Research On Server Thermal Fault Diagnosis Based On Infrared Images

As a key facility,data centers typically contain thousands of data processing devices such as servers,switches,and routers.With the development of the "Internet Plus" era,the number and scale of the data center have shown explosive growth.The abnormal conditions such as cabinet fan damage and long-term overload operation of server can easily lead to excessive temperatures of data centers.The method of further reducing the air conditioner temperature is usually used to maintain the safe and reliable operation of the server,resulting in huge power consumption.Therefore,the thermal fault diagnosis of the server can not only greatly improve the management efficiency of data centers,ensure the safe and reliable operation of the data center,but also can improve the air-conditioning cooling efficiency to some extent.By comparing and analyzing data center temperature field sensing methods,combined with the status of image processing and artificial intelligence technologies,this study proposes a server thermal fault diagnosis method based on infrared images.The purpose is to make energy-efficient thermal management decisions based on diagnostic results to optimize data center's energy allocation and thermal management and ensure that the data center operates reliably and reduces data center energy consumption.In order to verify the effectiveness of the method,this study designed four common thermal faults of the server.The infrared image of the server under different operating conditions was collected by the infrared camera,and the infrared image database of the server was established.This study starts with two key technologies of image feature extraction and classifier design,and designs two kind of server thermal fault diagnosis methods.One is a server thermal fault diagnosis method based on the combination of features and support vector machines.Firstly,the image enhancement algorithm highlights the heat distribution area of the server,and then the image normalization method is used to eliminate the geometric distortion caused by the camera's shooting angle and distance and map it to a unified image template.Then the texture features,the HOG features and the improved entropy features proposed in this paper are extracted from the normalized image.These features are used to train the SVM model.Experiments show that in this method the improved entropy features combined with the SVM can achieve better results,the highest accuracy of thermal fault diagnosis is 85.3%.The other is a server thermal fault diagnosis method based on convolutional neural network.Convolutional neural network has the characteristics of automatic learning characteristics from end to end,it can eliminate the cumbersome process of manually extracting features.This study optimizes the parameter settings based on the Alex Net network model.The experimental results show that the accuracy of the server thermal fault diagnosis obtained by the AlexNet network model after optimizing the parameters can reach 95.3%.Comparing and analyzing the performance of the two methods of server thermal fault diagnosis,the server thermal fault diagnosis method based on convolutional neural network is more suitable for server thermal fault diagnosis.