Abstract Preview

Here is the abstract you requested from the thermal_2013 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.

Transient Heat Exchanger Compact Models for Data Center Applications Using Artificial Neural Networks
Keywords: Data Center, Hybrid Cooling Systems, Transient Systems
Data center hybrid air/liquid cooling systems are capable of providing more efficient cooling due to their proximity to the heat dissipating servers. They offer the potential of providing local cooling only when it is needed, thereby reducing the overprovisioning that is endemic to traditional systems. The greatest efficiency can be gained when these cooling systems can be operated dynamically and synergistically with IT load scheduling under control systems that anticipate IT load shifts and dynamically provide localized cooling. The main component of every hybrid cooling systems is an air to liquid cross flow heat exchanger that regulates the amount of cooling provide by the system by modulating the liquid or air flows or temperatures. In dynamic cooling systems, understanding the transient response of the heat exchanger is crucial for precise control of the system. This presentation will present recent work performed in the NSF Center on Energy Smart Electronic Systems towards developing compact models of air/water heat exchangers that will be embedded in large thermodynamic simulators of dynamic data centers and eventually in sophisticated control algorithms for real time control of data center close-coupled cooling systems. The transient behavior of the cross flow heat exchangers is governed by three coupled partial differential equations that can be solved using numerical methods. The large computational time required by the numerical solutions make them unsuitable for any control application therefore fast compact models that can predict the transient behavior of the heat exchanger are required. In this paper heat exchanger compact models are developed using Artificial Neural Networks (ANN). The detailed transient behavior of the heat exchanger is obtained using a Finite Difference (FD) approach and the results provide the necessary training data for the ANN. The ANN based compact model is then tested under different temperatures and flow conditions and compared with the results of the full finite difference solution. Results show that the ANN approach can produce accurate and fast compact models of liquid cooled heat exchangers if the training data is sufficiently large and parametrically robust. However, the compact model derived using this approach is a true “black box” model that does not provide any physical insight beyond what can be learned from the input data.
Marcelo del Valle, Research Assistant
Villanova University
Villanova, PA

  • Amkor
  • ASE
  • Canon
  • Corning
  • EMD Performance Materials
  • Honeywell
  • Indium
  • Kester
  • Kyocera America
  • Master Bond
  • Micro Systems Technologies
  • MRSI
  • Palomar
  • Promex
  • Qualcomm
  • Quik-Pak
  • Raytheon
  • Specialty Coating Systems
  • Technic