to the characteristics of fiber optic coil winding system image,the vision measurement system is designed;By analyzing the winding mechanism and defect principle,the types and judgment conditions of defects are obtained,and the actual requirements and boundary conditions of detection are established.(2)To identify the defects,the traditional image processing algorithm is used for the characteristics of the fiber optic coil image.Firstly,the region of interest is located,and preprocessing operations such as Gaussian filtering,Gabor transform,OUST threshold segmentation and Canny edge extraction are carried out;secondly,marker lines are set to remove irrelevant contours,and the contour points at the current winding position are subjected to polar calculation,and contour segmentation and least squares fitting are carried out according to the minimal value points;finally the coordinates of the center of the circle that can replace the position of the fiber optic wire are obtained,and the defect type and position are located according to the determination conditions to complete the detection of the defect.(3)A YOLO model-based defect detection method of winding of the fiber optic coil is investigated.Firstly,the images were collected and labeled with defects,and a defect dataset was established.In view of the poor recognition effect of the network for small targets,the YOLO algorithm is improved,and the network structure,anchor frame setting,loss function and other aspects are optimized to enhance the feature extraction ability and detection accuracy of the algorithm.Experimental verification show that compared with other defect detection algorithms,the proposed algorithm achieves the best detection accuracy and detection speed in defect recognition.To sum up,this dissertation has completed the design and verification of the algorithm based on the background of optical fiber winding system and the purpose of defect detection,so that the proposed algorithm can achieve the experimental purpose and complete the defect detection requirements,providing a certain reference value for the automatic research of optical fiber winding machine in China.