Control Systems & Virtual Experiments

5 September, 2017

With the development of HVAC systems, more advanced control strategies have been implemented to better meet the cooling/heating requirements. However, a data center is more complex. Constant changing of deployments makes a data center unique at different times during its lifetime. It is possible that an empty spot in a data center on day one may have a high-density rack on day 90, or even end up with housing an in-row cooler after several years. Obviously, there is no fixed control strategy that fits all the configurations. To this end, we use CFD for designing and optimizing cooling control systems in data centers instead of performing expensive testing or relying solely on rules of thumb. 

Like a regular HVAC control system, a data center control system consists of three main components: a sensor, a controller and a controlled device.  Sensors are placed at different locations in a data center – CRAC/H supply or return ducts for example, to measure variables like temperature, pressure, humidity, etc. The data read from sensors are then sent to a controller, which analyzes the input data and compares this data with the control set points, to generate control signals. Then, those signals are delivered to different end devices such as fan motors or pumps, to regulate the amount of airflow, coolant flow, etc.

 

Temperature Control

Take a simple 3D data center model, for example, which includes installed IT, cabinets, cooling units, PDUs, and underfloor pipes and cabling. Cabinets in top row have a loading of 7kW/rack, while the loading of rest of the cabinets is 2kW/rack. In this model, temperature is one of the controlling parameters used for airflow management; and two different control strategies are tested. The first scenario uses traditional return air control at a set point of 79°F, and the second scenario controls the supply air to 65°F. 

By looking at the results in the first scenario, there is a 4°F maximum supply temperature variance in the underfloor. To meet the same return temperature set point, cooling units on the lower half of the room are doing less cooling due to the unbalanced IT load in different rows, thus providing warmer supply air. Ideally, we would want to see a more uniformly distributed underfloor temperature, like the results of the second scenario. The images show a comparison of the underfloor temperatures between the two scenarios. A simple CFD simulation can help you visualize the impact of a given control system, and better assist a data center manager to make design and operational decisions.
 

Underfloor Temperature Control

Figure 1: (Left) Return Temperature Control; (Right) Supply Temperature Control

 

Pressure Control

Pressure is another important parameter as it helps drives the airflow. In most modern data centers, cooling units are equipped with Variable Frequency Drives (VFDs) that can be controlled to a given pressure set point to modulate the airflow, and dynamically meet the changing IT load. Running the VFDs at optimal speeds can aid in efficient operations. Depending on the configuration of the data center, pressure measurement setups could be different. For example, pressure sensors could be placed in the underfloor area of a traditional raised floor data center; while in a data center with aisle containment, pressure can be measured as the differential across the contained and uncontained space. 
 

Pressure Control

Figure 2: Pressure Section View Across the Data Center

 

However, choosing the ideal set point to deliver a suitable amount of airflow while maintaining efficiency is not that simple and straightforward. Multiple factors such as cooling unit type, underfloor obstruction density, etc. could affect the pressure distribution in the room. To get a better sense of how the control scheme affects the pressure distribution, we can modify the CFD model used earlier. 

In this model, the airflow is controlled to an average reading taken from 4 differential pressure sensors. The image below shows a schematic of this data center control system. 

 

Simple Control System Schematic

Figure 3: Simple Control System Schematic

 

Initially, a data center manager might set up a 0.05 in/H2O differential pressure set point based on experience. But when he runs a CFD simulation in the Virtual Facility, he notices that the target pressure difference of 0.05 cannot be in/H2O achieved as the ACU fans have already reached 100% of their rated flow.  Additionally, at this set point, the IT is overcooled making the intended energy saver -  VFDs –ineffective in this case. 

To locate a more reasonable and energy-saving set point, he can reduce the set point to 0.03 in/H2O. Fortunately, in this scenario, target differential pressure is met with only 88.6% fan flowrate compared to the initial baseline. More importantly, with a reduced set point, the IT is no longer overcooled. This reduction can be directly transfer to energy/cost saving, as expected by using VFDs. Of course, you can always try to reduce the target further more to maximize the saving, but it may affect the resiliency of your IT.  

Pressure Set Point

Figure 4: (Left) 0.05 in/H20 Pressure Set Point; (Right) 0.03 in/H20 Pressure Set Point

We just saw how CFD can be used to evaluate different types of control systems for cooling units in a data center; but the capabilities are not limited to just this. 6SigmaDCX also allows you to control various objects such as heat exchangers, rack-mounted fans, evaporative coolers, or even dampers, based on temperature, pressure, humidity and other parameters. 

Trying to seek an optimal operation for your data center environment? Set up the control systems in 6SigmaDCX, and explore your savings potentials.


Blog written by Chen Yang, Applications Engineer
 

Other Recent Posts


Data Centre World, 28-29 November, Frankfurt, DE

Come to Data Center World and visit Future Facilities at booth 560 with our local distributor Alpha…

Read More


Remove Modeling Headaches with Black Box Cabinets

Setting up CFD models can seem like a time intensive exercise, particularly when the user is forced…

Read More