Crack Detection And Optimization Design Of Section Quality For Self-piercing Riveting

As an important way for energy saving and emission reduction,automobile lightweight has become an inevitable trend of development.It’s a suitable way to realize the lightweight design by using the lightweight materials.Most automobile manufacturers use lightweight materials and steels in combination to reduce manufacturing costs,such as aluminum alloy steels.In the connection of dissimilar materials,self-piercing riveting(SPR)has greater performance than traditional welding,which has been widely used.The quality of self-piercing riveting will directly affect the safety and durability of automobile.Therefore,in the initial joint development process,the quality of SPR riveted joints should be inspected and classified to meet the required standards.This paper divides the SPR quality into riveting appearance quality and joint cross-section quality to carry out research.For the riveting appearance quality,the surface cracks under riveting will seriously affect the joint corrosion resistance and the riveting strength.Nowadays,state-of-art SPR joint inspection method is a manual visual process,which is time-consuming and rely on high-level trained engineers to distinguish features subjectively,which can not be applied on a large scale.Therefore,based on image processing and computer vision,a local-global strategy for automatic crack detection of SPR is proposed.For the joint cross-section quality,the finite element simulation model is established,and the riveting process and section quality are analyzed by numerical simulation.Besides,a multi-objective optimization method is proposed to improve the joint cross-section quality.Firstly,the full-size SPR images in different scenes are obtained.The sub regions in the full-size images are cropped and preprocessed to form the sub image sample dataset.Different kinds of feature values are extracted from the dataset,which are used as input to train several crack detection networks based on sub images.Considering that the crack detection network based on feature extraction still has a large room for improvement in accuracy and the limitation of image representation by feature description operator,two sub-image crack detection networks based on convolutional neural networks are constructed.The crack detection accuracies of these sub-image crack detection networks are tested,and the results show that the accuracy of crack detection network based on convolution neural network is significantly improved.Secondly,the traversal search algorithm is written to extend the application of the constructed sub-image-based crack detection network to the crack detection area of SPR full-size image.After that,the automatic detection of SPR cracks is completed.Four representative full-size images are used to show the detection results of different crack detection models.The final results show that the crack detection network based on convolutional neural network gets the best detection effect.Finally,a SPR finite element simulation model is established.The riveting process and section quality are analyzed according to the simulation results.Four parameters which have an important influence on the joint cross-section quality are selected as design variables,the joint cross-section quality parameters are taken as optimization objectives,and the quality standards which are widely used in factories are taken as constraints.Approximate models are established for multi-objective optimization design,and the Pareto optimal solution under the optimization objective is obtained.The optimized design parameters are substituted into the SPR finite element simulation model to prove the effectiveness of the optimization results.By comparing the quality parameters of joint section before and after optimization,the self-locking amount and bottom thickness value are significantly improved,and the quality of riveted section is improved.