Study On Statistical Process Control Of Joint Products In Rare Earth Separation

In recent years, the research of multiple auto-correlation control chart has causedwide attention of experts and scholars. The common methods for multiple auto-correlationprocessing are principal component analysis(PCA), multivariate CUSUM control chart,multivariate EWMA control chart and so on. Multiple auto-correlation control chart is fora variety of a product, but in many process in industrial production, there will be many co-products, such as in the oil industry, it will have kerosene and petrol as a co-product, soapand glycerin as a co-product in the soap industry, in the rare earth separation enterprise,there will be easy component extraction and difficult component extraction. Because ofthe characteristic of the process, the data we collected from the production site can havethe relevant features, and the multiple auto-correlation control chart cannot be applied tothe joint production process quality control. The common researches involving the jointproduct are cost accounting, production planning, sustainable development of the rareearth industry, etc., the control chart was not applied to process control on the quality ofthe co-product.Based on the separation of rare earth enterprise as the background, this papercombined the feature of co-product, by drawing the control chart of key technologicalparameters which can influence the co-product significantly to monitor the productionprocess is whether normal or not. Pattern recognition is used to diagnose the cause of theout-of-control chart influenced by key technological parameters. First, we choose the auto-correlation data from the export of the organic phase and water phase in the A rare earthseparation enterprise. Then we draw the Exponentially Weighted Moving Average(EWMA) control chart of the single product respectively, considering the correlation ofthe co-product, the multiple residual T2control chart of the joint product is drawled. In theend, we compare the two kinds of control chart.Second, we do the control chart pattern recognition to recognize the reason for alarm signal of residual T2control chart. The model of support vector machine for multivariatecontrol chart based on alarm signal to do classification processing. We put multivariateprocess control charts alarm signal corresponding to the quality of the feature data and thecorresponding characteristic value as the input samples, and choose50%of them as thetraining sets, the remaining50%as test sets, and then use the training sets based on theSVM classification model for training, and then the resulting model to predict the categorylabel for the test sets; Finally, according to the analysis of the SVM parameters influenceon its classification performance, Particle Swarm Optimization (PSO) algorithm ispresented in this paper, the PSO-SVM classifier is constructed and the result is analyzed.The results about the rare earth separation enterprise show that in the industrialprocess, the multiple residual T2control chart considering the feature of multiple auto-correlation can be used for monitoring the production process of joint product. Patternrecognition can diagnose the causes for the abnormality of the multiple auto-correlationresidual T2control chart. The multiple control chart drawled by the multiple attribute of aproduct appear abnormality, it can explain the product is not normal, while with the keytechnical parameter control chart drawing is abnormal, it cannot explain the joint productis abnormal, only that the key technical parameters affecting the product is abnormal.