基于热成像的机房热点成因自动诊断方法

Automatic Diagnosis of Hot Spots in Computer Rooms Based on Thermal Imaging

摘要:

服务器设备的异常高温在机房内部会形成热点,不仅会影响服务器的稳定和寿命,还会导致机房制冷效率的降低,从而增加机房的制冷能耗,增加运营费用。导致产生热点的原因有很多,例如空气流通不畅、风扇失灵、长时间满负荷运行等等。通过自动诊断热点的成因,可以有针对性的消除热点,为机房环境控制提供数据支持,有助于降低机房制冷能耗。根据热像仪拍摄的服务器出风口一侧的红外图像,利用人工智能技术,提出了自动诊断热点成因的方法。针对实际工程应用中热点样本数量不足的问题,提出了基于深度卷积对抗生成网络(DCGAN)合成热点样本的解决方案。通过多组实验验证了方法的有效性,热点成因的诊断准确率约为95%。

Abstract:

The abnormal high temperature of server equipment will form hot spots in the computer room, which will not only affect the stability and life of the server, but also lead to the reduction of the cooling efficiency of the room, thus increasing the cooling energy consumption and operating costs of the room. There are many reasons for hot spots, such as poor air circulation, fan failure, long-term full-load operation and so on. Through automatic diagnosis of the cause of hot spots, the hot spots can be targeted to eliminate, provide data support for the environment control of the computer room, and help to reduce the cooling energy consumption. According to the infrared image taken by the thermal camera on the side of the outlet of the server, a method of automatically diagnosing the cause of hot spots is proposed by using artificial intelligence technology. Aiming at the problem of insufficient hot spot images in practical engineering application, a solution based on Deep Convolution Generative Adversarial Networks (DCGAN) composite hot spot images is proposed. The validity of the method was verified by multiple experiments, and the diagnostic accuracy was about 95%.