Gráficos de controle para dados do tipo taxas e proporções autocorrelacionados

• Autor(a) principal: Tondolo, Cátia Michele
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do UFSM
• Texto Completo: http://repositorio.ufsm.br/handle/1/8392
• topic: Gráficos de controle; Autocorrelação; β; ARMA; Control charts; Autocorrelation; CNPQ::ENGENHARIAS::ENGENHARIA DE PRODUCAO

This work discusses one of the areas for quality improvement, defined as statistical process control (SPC). One of the most used tools in SPC is the control chart, which is used to monitor parameters of a process. In general, these charts are built under normality and independence assumptions of observations. However, sometimes these suppositions do not occur. The usual control charts work reasonably well if normal distribution assumption is moderately violated, but the violation of independence assumption reduces the applicability of them. When the data are autocorrelated, it is adequate to use residuals control charts, usually from ARIMA class. The residuals are used to produce the usual control charts, like the Shewhart, CUSUM and EWMA. In addition, variables restricted to the interval (0,1), such as rates and proportions, are naturally assumed to follow beta distribution. Thus, we propose the use of control chart with different residuals of the model βARMA to model and monitor autocorrelated beta distributed processes. The performance measures of the proposed control charts were evaluated by Monte Carlo simulations and the ARL (average run length) was analyzed under control and out of control. Proposed and traditional models were compared for autocorrelated data adjustment. Two applications were performed using real data associated to the volume of energy stored in southern Brazil and the levels of the sources of the Cantareira System (São Paulo, Brazil). The proposed control charts showed good performance for rates and proportions data, getting a better detection of special causes than usual modeling.