Avaliação de Técnicas Baseadas em Análise Externa para Detecção de Falhas em Processos Industriais Operando em Múltiplos Modos

Abstract

Over the years, industrial processes have become increasingly complex and more instrumented. The proper monitoring of multiple signals simultaneously is challenge, even more when their operation conditions often changes. In this context, it is necessary to use techniques and methods that can realize the process monitoring in its most diverse conditions. Thiswork presents a study about classic techniques of multivariate process control, such as PCA and PLS, as well techniques that use models to explain the process behavior. To addressthe multiple modes question, the external analysis is used, to explain the changes in the process operational conditions. Three case studies were developed: the first using the CSTR simulator, the second and third use industrial data from an industrial fan. In the situations analyzed in the case studies, the methodology was able to achieve a correct fault detection, regardless of the process operational condition.