论文部分内容阅读
Goes by the thought of Data Mining is more popular and the developed methodologies are mature, we would implement these techniques in another important application-the manufacturing.Within this paper, we would introduce three skills that might bring the most significant performance in quality control problems.They are (1) Root Cause Analysis, based upon the Feature Selection methodology, (2) Predictive Quality Control Charting, based upon the Neural Networks for Time-Series problem-solving and (3) the MSPC which is using PCA/PLS for multivariate statistical process control missions.We would use three fictitious datasets for presenting these three methodologies, the fundamental concepts and the applications.For (1) above, the dataset with 2,858 variables and 2,062 cases would be used and we would figure out the first 20 significant effects (i.e., including the 2-ways interactions) for wafer_yield based upon so-called ANCOVA-like skill.For (2) above, we would use Neural Networks to create a predictive model for further prediction.The dataset we use is with 3 variables and 978 cases.Latter, we would present the MSPC application based upon PCA (principal component analysis) and PLS (partial least squares).For this case study, we would use a specified Time-wise and Batch-wise processing dataset with 12 variables and 3,000 cases.From this presentation, we bring an interesting idea for implementing Data Mining concepts and methodologies onto quality control problems.And also, we believe Data Mining should be a good reference skill for real manufacturing.