Data-driven Manufacturing Product Quality Defect Analysis And Prediction Research

The manufacturing industry which has been rapid development and gotten long-term achievement from our country' founding to now,has become one of the major industries affecting our country's GDP.In resent year,with China's accessing to the WTO,foreign manufacturing enterprises have poured into the market.With the continuous improvement of the quality requirements of the people,the control of product quality was not satisfying with the quality inspection.So it have turned to the production process monitoring and analysis,hoping to improving and eliminating the poor quality of the factors to ensure that production's process is running smoothly and the production costs are reduced.The traditional quality control methods used artificial experience or SPC control chart to control the quality of products.This is a method after the control which can not meet the needs of manufacturing enterprises whom wants to monitor the quality of product requirements online.In order to solve the above problem:(1)In this paper,a quality defect analysis method based on knowledge discovery is proposed which is to replace the method that based on the experience or SPC control chart.Then Process quality defect correlation analysis model,And proposed a process of quality defect analysis and control of the association rule extraction algorithm based on GA algorithm.Finally,the effectiveness of the method is verified by the case of injection molding.(2)In the manufacturing process,the process quality data is in the form of data flow,and with the cloud computing and cloud storage development,make it possible that the cloud through the end to assist the process of quality prediction.In this paper,an on-line prediction architecture of the manufacturing process is proposed,which is based on the continuous training of AG-ELM through the mass production data at the cloud end.A prediction model of AG-ELM is constructed and the genetic algorithm(GA)to optimize the input weight and hidden layer threshold of AG-ELM.On-line part of improved K-means algorithm,using sampling method to determine the initial clustering center,then introducing the data flow calculation framework to the method.the K-Means parallel processing which greatly reduces the time required for the execution of the algorithm.Finally,the online processing of the data was input to thecloud end model to complete the prediction of the quality of the manufacturing process.The experimental results shows that the improved K-means algorithm has superiority in time complexity which the time complexity of the method is more obvious than traditional method when the data size and the number of parallel processing nodes increase.(3)Taking the spot welding manufacturing process as an example,the feasibility of the method of online forecasting of fusion products is verified.