| In actual production,the monitoring feature value is not limited to a single variable,and often is dependent multivariate,leading to traditional exponentially weighted moving aver-age(EWMA)control monitor invalid.Among the machine learning algorithms,the support vector data description algorithm(SVDD)is a single classifier,which does not require abnor-mal data during the learning process,and is suitable for high-dimensional situations with no requirements for data distribution,which just makes up for the defects of the MEWMA con-trol chart.Therefore,the introduction of the SVDD model into the MEWMA control chart has certain research significance.First,the paper independently generates training samples and monitoring samples that obey the normal distribution through simulation experiments,and uses the support vector data description algorithm to train the training samples to determine the center and radius of the hypersphere.Using support vector data to describe the misjudgment rate of the algorithm as a measurement indicator,continuously adjust the values of the algorithm parameters to obtain the optimal values of the parameters.Secondly,based on the MEWMA control chart,the paper introduces the idea of con-structing time series and historical data,combined with the SVDD algorithm,to design the MEWMA control chart based on the support vector data description.And build monitoring statistics.when a given control chart is in the control average running length,the control chart control limit solution step is designed to explore the performance of S-MEWMA con-trol chart prepare.Finally,compare the S-MEWMA control chart with theD~2control chart and the MEW-MA control chart,and use the out-of-control average running length as the criterion for good control chart performance,and continuously adjust the monitoring sample offset coefficient δ,S-MEWMA control chart parameters λ and sample variable correlation coefficient ρ are used for simulation experiments.Finally,the simulation results show that when the samples are not independent,regardless of whether the samples follow a two-dimensional normal distribution or a three-dimensional normal distribution,the performance of the S-MEWMA control chart is better than that of the MEWMA control chart and theD~2control chart. |
