Research On Welding Seam Detection And Path Generation System Based On Deep Learning

With the proposal of the China Manufacturing 2025 strategy,all areas of my country's manufacturing industry are developing in the direction of automation and intelligence.As one of the important components of industry,welding technology is very important for industrial development.Weld seam recognition technology,as one of the core technologies to realize welding automation,has also aroused widespread concern in academia and industry.The current research on welding seam identification and detection technology mainly focuses on the welding seam identification and positioning in a small window specific environment,and the research on multi-weld target recognition in a complex environment is not deep enough.Moreover,the relevant algorithm parameters are fixed,and the adaptability to the environment is poor.In recent years,the development of deep learning has provided new ideas for weld recognition.Therefore,in view of the problem that the current weld recognition algorithm has a small recognition window and poor algorithm adaptability,this paper proposes a deep learning-based weld detection system,which aims to realize the identification and detection of welds in complex environments and can adapt to changes in the target environment for parameters Learning adjustment,and finally complete the function module packaging and software development work.First,the construction work of the weld seam image data set is carried out.The Welding Seam data set constructed in this paper contains about 2000 weld image data,including butt joint welds,lap joint welds,and T-joint welds,and other types of welds.At the same time,flip,noise and other methods are used for data The expansion of the set and label the weld type and target as the location.This paper builds a deep neural network model based on Faster-RCNN.The model is divided into target extraction module,pre-selected area generation module and target recognition and position regression module according to function.The target extraction module is composed of multiple convolutional layers and pooling layers,and is used to extract image feature maps.The pre-selected area generation module mainly generates candidate areas on the basis of feature images,and finally completes the classification recognition task in the target recognition part,and completes the target positioning work in the position regression part.Through the training of the model and the debugging work,the identification and positioning of the weld seam in a complex background is realized.In order to realize the generation of welding path,this paper conducts the research of welding seam location and centerline extraction based on depth map in virtual environment.It mainly includes the segmentation of depth images,filtering and welding edge extraction,using the local maximum to extract the center line of the welding seam,and according to the coordinate transformation to realize the spatial positioning of the welding seam position and generate the welding path..Finally,the relevant software development work was carried out to realize the packaging of the welding seam recognition system function and the user-oriented design.The software is divided according to functions mainly including: Dataset module,Net Work module and image processing module.Use PyQT to design related user interfaces and use multithreading to separate the interface from the businessDeep learning technology provides new ideas and solutions for intelligent welding.This article uses deep learning technology to realize the identification and positioning of welds,and cooperates with the deep camera design path generation system to provide "smart" eyes for automation and intelligent welding.