Research On Multivariate Statistical Process Control Based On Autocorrelation Process

Statistical Process Control(SPC)is an effective method to improve product quality and save production costs.It has played a key role in economic development.Control charts,as the main tool of statistical process control,are an important means to monitor the quality of products in the production process.With the improvement of people's quality requirements,people's attention has been transformed from one quality characteristic to multiple quality characteristics,and the rapid development of today's data collection technology makes the collected data usually have autocorrelation.This article mainly focuses on the online quality control of multivariate autocorrelation processes:First,the influence of different autocorrelation structures on conventional multivariate control charts is analyzed.The MEWMA control chart is used as the research object.When the overall covariance value is different,the performance of the traditional control chart and the residual control chart is analyzed.Monte Carlo simulation compares the following two cases:(1)ignoring the autocorrelation of the original data;(2)considering the autocorrelation,the residual data constructed according to the multivariate time series model VAR(1)of the synthesized raw data.The average run length(ARL)under different autocorrelation structures and different average correlation lengths are compared,and then the performance of the control chart monitoring offset under each condition is analyzed.The analysis results show that the performance of the residual MEWMA control chart is the best,and when S2 is used as the overall covariance value,the controlled average running chain length ARLO of the residual MEWMA control chart is lower,but the change of the mean can be found faster.Secondly,in order to improve prediction accuracy and achieve effective monitoring of multiple autocorrelation processes,a Support Vector Regression(SVR)model based on chaotic genetic algorithm(CGA)optimization is proposed,referred to as CGA-SVR,this method can not only filter out the autocorrelation effectively,but also overcome the shortcoming of hyperparameters of the SVR model falling into the local optimum prematurely,and then build a multivariate residual MEWMA control chart based on the model to monitor the mean of the statistical process.Through an example analysis of the two quality characteristics of the rotary vane pump noise and temperature,compared with the traditional SVR model,this model has the characteristics of predictive stability and high generalization performance,and the monitoring effect of the autocorrelation process is more obvious.Finally,in order to improve the monitoring sensitivity of the multi-element correlation statistical process,the cost loss caused by the monitoring process is reduced.In this paper,the Particle Swarm Optimization(PSO)algorithm is used to solve the economic and economic statistical design model of the autocorrelation MEWMA control chart.Firstly,based on the Taguchi loss function,the economic model of the autocorrelation MEWMA control chart is constructed.The differential particle swarm optimization algorithm is used to solve the optimal combination of the five parameters of the control chart,so that the unit time cost is the lowest.Finally,the experimental design method is used to analyze the sensitivity of the main parameters of the cost function and the control chart,thereby reducing the quality loss.