Study Of Fault Diagnosis Technology About Chilled Water Set Of Screw Based On Artificial Neuron Network

As central air conditioner more and more widely used, the fault detection and diagnosis (FDD) of the system is becoming an important task. The research in this field is quite limited in our country, especially on chillers with which traditional manual or half-automatic diagnostic tools can't meet the needs. How to use artificial neuron network to detect and diagnosis faults of chillers is discussed in this paper. This paper introduces several methods used for FDD and their characteristics. ANN model is influenced by many factors, people always used many styles to optimize the model in the past which is not a perfect way, so orthogonal experiment method is adopted to seek the best ANN model to FDD program.At first the basic research about FDD is summarized, and a detection model based on ANN is initially set up in the next section. . In the third section the paper presented experiment which simulate six faults including: change flow rate of chilled water, cooling water and refrigerant, charge non-condense gas, shift temperature of cooling water and alter outside cold load. Also a set of characteristic parameters are defined in order to differentiate these faults and clarify the reasons. At last a FDD tool is programmed based on ANN with experiment results which formed training stylebook and test stylebook.The paper adopt orthogonal experiment method to seek optimal network model, and the result shows the following reasons influence it gradually: learning rules, neurons of the hide layers, training times, normalization in literature and training goal. And the best model is Bayesian, 20 neurons of hide layers, max training time is 1500 and training goal is 0.001. Experiment results confirm the program can correctly judge the faults.