Quality Control Of Product Composition Data Based On MCUSUM Control Chart Based On Integrated Algorithm

The increasing competition from downstream enterprises in the supply chain caused by the diversity of market demands makes it particularly important to ensure the quality of their products.Abnormal diagnosis technology can effectively improve the quality of products by determining the manufacturing status and identifying the causes for outliers while production.However,conventional control chart can only monitor the manufacturing status without identifying specific exception sources.Accordingly,in order to effectively carry out quality control,the approach of integrating statistical process control with intelligent algorithms has become the current research direction in quality diagnosis technology.In this approach,manufacturing status is monitored via the control chart.When the manufacturing process is in a runaway state,the exceptional patterns in the control chart are identified by intelligent algorithms and timely alarms are issued.Subsequently,the causing factors at the source of the outliers are diagnosed through the expert knowledge base,thereby resetting the process to normal status promptly.In the actual manufacturing process,the composition distribution in certain products affects the quality of the finished products.For the quality control of this sort of products when outliers occur,a multivariate control chart is designed to monitor the compositional data of the product and an improved Ada Boost-SVM recognition model is developed for the exceptional pattern identification.Firstly,a MCUSUM control chart based on Compositional Data(CODA)is constructed to monitor the manufacturing process.As compositional data are constrained by the fixed sum,the data are transformed by irl transformation.A control chart for detecting the mean shift of the compositional data is constructed based on the idea of MCUSUM control chart.Additionally,the optimal parameters and the average runaway chain length of the MCUSUM-CODA control chart are calculated using Markov Chain and Monte Carlo method.So as to compare the performance of the MCUSUM-CODA control chart and the T~2-CODA control chart.The results show that generally the MCUSUM-CODA control chart performs better than the T~2-CODA control chart.Secondly,an improved Ada Boost-SVM outlier identification model is constructed to diagnose the sources for the outlier in the manufacturing process.Based on the features of the compositional data,twelve mean-value exceptional patterns are designed to simulate the outlier while manufacturing by Monte Carlo method,with step shift in the control chart as an example.Following wavelet transform noise reduction,the SVM classifier is optimized by a difference algorithm,and subsequently the optimized SVM classifier is used to construct an Ada Boost-SVM outlier recognition model.The simulation experimental results show that the Coif4 wavelet function provides better noise reduction;the recognition effect of single classifier is inferior to the integrated classifier in all cases,moreover,the improved Ada Boost-SVM classifier offers the best recognition accuracy;with respect to the average runaway chain length and function,the applicability of the approach in this study is supported by the comparison with two separate approaches,the traditional control chart and the intelligent algorithm.Lastly,the application of the MCUSUM control chart based on integrated algorithms for product compositional data in port coal blending is exemplified by a simulated case.The result indicates the validity of the proposed monitoring and outlier diagnosis model.