Diagnosis And Evaluation For The Thermal Faults In Heat Pump Air Conditioning System

The failure of the heat pump air conditioning system will result in increased system energy consumption,decreased cooling capacity,COP reduction,deviation from the design conditions,shorten the equipment life.In this paper,seven common thermal faults of heat pump air condition system are researched.Seven common thermal faults researched in this paper is compressor valve leakage;insufficient air flow in the condenser coil;insufficient air flow in the evaporator coil;liquid line restriction;under charge;over charge and non-condensable in the refrigerant.The experiments were carried out under the condition of fault free and seven kinds of faults under the four typical operating conditions.Fault detection and diagnosis is researched in this heat pump system.In this paper,two kinds of fault diagnostic methods are proposed.One fault diagnostic method is based on the radial basis of neural network model.The other fault diagnostic method is based on adaptive machine learning.The first diagnostic method needs to establish the radial basis of neural network as system model.The method comperes the system model output and experimental data and diagnoses faults through analysis rule of fault characteristic.First,we need to establish a system model.Then,establish a diagnostic rule based on fault characteristics.Finally,compare the system model output and experimental data to diagnose faults.The analysis rule of fault characteristic is established with the experimental analysis system under various failures.The second method is based on subtractive clustering and support vector machine.First,the subtractive clustering model needs to be established based on experimental data.The subtractive clustering model is used to adaptive identification conditions.Then,we need to establish a support vector machine fault diagnostic model to diagnose data.For the support vector machine model,this paper uses "one to one" method to diagnose seven kinds of faults and uses 5-fold cross validation method to optimization model parameter.The method is based on artificial intelligence methods,independent of the precise system model.In this paper,the two methods proposed in this paper are validated according to experimental data and evaluated by five evaluation indexes proposed in this paper.The five evaluation indexes are correct rate,false alarm rate,leakage alarm rate,misdiagnosis rate and no response rate.At the same time,this paper compares the two methods with the method based on RCA rules.According to the diagnostic results and the evaluation index,the diagnostic accuracy of the diagnostic method based on adaptive machine learning is the highest.The diagnostic method based on the radial basis function model is the second.The diagnostic accuracy of the two diagnostic methods is better than diagnostic method based on the RCA rule.The diagnostic method based on the RCA rule is the lowest.But the diagnostic method based on the radial basis function neural network model diagnoses small fault data incorrectly.The method based on the adaptive machine learning diagnostic method diagnoses medium fault data incorrectly restricted to the training data.Therefore,the diagnostic method based on the radial basis of the neural network model is more suitable for users.Finally,according to the diagnostic results and practical application,it can be seen that the diagnostic method based on the radial basis function model and the diagnostic method based on the adaptive machine learning are better than the method based on RCA rule with experimental data.If we don't have experimental data,the RCA-based diagnostic method is better.