Big Data-driven Quality Management Of Small Batch Multi-variety PCB Templates

China has become a major nation of electronics manufacturing,and the corresponding printed circuit board(PCB)production capacity has also been gradually shift to China.In recent years,the rapid updating of electronic products has led to a substantial increase in the number of individualized PCB templates.Studying the quality rule of PCB template,guiding the optimization of material feeding and controlling the production process are the key to reduce the comprehensive cost of workshop.At present,the quality control of most domestic PCB template production enterprises mainly depends on manual experience.Multi-variety and small-batch orders lack overall guidance.It is difficult to determine the quality rule of each order by basic statistics because of the great differences in the requirements of sample design and many factors affecting the quality.There are many production processes,so it is difficult to control the quality of the process.Because of the small batch production,it is difficult for workers to form high-quality operation habits.In the early stage of the project,we used data mining methods such as multiple regression analysis,decision tree,support vector machine and BP neural network to carry out systematic research on PCB template.However,based on the research results and further investigation,the following problems still need to be solved:(1)There is a lack of quantitative evaluation on the factors affecting the quality of PCB templates.Combined with feature selection,the key factors for the forecasting model construction can be better optimized.(2)There are great differences in order structure.It is helpful to improve the accuracy of the forecasting model by dividing the orders and optimizing the key factors in each group.(3)Combining the characteristics of PCB template,the research of advanced prediction algorithm will further improve the accuracy of prediction model.(4)Order grouping has the characteristics of fuzziness and randomness in initialization.It is necessary to further study the prediction mechanism incorporating the characteristics of fuzziness and the optimization mechanism combining heuristic algorithm.Based on the above reasons,the research is carried out from the following four aspects:(1)Identify attributes that affect quality and prioritize them to provide a high-quality data source for building models.The basic analysis of data attributes and feature selection are carried out to find the key attributes that continuously produce high-quality products.(2)The order classification model is established based on the CHAID decision tree,by which the scrap rate of the order is subdivided into different categories.Analyzing the relationship between scrap rate and attributes is helpful for enterprises to manage orders more reasonably.(3)Based on FCM clustering and BP neural network,a PCB scrap rate prediction model is constructed.The prediction model aggregates all data into several groups and makes prediction.Compare the difference between the predicted results after clustering and those without clustering.(4)A new model combining FCM clustering,GA and BP neural network is used to verify the ability of the prediction model combining fuzzy characteristics and heuristic algorithm to solve problems.Based on the big data-driven quality management of small batch and multi-variety PCB templates,will promote enterprise product improvement,improve production quality and customer relationship,reduce the risk loss of design,production and sales,and enhance enterprise efficiency.The SOM-BP model reduces the surplus rate and supplemental feeding rate from 27.95% and 17.91% of manual feeding to 10.13% and 9.37% respectively.In addition,after using FCM clustering to solve the fuzzy problem of data and GA to optimize the neural network,the FCM-GABP model reduces surplus rate and supplemental feeding rate to 8.42% and 12.24% respectively.