Research Of Server Working Condition Classification Based On Infrared Thermal Image

In Cloud computation times, the safety and stability of network services become more and more important. Server temperature overhigh in data centers is a very important reason of server crashing down, which may cause internet service disrupt. Generally speaking, the temperature in data center can be maintained in a proper range by means of increasing power of air conditioner or optimizing server layout or other methods. But partial hot spot will be generated when abnormal working conditions occur such as server cooling fan failure or server computation overload, which further harms to internet service stability.In order to meet data center administrators'requirements of automatically recognizing server abnormal working conditions, this paper researches how to classify 3 common kinds of working conditions which could cause server temperature overhigh using infrared diagnosis technology. The contributions of this paper includes:On the basis of analyzing server infrared image and its features, according to the basic principles of the infrared diagnosis technology, this paper realized the whole algorithm process of server working conditions recognition based on infrared image from region of interest extraction, server working condition feature extraction and working conditions classification.This paper proposed an cabinet region extraction algorithm based on threshold to segment the cabinet region and background region. In the process of inspecting server edge in the server infrared images, the server edge in infrared images exists perspective distortion because of the specific structure of rack server and the unfixed infrared camera position. In order to solve the above difficulties, this paper proposed an algorithm based on gray-scale difference to inspect server edge in cabinet, which realized the server edge inspection without any inner and external parameters of infrared camera.Three common kinds of working conditions which could cause server temperature overhigh are designed in this paper, includes main fan failure mode, server single fan failure mode and CPU workload 100% mode, as well as idle mode compose the 4 test working conditions. In order to find the features that can distinguish different working conditions, on the basis of analyzing the characters of the above 4 working conditions, this paper proposed a feature extraction algorithm based on space and wavelet, and extract 28-dimensional feature used for classification. The feature extraction algorithm based on space develop the feature vector by means of the gray-scale distribution difference, while the feature extraction algorithm based on wavelet develop the feature vector by means of energy distribution in different frequency after infrared image wavelet decomposition. In server working condition classification phase, this paper takes the commonly used Fisher classifier and SVM as the classifier; use one-against-one method to make the SVM fit for multi-classification of 4 working conditions, and reached the goal of server working conditions automatic recognition. The experiment shows that the methods proposed by this paper can classify 4 known working conditions effectively.