Detection And Diagnosis For Refrigerant Leakage Fault In Data Center Air Conditioning System

Refrigerant leakage is a common fault during the operating of refrigeration system.For a data center air conditioning system,if the fault of refrigerant leakage cannot be found in time,the temperature of the room might be too high,even caused the consequence of downtime.Based on the existence of the complete structural parameters of refrigeration system,this paper proposes a refrigerant leakage diagnosis method based on the hybrid logic of physical-data mining,and uses the experimental data to verify the diagnosis method.The results show that hybrid logic method has good detection and diagnosis effects on refrigerant leakage.In this regard,this paper carried out the following work:First of all,for the system of the structural parameters of the known refrigeration system components,a physical-based refrigerant charge estimation method is proposed.According to the approximate linear relationship between the flow rate and frequency of inverter compressor,a virtual sensor model of compressor mass flow rate is established.Using the principle of conservation of energy and conservation of mass,a physical simulation model of each component is established.The temperature,pressure and control signals of the acquisition system at different positions are input to calculate the refrigerant quality of each component.The total system charge is the mass of all components.Since the void coefficient has a great influence on the mass calculation of the two-phase region of the refrigerant,six kinds of void coefficient models are modeled separately and the one with the highest accuracy is selected to optimize the physics.Secondly,for the system with unknown structural parameters of each component,but with historical operating data,a refrigerant leakage fault diagnosis method based on data mining is established.According to the characteristic of liquid pipe length in air conditioning system of data center,an improved gray model of filling quantity to increase the characteristic index of liquid pipe pressure drop is proposed,and the neural network is established to establish the gray box-neural network diagnosis model.The accuracy of the gray box neural network model is greatly improved compared to a single gray box model of undercooling and superheating.Finally,a hybrid logic fault diagnosis method based on physical-data mining is proposed.Based on the experimental data of a data center air conditioning system with a rated capacity of 25 kW at different recharge levels,the model is tested.The results show that the model can effectively detect and diagnose the current system refrigerant charge level and has good effect.