# CEC’2022 Competition on Heat Pipe-Constrained Component Layout Optimization

The performance of electronic devices is significantly influenced by the placement of its components. For example, an increasing number of electronic components are required to be installed within a small size of device space (such as print-circuit boards), thus easily causing a severe heat concentration. An irreversible device damage would probably happen under over-high temperature if their positions are not carefully arranged. Therefore, it raises an important practical engineering-driven optimization problem, the component layout optimization (CLO) problem. The objective of this problem is to maximize the thermal performance by optimizing the layout scheme of components while satisfying some necessary design constraints (including non-overlapping constraints, system centroid constraint, etc.). To obtain the global optimal component layout design, population-based evolutionary algorithms show their unique advantages over than gradient-based algorithms. However, the complexity in such problems still poses great challenges to common evolutionary optimization algorithms.

Firstly, taking position coordinates as design variables, the CLO problem can easily become a high-dimensional optimization problem with the number of components increases. Secondly, a huge number of complex constraints should be strictly satisfied. One type of basic geometric constraint is the spatial non-overlapping constraint between components or between the container and components, which requires that no overlap exists between any two objects. Thirdly, there may exist two or more distinct component layout design schemes corresponding to the same or similar performance, which makes the CLO a multimodal optimization problem. All these points make the CLO problem intractable in efficiently and effectively searching optimal or suboptimal layout solutions using common evolutionary algorithms, which highly hampers their industrial applications. The aim of this competition is to promote the research on continuous constrained single-objective optimization problems and hence solve complicated real-world application problems.

In this competition, one typical layout design scenario, the Heat pipe-constrained Component Layout Optimization (HCLO) problem, is carefully selected and simplified from the real-world engineering layout application. As shown in Figure 1, the largest red rectangle denotes the layout container, determining the available layout domain boundary. Green rectangles mean heat pipes where the generated heat by components can be directly dissipated outside by heat conduction. Blue shaded boxes represent the electronic components that are required to be placed right above the heat pipes and within the layout domain. The design objective is determined to minimize the maximal real heat dissipation power of heat pipes, that is, to improve the heat dissipating uniformity among all heat pipes. Therefore, the HCLO problem is a continuous constrained single-objective optimization problem. Note that the source code for calculating the relevant performance indicators will be provided in two programming languages (MATLAB and Python). Researchers can choose to use either of them for convenience.

Figure 1. The illustration of Heat pipe-constrained Component Layout Optimization (HCLO) problem

In this competition, there are five HCLO problems with different optimization scales under different problem settings: four public HCLO problems and one private HCLO problem. Four of them are public, which will be totally uncovered to help participants to better understand this problem. The last one is private with the same problem type, where the evaluation code involved will not be released. Participants are encouraged to develop the algorithm to solve this type of optimization problem, not just specific one of them. Participants may propose a new optimization algorithm or utilize a hybrid form of previously proposed algorithms. However, it must be restricted in the field of evolutionary computing. Besides, using commercial optimization software is not allowed. Participants are required to submit their own source codes along with a brief description of the optimization algorithm, as well as a brief code instruction. The performance evaluation of your proposed algorithm in all five HCLO problems will be carried out by organizers to guarantee its fairness. With the same computational budget, the best layout solution of each problem obtained by randomly running your algorithm five times will be compared directly. Note that for public HCLO problems, participants can submit their obtained best optimization solution as supplementary files. But these results will only be taken as reference. To fairly assess the performance of compared algorithms, we will directly evaluate the objective value of the best layout design when all constraints are satisfied. If no feasible solution is returned when the termination condition is satisfied, their constraint violations will be compared. The algorithm with less constraint violation will be better.

A brief problem description:

As shown in Figure 1, the design variables of this problem are their two-dimensional position coordinates. In order to meet the requirement of heat dissipation and system mass characteristic, three type of design constraints are established. One is the basic geometric non-overlapping constraint, which requires that any two components cannot overlap and any component cannot protrude from the layout domain. The second constraint is that the system centroid should be controlled below a specific absolute error with the expected centroid. The third one is involved in the heat dissipation process, which requires that components must overlap with heat pipes. Only if components intersect with heat pipes, the generated heat can be transferred outside, avoiding heat concentration. Note that for each heat pipe, there is a maximal heat-dissipating power capacity (denoted by $P_{max}^{HP}$). Therefore, the summation of heat-generating power of components that are placed above each heat pipe should not exceed its maximal capacity. When one component occupies two heat pipes simultaneously, it is assumed that its thermal power is equally conducted to two heat pipes. It is desired that the heat distribution over this layout area should be uniform. For simplification, the design objective is determined to minimize the maximal real heat dissipation power among all heat pipes. To summarize, the HCLO problem is a continuous constrained single-objective optimization problem. Four public HCLO problems and one private problem are introduced briefly below.

• Problem 1 (Public): 6 components need to be placed properly within one layout domain, involving 12 design variables.
• Problem 2 (Public): 15 components need to be placed properly within one layout domain, involving 30 design variables.
• Problem 3 (Public): 40 components need to be placed properly within two layout domains, involving 80 design variables.
• Problem 4 (Public): 60 components need to be placed properly within four layout domains, involving 120 design variables.
• Problem 5 (Private): 119 components need to be placed properly within four layout domains, involving 169 design variables. Some of them are fixed during optimization.

Problem Statements and codes:

The detailed problem statements for this competition can be downloaded here. Codes for evaluating four public problems are supplied with two versions (MATLAB and Python).

Rules:

• Participants will propose and implement metaheuristic algorithms (e.g., evolutionary algorithms, swarm intelligence, estimation of distribution algorithm, etc.) to solve complex problems in the layout design domain.
• The code for evaluating objective and constraints is strictly not allowed to be changed.
• A maximum number of 10,000 function evaluations is allowed.
• Only random seed initial solutions are allowed in this competition. Heuristics and special tweaks for initial solutions are not accepted.

Results Submission:

Please send your submission file (a zip file containing source codes) directly to Dr XXX ().

Organizers:

Wen Yao, Defense Innovation Institute, Chinese Academy of Military Science, China

Weien Zhou, Defense Innovation Institute, Chinese Academy of Military Science, China

Xianqi Chen, College of Aerospace Science and Engineering, National University of Defense Technology, China