Fault Diagnosis On Refrigeration System Based On Compensatory Fuzzy Neural Network

With the development of artificial intelligence and industrial automation, the refrigeration and air condition equipment get more and more complicated than before according to the increasing automation function and the wide use of big-scale refrigeration equipment. As a result, the latency trouble points increased, it demands more ability in the system's trouble shooting。So, the fault detection and diagnosis of the refrigeration and air condition equipment have been widely given attention, and the investigation of it is becoming a hotspot in the world. It is just the whole background of this investigation.The following is the main contents:This paper is on the basis of a heat pump water chilly set, has illustrated the essential theory of system and neural network, and has also discussed some issues on compensative fuzzy neural network on the fault diagnosis of refrigeration system.On the basis of analysis of frequent fault character of refrigeration, gathers characteristic parameters while the refrigeration system goes wrong by data gathering system, then unifies them, at last gets the training examples by means of calculation. This paper mostly diagnoses five common faults, i.e. too little refrigeration, thermal expanding valve(TEV)open too big, the alterative flux of cooling water, suction and discharge valve destroy of compressor, the clogging in the suction pipe of compressor.The compensative fuzzy neural network model has been established with eight characteristic parameters,namely compressor input temperature, evaporator input temperature, cooling medium water input temperature, cooling medium water output temperature, compressor suck pressure, compressor discharge pressure, compressor output temperature, condenser output temperature as net input and fault patter as net output.There has programmed the training and verifiable simulation software of fault diagnosis on refrigeration system based on compensatory fuzzy neural network by MATLAB 7.0. By comparison, choose the appropriate fuzzy division number when the net is being trained to make the net reach faster expectant error.The net model has been trained by means of the training examples and has simulated the other training examples. Simulation results show that the system has a high rate of accuracy. As long as the refrigeration system and the training samples of the same fault, diagnosis system will be given the correct cause of the malfunction. So, the compensative fuzzy neural network is a useful tool on the diagnosis of the refrigeration fault.