Metodologia detecção e diagnostico de falhas em sistemas de refrigeração usando redes neurais

Abstract

This work presents a methodology of failure identification based on the inter-relation analysis of some sensitive parameters to specific fails artificially imposed on a prototype refrigeration system using Artificial Neural Network (ANN) modeling. This methodology is tested in a small refrigeration system, the ANN are initially trained in a normal operation of the system (without faults). Were introduces three artificial faults on the refrigeration system: reduction of evaporator water flow, suddenly expansion valve closure and obstruction on the compressor suction line. The ANN compares the normal and failure operation using a residual error analysis in the majority of two sensitive parameters to detect an imposed fault. Also, is presented the training method used in the basic structure of the logic used to operate the proposed detection and failure diagnostic method (DFD). The methodology proposed is robust, achieving a good precision levei in the identification of faults introduced to the system.