Multi-Split Configuration Design for Fluid-Based Thermal Management Systems

Saeid Bayat, Nastaran Shahmansouri, Satya R. T. Peddada, Alexander Tessier, Adrian Butscher, James T. Allison

TL;DR

This paper introduces an automated framework for designing "multi-split" fluid-based cooling systems, which allow for complex branching architectures rather than simple parallel paths. By combining spatial clustering, graph-based physics modeling, and optimal control, the researchers can automatically generate and identify the most efficient cooling layout for high-density electronics, significantly extending the time devices can operate before overheating.

What Problem Does This Paper Address?

As electronics shrink and power densities soar, managing heat has become a primary bottleneck in engineering. Traditional thermal management systems (TMS) often rely on "single-split" architectures—where a single source sends fluid to various components in parallel branches. While simple, these designs are often sub-optimal for complex layouts with "hot spots."

The challenge is that as you move toward "multi-split" systems (where fluid can branch off at any point), the number of possible configurations grows exponentially. Designing these by hand is nearly impossible, and standard optimization struggles with the "discrete-continuous" nature of the problem: you have to choose a discrete layout (the graph) and then solve for continuous flow rates (the control) for every single layout to see which one is best.

Methods

The authors propose a multi-step automated pipeline:

  1. Spatial Clustering: They use K-means clustering on the physical coordinates of heat-generating devices to group them into "super-nodes." This helps determine where junctions (splits) should physically be located.
  2. Graph-Based Representation: Architectures are represented as acyclic undirected graphs. The "Tank" is the root, and Cold Plate Heat Exchangers (CPHXs) are nodes.
  3. Automated Physics Modeling: The framework automatically converts these graphs into dynamic state-space equations. It models the physics of advection (fluid moving heat), convection (heat moving from wall to fluid), and bi-directional advection (at the heat exchanger interface).
  4. Optimal Open Loop Control (OLOC): For every generated architecture, the system solves an optimization problem to find the ideal flow rate for each valve. The goal is to maximize thermal endurance—the total time the system can run before any single component hits its temperature safety limit.
  5. Enumeration: The framework generates all unique, feasible configurations within a class and ranks them based on their optimized performance.

Datasets

The paper does not use external datasets. Instead, it evaluates the framework using three synthetic but representative case studies:

  • Small-scale systems: 3 and 4 CPHX nodes with asymmetric heat loads (e.g., one node at 12kW, others at 1kW).
  • Disturbance analysis: A 6-node system tested under different "workloads" (heat load distributions).
  • Large-scale system: A complex 18-node graph (1 tank, 14 CPHXs, and 3 junctions) to demonstrate the framework's scalability.

Key Insights and Ideas

  • Multi-split > Single-split: The results consistently show that multi-split architectures outperform single-split ones. By allowing branching closer to the heat sources, the system can deliver "fresher" (cooler) fluid to high-load components more effectively.
  • Spatial Intelligence: Using the physical location of devices to define junctions is a powerful heuristic. It drastically reduces the search space of possible graphs to those that are physically sensible.
  • Load-Dependent Optimality: The "best" architecture isn't fixed; it changes based on the heat load. If one chip is significantly hotter than the rest, the optimal graph will almost always place that chip as close to the pump as possible to ensure it receives the maximum cooling capacity.
  • The "Aha" Moment in Flow Control: In some optimal multi-split cases, the controller actually restricts flow to certain branches initially. This "saves" cooling capacity for later, allowing all components to hit their temperature limits at exactly the same time, which maximizes the total system runtime.

Key Results

  • Performance Gains: In the 3-node case study with an extreme load (12kW vs 4kW/1kW), the multi-split architecture achieved a thermal endurance of 69.9 seconds, outperforming simpler series or parallel configurations.
  • Scalability: The authors successfully generated and solved an 18-node system, proving the code can handle the complexity of modern high-power electronics.
  • Trade-offs: The paper quantifies the trade-off between system performance and complexity (number of valves and junctions), helping engineers decide if a more complex layout is worth the extra cost.

Limitations and Future Work

  • Search Strategy: The current method uses "enumeration" (checking every possible graph). While effective for small systems, it won't scale to hundreds of nodes. The authors suggest moving toward Genetic Algorithms (GA) or Machine Learning to navigate the design space more efficiently.
  • Simplified Hydraulics: The model assumes zero heat loss through pipes and simplified fluid dynamics. Future iterations plan to include more sophisticated hydraulic models and 3D pipe routing constraints.
  • Knowledge Extraction: A major goal for future work is using Machine Learning to look at the "winning" designs and extract human-interpretable "design rules" that engineers can use as best practices.

Summary updated

Abstract

Abstract High power density systems require efficient cooling to maintain their thermal performance. Despite this, as systems get larger and more complex, human expertise and insight may not suffice to determine the desired thermal management system designs. To this end, a framework for automatic architecture exploration is presented in this article for a class of single-phase, multi-split cooling systems. For this class of systems, heat generation devices are clustered based on their spatial information, and flow splits are added only when required and at the location of heat devices. To generate different architectures, candidate architectures are represented as graphs. From these graphs, dynamic physics models are created automatically using a graph-based thermal modeling framework. Then, an optimal fluid flow distribution problem is solved by addressing temperature constraints in the presence of exogenous heat loads to achieve optimal performance. The focus in this work is on the design of general multi-split heat management systems. The methods presented here can be used for diverse applications in the domain of configuration design. The multi-split algorithm can produce configurations where splitting can occur at any of the vertices. The results presented include three categories of problems and are discussed in detail.
BibTeX
@article{bayat2024multi,
  author = {Saeid Bayat and Nastaran Shahmansouri and Satya R. T. Peddada and Alexander Tessier and Adrian Butscher and James T. Allison},
  title = {{Multi-Split Configuration Design for Fluid-Based Thermal Management Systems}},
  journal = {Journal of Mechanical Design},
  year = {2024},
  doi = {10.1115/1.4066425},
}
Figure