Detecção de Falhas em Processos Industriais Operando em Múltiplas Regiões via Análise Externa com Múltiplos Modelos Lineares

• Ano de defesa: 2018
• Autor(a) principal: Fernandes, Renata Teixeira das Neves
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Industrial processes monitoring; Fault detection; External analysis; CSTR; Principal component analysis; Reator CSTR; Análise externa; Controle de processos industriais; Detecção de falhas; Análise de componentes principais; Controle de processo; Localização de falhas (Engenharia); Modelos lineares (Estatística); Eletrônica Industrial, Sistemas e Controles Eletrônicos; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9566

The conventional methods of Multivariate Statistical Monitoring have been widely applied to industrial processesmonitoring. A unique operation condition is usually assumed in these methods,i.e., theyoperate in a stationarystate. To handle processes that operate in multiple operation conditions,monitoring methods capable of differentiating normalchanges at operation conditions from processes faultsare required. Otherwise, high levels of false alarms are generated. Severalrecent studies propose solutions to this problem, among themtheexternal analysis, which does not require the definitionof delimited operating regions and neither trainings for each of them. However, when nonlinearities are present in processes,the application of linear external analysis impairs the quality of the monitoring. In these cases, non-linear external analysis can be used as an alternative. This solution requires complex definitions of functions and parameters that has significant influencein thequality of the model. Another option is the use of multiple linear models in external analysis.The main goalof thiswork is to investigate the possibility of using multiple linear models applied to the external analysis to deal withthe presence of nonlinearities in the process.Amethodology is proposed to detect ifmultiple linear modelsare necessaryandto carry out the construction andthe application of the multiple models in the process monitoring.Theproposed methodology is applied in the supervision of the simulated industrial process of a Continuously StirredReactor Tank(CSTR) and its performance is compared with the performance of the traditional external analysis method using a single linear model.The results haveshown improvement in the performance of fault detection when applied the multiple models methodology, which has also presentedbetter sensibility to faults then the single model methodology.Thus, based on the results obtained in this work, the application of multiple linear modelsis a viable alternative for fault detectionsin industrial processes.