Research On Early Warning System For SMT Fault Diagnosis Based On Deep Learning

Electronic products are gradually becoming an indispensable part of the general population’s lives as time passes.SMT,as an advanced technology in the production of electronic products,is the most important part of the SMT production line in terms of production efficiency and product quality.There will inevitably be equipment operation failure problems,the need to perform temporary maintenance,or a heavy stoppage of production,all of which will have a negative impact on product production.As a result,equipment fault diagnosis is a much needed task,and optimizing this process is an area that requires urgent research.In the paper,deep learning algorithms are used to optimize the equipment fault diagnosis process.The use of deep learning algorithms to diagnose equipment faults can significantly improve detection accuracy and speed,increasing the productivity of SMT production lines.The details are as follows:First,this paper creates a Printed Circuit Board Assembly(PCBA)solder joint defect detection dataset by capturing images of PCBA solder joint defects collected by automated optical inspection(AOI)equipment on SMT lines.The dataset contains 759 images of PCBA solder joint defects with three types of defects: billboard,false welding and insufficient solder.Training was performed on this PCBA solder joint defect detection dataset using multiple neural network models such as Faster R-CNN,YOLO series,and YOLOX series,and the training results were compared and assessed to select the best model as the base model for subsequent model improvement.Second,the ConvNeXt-YOLOX network is created by incorporating various structures into the ConvNeXt image classification neural network and upgrading the YOLOX neural network’s backbone feature extraction network CSPDarknet53.The PCBA solder joint defect detection dataset is used for training,and comparison experiments are run to compare the training outcomes with the original model YOLOX and the lightweight model YOLOX-s for analysis.The results show that the improved model ConvNeXt-YOLOX has a mean average precision(m AP)of 97.21%,which is 0.82% and3.02% higher than YOLOX and YOLOX-s,respectively,and the m AP@(0.5:0.95)is increased to 77.5;the detection speed is increased from 27.06 fps to 27.88 fps.The detection results for the different types of false welding reveal that the enhanced model ConvNeXt-YOLOX has better small target feature extraction and detection capacity,which is more in accordance with the actual needs of solder joint defect detection.Finally,using the PCBA solder joint defect detection results and SMT production line equipment fault history data,the SMT equipment fault diagnosis dataset was created,and the SMT equipment fault diagnosis model and PCBA solder joint defect detection model were obtained using probabilistic neural networks(PNN)training on this dataset.The SMT equipment fault diagnosis algorithm’s deployment environment are analyzed,the fault diagnosis process is designed,and the algorithm’s incremental update model is designed to realize the SMT production line equipment fault diagnosis early warning system to update itself along with each life cycle of the production line equipment to reduce the possibility of misjudgment.