Research On Performance Of MEWMA Control Chart Based On Probability SVDD

Statistical process control is mainly used in industrial production to ensure that the production process is in a stable state or within a certain error range.The control chart is a main tool for statistical process control,which has been developed for nearly one hundred years.However,the traditional control chart represented by the multivariate exponential weighted moving average(MEWMA)control chart requires the initial data to follow the normal distribution,which is not in line with today’s actual situation.With the rise of machine learning theory,support vector data description(SVDD)algorithm is widely used,and the initial data is no longer required to obey the normal distribution.The real-time contrast method(RTC),which is also a machine learning theory,has great advantages in dealing with highdimensional data.Therefore,it is of certain theoretical and practical significance to introduce SVDD algorithm and RTC method into MEWMA control chart.In this paper,the SVDD model is trained by stochastic simulation method,and it is improved to SVDD classifier based on probability(P-SVDD).After the model parameters are determined,the sample data is classified by classifier,and the classification probability of the sample is calculated by RTC method.Then,the P control chart is constructed by taking the classification probability as the statistic.Based on the theory of MEWMA control chart,the exponential weighted moving average control chart(P-MEWMA)based on P-SVDD is constructed.Under the premise that the average running length in the controlled state is fixed,the control limit of the control chart is determined by Monte Carlo stochastic simulation,so as to facilitate the subsequent research and discussion.After the model was constructed,the offset of the monitoring sample and the smoothing parameter λ of the P-MEWMA control chart were continuously adjusted by Monte Carlo stochastic simulation in the case of different distributions,and the average running length ARL1 under the out-of-control state was recorded.Compared with the original exponential weighted moving average control volume based on support vector data description(SMEWMA),the performance of them was judged.The simulation results show that the performance of P-MEWMA control chart is better than that of S-MEWMA control chart in detecting small offsets under normal or non-normal distribution.