Raciocínio baseado em casos: uma abordagem utilizando o sistema imune artificial

• Ano de defesa: 2009
• Autor(a) principal: Eduardo Enrique Ostos Carvalho
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Sistema imune artificial; Raciocínio baseado em casos; Detecção de falhas; Motor elétrico; Anticorpo; Antígeno; Engenharia elétrica
• Link de acesso: http://hdl.handle.net/1843/BUOS-8CZLVX

Case-Based Reasoning (CBR) is a powerful method for solving problems and is fundamented on the human cognitive system, which is strongly based on solving problems based on known similar ones. The solution is given by searching a memory for a similar situation, which in turn can or cannot be adapted to the current state. After searching for the more similar situation, it's analyzed to detect the adaptation needs, in order to avoid unnecessary effort and at the same time reuse previously validated knowledge. Then, if necessary, the solution is evaluated to ensure its effectiveness. Finally, the case, or problem solved, is stored together with others, helping to increase the system knowledge. These four steps, or processes, operated together, constitute the CBR life cycle. CBR is first and foremost a methodology, and as such, does not specify operational details of each of its processes, what opens the possibility for the use of different paradigms and computational tools. On this point of view, the Artificial Immune System (AIS) is an interesting computational paradigm that can be used in conjunction with CBR. The reason is that it relies on sophisticated attributes of the human biological system, such as pattern recognition, data compression, population generation and dynamic environment adaptation, all those relevant characteristics to CBR. This work presents a hybrid model of CBR and AIS, responsible for generating contributions in the process of recovery (search and evaluation), adaptation (reuse and revision) and retention (storage) of cases. Among the most relevant contributions are the creation of an alternative way of grouping cases, identifying high density areas, improve search efficiency in the case space and store the relationships of similar cases. The proposed model is applied to fault detection and diagnosis problem of a DC machine, which operates with simulations on supervised an unsupervised mode. The hybrid method of CBR and AIS is confronted with other more traditional CBR search and storage techniques. Finally, the results are compared using specific CBR performance metrics, what at the end showed positive perspectives for the proposed model.