Research On BGA Solder Joint Fault Diagnostics And Traceability Approach Based On Multi-model Fusion

As the development trend of electronic products is gradually oriented to high integration and miniaturization,the traditional packaging form can no longer meet the demand of high reliability,high quality and small and thin packaging,so the search for high-quality packaging technology and soldering process has become the focus of the industry's development.The quality of BGA solder joints depends on the regulation of the soldering process and directly affects the mechanical and electrical properties of the components.Therefore,the combination of solder joint fault diagnosis and process fault tracing technology is the most important to optimize the soldering process and ensure the good quality of solder joints.This paper focuses on the probabilistic fusion of machine learning and deep learning for solder joint fault diagnosis,and the traceability of the main process causes of solder joint faults,i.e.process fault traceability methods.In this paper,we define the failure types of solder joint bridges,solder joint voids,oversized solder joints and undersized solder joints according to IPC-7095D and IPC-A-610G standards,and design a BGA soldering test program through in-depth investigation of the reflow soldering process,using X-ray inspection equipment to collect50,000 geometric parameters and 40068 temperature parameter sets.In order to make each feature in the data is in the same range and in the same order of magnitude,the data is divided into 30,000 training data and 2,000 test data by random division in the fault diagnosis model,while the data is divided into 75%as the training set and 25%as the test set by cluster analysis in the process fault traceability model,and the data is touched up before inputting into the model.in order to achieve better model training and testing.Secondly,a solder joint fault diagnosis model and a process fault traceability model are designed and their model performance is analyzed separately.The proposed multi-model fusion fault diagnosis method(MMF)is performed by fusing a 20%weighted SVM classifier,a 50%weighted RF classifier and a 30%weighted 2D CNN network structure for BGA weld joint fault diagnosis and comparing the diagnosis results with those of SVM and RF classifier respectively to verify the experimental results show that the MMF diagnosis model improves the accuracy to 98.36%,which both Both the interpretability and the model has good stability.Further,for the reflow soldering process temperature characteristics of parameters including temperature rise rate(K_p),temperature drop rate(K_c),preheating time(T_p),holding time(_sT),liquid phase line time(_rT)and peak temperature(R)unreasonable settings based on MMF diagnostic model for the tracing of welding process failures,but in order to improve the model performance using SENe T network to improve the model,and with MMF model for comparative analysis.The results show that the improved model traceability results for,they have a high failure rate of 88.21%and achieved a traceability accuracy of 97%,an improvement of 3.8%.Finally,the cloud-based visualization system is developed based on Py Qt5cross-platform tool library,Django framework and Vue framework.In order to enhance the integration and practical application feasibility of the fault diagnosis model and fault tracing model,Django is used to build the back-end framework to enhance security and mysql is selected as the database,and Py Qt5 is used to avoid main thread jamming when data prediction is performed;and Vue is used as the front-end framework of the cloud system to facilitate the engineers to periodically regulate the model.