Air Conditioning Fault Diagnosis System Based On Deep Learning

The external air conditioner is the main noise source of the air conditioner noise,and the noise of the external air conditioner has a direct relationship with the vibration.Before the air conditioner leaves the factory,the air conditioner manufacturer detects the fault according to the vibration of the outdoor unit of the air conditioner.In the production line,it mainly relies on the manual detection method to diagnose the fault of the outdoor unit of the air conditioner.To this end,this paper designed and developed a non-contact air conditioning outdoor unit fault diagnosis system.In this paper,the vibration excitation source and vibration transmission path of the air conditioner outdoor unit are analyzed first,which lays a foundation for the selection of subsequent vibration signal collection points.A six-dimensional vibration detection method is proposed.The six-dimensional vibration detection of air conditioner is realized by laser Doppler vibrometer and laser displacement sensor.The feasibility is proved by comparison experiments.Simulated seven different operating states of the air conditioner as a source of fault signals.Aiming at the disadvantages of the traditional variational mode decomposition denoising method,it is difficult to select the noise component.Based on the energy concentration ratio and correlation coefficient,an improved vibration signal denoising method based on variational mode decomposition is proposed.Experimental signal analysis proves its effectiveness.The empirical mode decomposition combined with support vector machine and variational mode decomposition combined with support vector machine are used to diagnose the fault of the outdoor unit of the air conditioner.The two fault diagnosis methods are proved to be insufficient in the fault diagnosis of the outdoor unit of the air conditioner.The experimental results show that the two fault diagnosis methods have difficulty in unifying the feature vector dimension when the feature is extracted,and the correct rate of model recognition is not high enough.Aiming at the shortcomings of the above two fault diagnosis methods,this paperproposes a method of fault diagnosis for external air conditioner by using deep learning network based on stacking auto-encoder and softmax classifier.The hidden layer and node of deep learning network are analyzed through experiments.The influence of the training parameters of the number,pre-training stage and fine-tuning stage model on the correct rate of the model.The experimental results show that in the six-dimensional vibration signal of the outdoor unit of the air conditioner,the correct recognition rate of the normal direction vibration signal of the detection surface is the highest,and the correct rate is 99.86%.In the process of training the deep learning model,the influence of the training parameters of the fine-tuning stage on the correct rate of the model is more important than the training parameters of the pre-training stage.The number of batch processing samples is controlled below 10% of the total sample,and the correct rate of the model is high.Smaller learning rates,exponential decay,and Dropout can improve the accuracy of the model,both in the pre-training phase and in the fine-tuning phase.Regularization has a positive impact on the correct rate of the model in the pre-training phase and a negative impact in the fine-tuning phase.Finally,using LABVIEW and MATLAB hybrid programming technology,the six-dimensional vibration detection fault diagnosis software for air conditioner outdoor unit was developed,and the air conditioning fault diagnosis was realized by deep learning.The software has real-time display of acceleration,velocity and displacement,real-time fault diagnosis,signal filtering,off-line analysis,excitation component analysis,correlation analysis,frequency doubling analysis and other functions.