2016 Volume 41 Issue 235 Pages 1-8
A data center is very important facility nowadays for supporting Internet services. In a data center, an air conditioner is used to maintain a suitable room temperature. Because the problem that occur in an air conditioner result in serious damage to a server system, the maintenance of an air conditioner is very important. Although many problems that occur in electrical circuits are easily detected, it is difficult to detect problems in a condenser, e.g. the debris within it. The most popular technique for maintaining a condenser is inspection and cleaning by service staff. This procedure is very simple, but it requires large costs and time. In this paper, we propose a new technique to estimate the condition of a condenser by using a heat balance model with dynamic data. The procedure of our technique is as follows. (1) Building and tuning of a heat balance model The heat balance in a condenser was calculated on the basis of Mollier chart. The heat carried by a refrigerant was transferred through a pipe to air by a cooling fan. After this process was modeled, the process parameters were determined by using dynamic data for a clean heat sink in a condenser. Here, the dynamic data were the data measured immediately after a compressor started. Moreover a clean sink was set without a net. On the contrary, a dirty sink was set by covering a condenser with nets. Experiments were carried out by changing the number of nets and the outside temperature. (2) Diagnosis of a condenser First, the output was calculated by using a model with the measured input data. Then, the calculated output and the measured output were compared to estimate the cleanliness of a sink in a condenser. Experiments were carried out under the following conditions. (a) Time to stabilize an air conditioner For experimental condition A, it took 3 h to stabilize an air conditioner. For experimental condition B, experiments were carried out 5 min after a compressor started. (b) Outside temperature The outside temperature was selected to estimate the effects of a season. The two temperatures of 15 and 35 °C were selected. (c) The number of nets The level of dirtiness of a condenser was controlled by covering a condenser with 5, 10, and 20 sheets of nets. At the beginning, the proposed technique was verified. A set of experimental input data for tuning the model parameters was input into the model and the difference between the calculated and experimental outputs was checked. In the ideal situation, no difference was assumed. In our case, a small difference remained. This difference was assumed as an index of diagnosis. If the estimated difference is almost as small as this difference, the condenser was assumed to be clean. Next, experiments were carried out under the above conditions. Finally, our proposed technique indicated an obvious difference for 10 and 20 sheets of nets under all experimental conditions.
