Research On Exponentially Weighted Moving Average Control Chart With Real-time-contrast Method

Detection of product quality characteristic changes is significant to ensure thestable operation of multivariate process in modern manufacturing industries.Statistical process control is used to detect the product quality characteristics changesin normal production process and give warnings to process exception. Control chartsare the most important statistical process control tools. Shewhart control chart, as theearliest control chart, is proposed by Dr. Shewhart in1920s. It only considers thecurrent data, so it is not sensitive to small shift in a process. Later, cumulative controlcharts and exponentially weighted moving average control chart appeared, they wereeffective in detecting small shifts. All these control charts are used to detect singlevariable change, so they are called univariate control charts. However, in modernproduction process, quality attributes are often associated with a multi-group ofvariables, the monitoring of these production process is known as multivariatestatistical process control (Multivariate Statistical Process Control, MSPC).In the field of multivariate process, control charts such as T2chart, MEWMAchart, MCUSUM chart, have already been used to detect process changes. Althoughthese control charts can be used to monitor multivariate process, they always assumethat variables follow normal distribution. However, with the development of modernmanufacturing technologies, many real-life data sets we get are often mixed numericand categorical, high-dimensional data or follow non-gaussian distribution, thuschallenging the traditional monitoring methods. With the evolving of moderncomputer technology, machine learning has become a research hotspot in this field.Artificial Neural Network (ANN) and Decision Tree (DT) algorithm, support vectormachines etc. have been used MSPC fields.This paper uses RTC (Real-Time-Contrast) method and applies random forest asclassifying tool to construct Shewhart-type control chart and EWMA control chart ofmultivariate process monitoring respectively. Then, numerical examples are used tocalculate and analyze ARLs of10&100dimensional normal data and2dimensionalnon-normal data. Results show that the method of EWMA control chart combiningRTC statistic is effective and better than other methods in monitoring highdimensional complex data.