Metodologia para a detecção da fonte de variabilidade em gráficos de controle multivariados para processos com dados de autocorrelação serial

• Autor(a) principal: Ueda, Renan Mitsuo
• Data de Publicação: 2023
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• dARK ID: ark:/26339/0013000015bf9
• Texto Completo: http://repositorio.ufsm.br/handle/1/28024
• topic: Multivariate control charts; Autocorrelated variables; Vector error correction; Autoregressive vectors; Hotelling’s T2 decomposition; CNPQ::ENGENHARIAS::ENGENHARIA DE PRODUCAO

The modernizing of the industrial manufacturing process requires simultaneous monitoring of product and process quality characteristics. Multivariate Statistical Process Control (MSPC) are able to assess the existing dependence between the investigated variables. In several hypotheses, the autocorrelation between the analyzed variables is present, where one of the main causes is the gradual wear and tear of machines and equipment. The use of multivariate control charts allows the process to be monitored in real time, ensuring the reduction of variability in the production system. The objective of this research is to propose a methodology based on residuals from vector autoregressive (VAR) models and vectors error correction (VEC), combined with Hotelling's T2 decomposition to identify the variable causing instability in the process. The methodology was tested through simulations and application to real data. Hotelling's T2 control chart, prepared from the residuals of the vector models, accurately indicated the intentionally incorporated outliers, and the T2 decomposition technique effectively pointed out the variable causing the variability. In terms of practical contributions, there was a combination of two distinct areas of knowledge: quality engineering and econometrics. The research sought to bring the academy closer to the industrial sector, since the focus of this methodology is to help managers and professionals who work in the area to deal with the presence of autocorrelated data in production processes. The use of this methodology helps in manufacturing competitiveness, and consequently, in the generation of employment and income for the sector. For future research, it is strongly suggested the application of this methodology in other industrial processes, combining other tools, and confronting them with different types of MSPCs.