Tecnicas estatisticas multivariadas para o monitoramento de processos industriais continuos

AUTOR(ES)

Celeste Maria Diaz Consul

DATA DE PUBLICAÇÃO

2002

RESUMO

In the industry of chemical processes, a lot of variabIes are manipulated and monitored at the same time. In these cases, it start to be of extreme importance the stages of data treatment and the deveIopment of models for representation of the processo One of the most important goals are detection and identification of faults in the processo Using multivariable statistical techniques, as Principal Components Analysis (PCA) and Fisher s Discriminant Analysis (FDA), is possible to take advantage of the data multivariabIe nature and it is possible to proceed with the detection of monitoring problems as well as diagnosing which the causes of these behaviors. In this work is considered as study case a hydrogenation of phenol to cyclohexanol reactor. Historical data, with a great number of variables and obsexvations, were collected during the operation of the processo The general idea of the method of PCA is to expIam the covariance structure of the data through some few lineal combinations of the original variabIes, which try to reflect the dimensions tru1y important. The acting of the process then can be monitored in the space of the principal components, of smaller dimension. Using the model PCA was possible the evaluation and identification of a group of faults in the processo On the other hand, using a bank of faults, adequately built, FDA got to classify the obsexvations with a good classmcation taxo A reflection on the importance of the use of these multivariabIe techniques for detection and fault diagnose is presented with the evaluation of the obtamed results.

ASSUNTO(S)

controle de qualidade localização de falhas (engenharia) analise discriminante analise de componentes principais engenharia - metodos estatisticos

Documentos Relacionados

• Aplicação de tecnicas estatisticas multivariadas e de redes neurais na modelagem de um sistema de tratamento de efluentes industriais
• Análise da percepção sobre assentos de trabalho utilizando técnicas estatísticas multivariadas
• Utilização de técnicas estatísticas multivariadas para definição de ambiente de produção do peso ao sobreano para o estudo da interação genótipo-ambiente em bovinos Canchim.
• Cartas de controle multivariadas baseadas no método Kernel-Statis para monitoramento de processos em bateladas
• Formação de grupos homogêneos de unidades de produção de leite bovino através de técnicas estatísticas multivariadas