Detection Of Server Thermal Distribution Based On Infrared Image

With the great energy consumption in data center,data center energy management issignificant for guaranteeing its safe and reliable operation.One of the main factors that lead to the high energy consumption of the data center is the temperature imbalance.However,the existence of "hot spots" in the data center is the main reason for the imbalance of temperature distribution.If there is a solution to the "hot spot" problem,Data center energy consumption can be reduced.Therefore,accurate positioning of the hot spots in the server surface temperature distribution is very important to study the automatic diagnosis method of data center hot spots.In this paper,the surface temperature distribution of each server is automatically extracted based on thermal image technology,and it provides data for server state detection.This paper proposes a method based on image registration.First,the visual and corresponding thermal images are matched based on the texture characters of the thermal images obtained in data centers.Then,the thermal distribution of each sever is standardized.The feature of the hotpots or server extracted from the standard distribution can diminish the impact of the position and direction of the cameras.The experimental results show that the visual and corresponding thermal images can be registered correctly based on the proposed method.The position and direction of the cameras can affect the thermal features of the server or hotpot.Thus,if the features are used to recognize the causes for the generation of hotpots or other thermal information,the direction of the cameras should be perpendicular to the surface of the servers.Otherwise,the proposed standardization procedure can decrease the error to a certain extent.According to low SNR of the thermal infrared images and the particularity of environment in data center,the method can obtain horizontal edges and vertical edges of cabinet's optimal angle in thermal infrared image with steerable filters,and then complete the server equipment automatic detection in infrared image with hough linear detection and linear fitting method.In our research,the experiments were conducted by thermal infrared images of different camera angles.But frankly speaking the shooting angle will affect the detection results.For the same shooting angle,compared with the traditional hough linear detection.The method proposed in this paper can more accurately extract the temperature distribution in the area of the servers.