Abstract
This article presents a novel three-dimensional topology optimization framework developed for 3D spatial packaging of interconnected systems using a geometric projection method (GPM). The proposed gradient-based topology optimization method simultaneously optimizes the locations and orientations of system components (or devices) and lengths, diameters, and trajectories of interconnects to reduce the overall system volume within the prescribed 3D design domain. The optimization is subject to geometric and physics-based constraints dictated by various system specifications, suited for a wide range of transportation (aerospace or automotive), heating, ventilation, air-conditioning, and refrigeration, and other complex system applications. The system components and interconnects are represented using 3D parametric shapes such as cubes, cuboids, and cylinders. These objects are then projected onto a three-dimensional finite element mesh using the geometric projection method. Sensitivities are calculated for the objective function (bounding box volume) with various geometric and physics-based (thermal and hydraulic) constraints. Several case studies were performed with different component counts, interconnection topologies, and system boundary conditions and are presented to exhibit the capabilities of the proposed 3D multi-physics spatial packaging optimization framework.
1 Introduction
Effective design automation strategies are key to meeting the present and future needs for 3D physics-based systems involving spatial packaging problems. Systematic, flexible, and efficient design optimization methods are essential for achieving better system performance, compactness, and life-cycle cost, across a wide range of engineering industries. The optimal 3D spatial packaging of interconnected systems with physics interactions (or SPI2) consists of three individual sub-problems: component placement, interconnect routing, and physics-based performance evaluation [1–7]. Current practices treat these problems independently, relying heavily on packaging and routing design rules [8,9], design heuristics [10–12], and expert intuition [13,14], and solution of these design problems often relies upon time-intensive direct human effort. Solving the SPI2 problem requires knowledge of the system requirements for manufacturing, assembly, testing, operation, maintenance, and repair. Often, engineers start with an existing system design and adapt it to meet new requirements. This process is not only time-consuming but also usually leads to designs that are sub-optimal in terms of density, system performance, and reliability. The associated design space involves many interrelated geometric constraints and physics-based considerations, which are cognitively difficult to navigate. Several inefficiencies, including designer bias and human error, lead to inferior designs; these inefficiencies may increase as product requirements grow more complex and required packaging volumes shrink. Therefore, there exists a critical need for design automation methods that support optimal solution of the combined component placement, routing, and physics evaluation problem for systems of increased complexity, and that reduce overall solution expense. Figure 1 shows a 3D representation of one of the many possible solutions to a simple SPI2 packaging problem.

Three-dimensional design representation of a system with spatially interconnected components that shows one of the many possible solutions to a simple SPI2 packaging problem
Here are some of the specific challenges that indicate the complexity of the 3D spatial packaging problem from various angles: (1) Both 3D component placement and interconnect routing are NP (nondeterministic polynomial time)-hard problems. Therefore, as the scale and complexity of the system increases, the number of possible solutions explodes combinatorially, increasing decision-making cost significantly. The 3D topological space is vast and challenging to navigate as there can be infinite design options depending on the tuning parameters. Therefore, it is essential to have sampling strategies that can cover the design space thoroughly and efficiently, (2) the 3D spatial packaging problem is a highly nonlinear optimization problem that simultaneously addresses component placement, routing, and physics performance evaluation. Therefore, there is a greater possibility of encountering local solutions with continuous spatial or parameter tuning when compared to design optimization of individual problem elements, (3) one key challenge in using gradient-based solution methods is that changes in interconnect spatial topology may impact the lumped-parameter system models (such as fluid loops) in ways that either prevent simulation of certain designs or at a minimum introduced non-smoothness, (4) creating design representations that can support topology, geometry, and physical aspects of the 3D spatial packaging problem in a unified way is one of the most challenging aspects. Conventional methods address at most a pair of them while solving multi-physics optimization problems. Previous work exists where all three aspects are included but they are specific to their applications and do not generalize, and (5) finally, human expertise solving these packaging problems is useful but can lead to bias or errors when the scale and complexity of the decision-making jas multiple attributes are involved. All these make the spatial packaging problem extremely challenging to tackle with.
Three-dimensional component placement (or packing) [15–20] and 3D pipe routing or cable wiring [21–40] problems are individually NP-hard problems; solving the packing and routing problem combined with multi-physics interactions and couplings between system elements is thus especially challenging. The 3D component placement, interconnect routing, and physics performance evaluation sub-problems, however, should be solved synchronously to achieve system-optimal designs. A sequential effort, such as place-then-route or vice versa, cannot fully exploit design coupling between all sets of decisions. The challenges grow in significance as system compactness and performance requirements intensify. For example, commercial aircraft engines a few decades ago had larger core diameters compared to current designs; consequently, modern aircraft engines have a much smaller core surface area than prior designs on which are package comparable externals (e.g., wires, pipes ducts, sensors, and actuators).
The authors have made recent advancements in simultaneous 2D spatial component placement and routing (PR) of fluid-thermal systems. A two-stage design framework [41–44] was previously developed by the authors utilizing a 2D geometric projection-based topology optimization method [45] for solving SPI2 problems using a continuous design representation. The continuous representation enables the use of gradient-based methods to efficiently search the component packing and routing space. This methodology is centered on the use of simple geometric bars as basis elements to approximately model both the component geometry and routing paths. Two-dimensional bars have favorable geometric properties that can be exploited to represent both the packing and routing problem, and solve them simultaneously. This simple geometric representation was used in two stages to perform physics-based packing and routing for different performance objectives; the staged method ensures identification of a system-optimal solution. The success of this method has been demonstrated for simple 2D PR test cases in Refs. [41,46]. While the authors had success in the 2D design problem, the 3D problem is much more challenging due to the fundamentally different topological design space and many other features such as crossings between interconnects. This work focuses primarily on extending the prior 2D SPI2 design optimization capabilities to 3D systems by utilizing the 3D gradient-based geometric projection techniques [47].
In must be noted that the current scope of this proposed work only focuses on the demonstration of the 3D geometric projection method (GPM) applied through a gradient-based topology optimization (TO) approach. However, a more comprehensive design exploration process would involve efficient enumeration of distinct 3d layout topologies and optimizing each of them using a multi-start method with the proposed 3D GPM approach. This would help explore the design space more thoroughly and get a close system-optimal design solution through exploration of several local minima points. The authors have demonstrated such a comprehensive approach for 2d systems as discussed in Ref. [41], and the 3d topology enumeration algorithms were successfully demonstrated using spatial graphs representations in Ref. [48]. Future work would involve combining both the 3d topology enumeration and the 3D GPM topology optimization steps into a single-integrated design optimization problem.
1.1 Objectives and Contributions.
The main objective of this article is to demonstrate simultaneous multi-physics 3D component placement and interconnect routing optimization using gradient-based topology optimization methods. This work is a novel attempt to explore a geometric projection method for development of a 3D simultaneous packaging and routing algorithm with broad industrial applicability. The ultimate plan is to incorporate all kinds of geometric, topological, and physics-based constraints that support efficient optimization solutions. This initial work is focused on finding solutions that minimize the bounding box volume, including multi-physics capabilities such as thermal and hydraulic models, non-interference geometric constraints, physics boundary conditions, etc. This will pave way to develop a robust design optimization framework suitable for real world 3D SPI2 applications. Here is a list of the major contributions in this article:
We present a novel technique that supports simultaneous multi-physics 3D optimization of component placement and interconnect paths, whereas existing methods treat component layout design and interconnect routing separately (e.g., optimal routing with fixed layout).
Thermal and hydraulic physics-based performance objectives and constraints were incorporated into the optimization problem, subject to various boundary conditions and geometric constraints that prevent interference between components and interconnects. Both 1D lumped parameter and 2D finite element physics models are used within a single optimization problem to support physics-based evaluation.
We use the 3D geometric projection method discussed in Ref. [49], which is an alternative to the well-established SIMP (solid isotropic material with penalization) method [50] for parameterizing and solving the design optimization problem [4].
Design variables include component locations, orientations, and interconnect lengths, diameters, bend radius at joints, and piping trajectories. The physics-based constraints incorporated are maximum allowable device temperatures, maximum fluid flowrate, and pressure head loss within the fluid network.
We demonstrated the effectiveness of the proposed method via the solution of several component placement-routing test cases that utilize multi-physics-based simulations.
The remainder of this article is organized as follows. Section 3 discusses the 3D GPM and the simultaneous component packing and routing (PR)-based optimization problem formulation. The sensitivity analysis performed to attain the first-order derivatives of the objective and constraint functions utilized by the optimization algorithm is derived in Sec. 4. Results from numerical case studies using the proposed optimization method are presented in Sec. 5. Finally, the conclusions and future work items are discussed in Sec. 6.
2 Geometric and Physics-Based Modeling
2.1 Geometric Projection.
Material distribution topology optimization (MDTO) methods optimally distribute material within a prescribed design domain to achieve optimal system performance, for example, structural compliance. Significant advancements have been made in the development of density-based topology optimization methods where the design space is parameterized by a discretized material density field. Here the local material properties are defined as differentiable functions of material density. The advantages of these MDTO methods are: (1) the ability to circumvent re-meshing due to changes in topology and shape of the structure during optimization and (2) support implementation of efficient gradient-based nonlinear programming algorithms by allowing computation of design sensitivities for optimization. Two examples of density-based TO methods are: (1) SIMP technique [51], which generates a solid-void design by penalization of intermediate density regions within the domain during the optimization process; and (2) level set TO methods [52,53] where the structural design is represented implicitly via the zero level set of a function. Level set TO methods do not require penalization of intermediate material densities as the level set representation inherently induces a solid-void design almost everywhere.
The above-mentioned density-based optimization approaches such as the level set TO and the SIMP-based techniques are quite popular as they generate complex, organic shapes that can be fabricated using additive manufacturing processes to generate structures that closely resemble the optimal structural topologies. However, in applications such as the proposed SPI2 design research where the designer chooses to utilize stock materials such as tubes, square or circular plates, bars, etc., it becomes very challenging to directly use the TO results from aforementioned methods due to limitations with the manufacturing processes. In such cases, the designer spends a significant amount of time in post-processing to achieve a design that simultaneously captures the intent of the result from the TO method and can be manufactured within the available processes. This in turn might lead to addition of weight on the structure and degradation of the overall structural performance. An alternative to this is to use high-level explicit design representations such as the GPM that renders designs based on solid geometric components. Thus, the GPM provides several advantages over conventional TO methods to represent complex geometries easily and optimize the locations and orientations of these objects with respect to multi-physics and thereby enabling easy post-processing for additive manufacturing such solutions. The GPM method is further described below.

Figure shows (a) the device and bar elements used to represent the components and interconnect routing, respectively, and (b) the geometric projection and parameters of the device and bar elements
2.2 Temperature Model.
2.3 Lumped-Parameter Pipe Flow Model.
Flow is incompressible.
All components are connected in series with no branches.
There is no pump or turbine.
The inlet and outlet cross-sectional area of each pipe are the same.
Flowrate at the inlet is known.
Specific weight is non-changing.
Flow is turbulent everywhere.
3 Problem Formulation With the Three-Dimensional Geometric Projection Method
The derivative with respect to the other axes can be obtained similarly. In our design representation for this work, the rotation matrix is used to rotate the device about its center by any angle between 180 deg (clockwise) and −180 deg (anticlockwise).
For the three-dimensional packaging and routing design framework, the objective is to minimize the bounding box volume of the optimal layout under certain geometric constraints. The system elements are predominantly defined as components and interconnects. The constraints as shown in Fig. 4 in the 3D system are enumerated below.
Device-device constraint (gdd(x)): This constraint ensures that the devices defined within the system do not collide with one another.
Device-interconnect constraint (gds(x)): This constraint ensures that the devices and the interconnects do not intersect one another.
Interconnect-interconnect constraint (gss(x)): This constraint prevents the interconnects from intersecting one another.
Minimum bar length constraint (gl(x)): This constraint ensures that the optimal interconnect segment lengths are not reduced to a point (or zero length). However, there could be cases where zero-interconnect lengths are common and very useful. For example, they represent direct connection of two devices, such that the outlet of the source device is directly connected to the inlet of the target device. For example, an engine component bolted directly to engine block.

The 3D non-interference geometric constraints imposed between different system elements are shown in (a) between two devices (gdd), (b) between a device and a routing segment (gds), and (c) between two routing segments (gss). The mathematical constraint equations for each case are provided in Eqs. (38)–(40), respectively.

The 3D non-interference geometric constraints imposed between different system elements are shown in (a) between two devices (gdd), (b) between a device and a routing segment (gds), and (c) between two routing segments (gss). The mathematical constraint equations for each case are provided in Eqs. (38)–(40), respectively.
These constraints are provided in the two-dimensional design framework and are modified to be included in the three-dimensional framework. The system layout is shown in Fig. 5.
The angular constraints were initially used as an alternative to physics constraints as a strategy to reduce interconnect bend sharpness. While physics-based constraints can help drive interconnects toward smoother designs (e.g., to reduce pressure losses due to sharp bends), the underlying combined finite element analysis and lumped-parameter models can be computationally expensive to evaluate. For some early-stage studies it may be beneficial to solve simplified problems where physics does not drive smoothness, but is still assured through direct angular constraints.
4 Sensitivity Analysis
5 Results and Discussion
This section presents five numerical case studies that were performed to demonstrate the proposed 3D spatial layout optimization method described in Sec. 3. All reported numerical simulations were performed using a workstation with an Intel Xeon E5-2660 HDD CPU @ 2.00 GHz, 64 GB DDR4-2400 RAM, Windows 10 64-bit, and Matlab 2021a.
The design domain for all the case studies is chosen as a box of size 0.5 × 0.5 × 0.25 cubic units (which equals 6.25 × 10−2 cubic units). This 3D volume is the entire design space that is available for the components and the interconnect network to continuously move during the optimization process. For all case studies, the components are represented as cubes having sides of length 0.06 units. For every case study, the initial design layout has an initial bounding box volume, i.e., the tightest bounding box enclosing the system components and interconnect network evaluated using Eq. (37). For example, in case study 1, the initial design layout shown in Fig. 7(a) has an estimated bounding box volume of 2.379 × 10−2 cubic units, which corresponds to the total initial space occupied by the three components and their interconnect network. The bounding box volume minimization was considered as the optimization objective for all the case studies. The optimization design variables are component center location, orientation, interconnect thickness, length, interconnect elbow bend angle, and bend radius.

Case study 1: three-component system with geometric constraints—(a) initial layout (volume: 2.379 × 10−2 cubic units) and (b) final optimal layout (volume: 1.044 × 10−4 cubic units). We observe a decrease in total system volume by
5.1 Case Study 1: Simple Volume Minimization Using Geometric Constraints.
In this case study, we perform a simple bounding box volume minimization of a three-component (as shown in Fig. 7(a)) and an eight-component layout (as shown in Fig. 8(a)). Only the component rotation, and geometric constraints to ensure non-interference between the components and/or the interconnect network are considered in this study. The final optimal layouts for the three- the eight-component layouts are shown in Figs. 7(b) and 8(b), respectively. The initial and final bounding box volumes for both the layouts are depicted in Figs. 7 and 8. We observe a decrease in total volume by and for the three- and eight-component layouts, respectively. The corresponding 2D XY plane (or top) views of the initial and final optimal layouts in both the cases are shown in Fig. 9. The computational time taken for the three- and eight-component systems are 219 and 118 min, respectively. The eight-component system converged faster than the three-component system as more components and interconnects helped the system to attain a quicker solution before the constraints are violated as shown in Table 1.

Case study 1: eight-component system with geometric constraints—(a) initial layout (volume: 2.534 × 10−2 cubic units) and (b) final optimal layout (volume: 1.668 × 10−3 cubic units). We observe a decrease in total bounding box volume by

Case study 1: (a) three-component system initial layout in the XY plane and (b) its final optimal layout, (c) eight-component system initial layout in the XY plane and (b) its final optimal layout
Summary of case studies: objectives and constraints
Case study | Problem description | Constraints under consideration (including with or without boundary (wall) conditions) |
---|---|---|
1 | Simple bounding box volume minimization with rotation and geometric non-interference constraints | Component rotation, and geometric constraints with no boundary conditions (wall temperature or heat flux constraints) |
2 | Finding optimal system volume with three components using angular constraints (the angle subtended between the interconnect segments is constrained within the given bounds) | Angular and geometric non-interference constraints for component layout with no boundary conditions (wall temperature or heat flux constraints) |
3 | Finding optimal system volume with same critical temperatures for components and heat dissipation rates | Geometric non-interference constraints, component rotation, physics-based constraints, same component temperature limits, and heat dissipation rates with boundary conditions (wall temperature or heat flux constraints) |
4 | Finding optimal system volume with different critical temperatures and heat dissipation rates for components | Geometric non-interference constraints, component rotation, physics-based constraints, varying temperature limits, varying heat dissipation rates with boundary conditions (wall temperature or heat flux constraints) |
5 | Finding optimal volume layout with pressure head loss constraints | Geometric non-interference constraints, component rotation, physics-based constraints, temperature limits, heat dissipation rates, fluid pressure head loss limits with boundary conditions (wall temperature or heat flux constraints) |
Case study | Problem description | Constraints under consideration (including with or without boundary (wall) conditions) |
---|---|---|
1 | Simple bounding box volume minimization with rotation and geometric non-interference constraints | Component rotation, and geometric constraints with no boundary conditions (wall temperature or heat flux constraints) |
2 | Finding optimal system volume with three components using angular constraints (the angle subtended between the interconnect segments is constrained within the given bounds) | Angular and geometric non-interference constraints for component layout with no boundary conditions (wall temperature or heat flux constraints) |
3 | Finding optimal system volume with same critical temperatures for components and heat dissipation rates | Geometric non-interference constraints, component rotation, physics-based constraints, same component temperature limits, and heat dissipation rates with boundary conditions (wall temperature or heat flux constraints) |
4 | Finding optimal system volume with different critical temperatures and heat dissipation rates for components | Geometric non-interference constraints, component rotation, physics-based constraints, varying temperature limits, varying heat dissipation rates with boundary conditions (wall temperature or heat flux constraints) |
5 | Finding optimal volume layout with pressure head loss constraints | Geometric non-interference constraints, component rotation, physics-based constraints, temperature limits, heat dissipation rates, fluid pressure head loss limits with boundary conditions (wall temperature or heat flux constraints) |
5.2 Case Study 2: Optimal Layout With Three Components Using Angular Constraints.
The second case study consists of a three-component layout. The initial design layout is shown in Fig. 10. The initial bounding box volume, i.e., the tightest bounding box enclosing the system components and interconnects, evaluated using Eq. (37) is approximately 2.379 × 10−2 cubic units.

Case study 2: different views of the initial layout: (a) isometric view and (b) a view in the XY plane
The optimization algorithm was run for about 350 iterations. The final layout obtained is shown in Fig. 11. Due to the computational expense associated with running a 3D geometric projection scheme, the optimization was terminated when the objective value approximately plateaued as shown in Fig. 12. The final optimal bounding box volume is approximately 0.018 cubic units. Thus for this simple three-component layout optimization case, an almost reduction in the bounding box volume was achieved. There is not much reduction in the total volume here compared to case study 1 because of the angular constraints applied. This affects the overall bounding box volume as components cannot come closer than they otherwise would due to the restrictions caused by the angle subtended at the joints of their interconnects. Further, density improvement might be possible by increasing the number of segments representing an interconnect to help enrich the design space. It must be noted that the angular constraint might not work well for similar case studies but can be useful in complex case studies where system specifications need pipe bends at joints to be at fixed angles for design and manufacturing purposes. Furthermore, the increase in the bounding box volumes seen in this case study also motivated the inclusion of component rotational constraints and incorporating component orientation within the optimization framework presented here.

Case study 2: different views of the final layout: (a) isometric view and (b) a view in the XY plane
5.3 Case Study 3: Same Critical Temperatures and Heat Dissipation Rates for All Components.
Unlike earlier case studies, in this study we demonstrate the use of physics-based constraints, system boundary conditions, and geometric non-interference constraints. Figure 13 shows the 3D design domain containing a representative eight-component interconnected system with physics-based boundary conditions. We assume there is heat convection on the colder face (on the right at an ambient temperature, 0 °C) and a hotter face in the front of the layout (at a fixed temperature 100 °C). It is also assumed that all the other faces (boundaries) of the box are thermally insulated. The heat transfer coefficient across the colder face boundary is 35.80 W/m3.

Three-dimensional design domain with physics boundary conditions used for case studies 3, 4, and 5. It is assumed that the hotter face (or boundary) is fixed at , and the colder face has heat convection due to the ambient air at . The remaining boundaries of the design volume are assumed to be thermally isolated. The heat transfer coefficient for thermal conduction is 35.80 W/m3.

Three-dimensional design domain with physics boundary conditions used for case studies 3, 4, and 5. It is assumed that the hotter face (or boundary) is fixed at , and the colder face has heat convection due to the ambient air at . The remaining boundaries of the design volume are assumed to be thermally isolated. The heat transfer coefficient for thermal conduction is 35.80 W/m3.
In case study 3, all the components have the same critical temperature of 30 °C and a heat dissipation rate of 3000 W/m3. Figures 14 and 15 show initial and final optimal layouts of the three- and eight-component systems, respectively. Figures 14(b) and 15(b) help visualize how all the components in both the three- and eight-component systems and their interconnect networks have moved toward the colder boundary and away from the hotter boundary during optimization in the final layout. This is a clear evidence that the physics-based optimization is successful as it tries to satisfy all the geometric and temperature constraints during bounding box volume minimization. Total reductions in bounding box volume of and were realized for the three- and eight-component systems, respectively.

Case study 3: three-component system with same critical temperatures and heat dissipation rates for all components—(a) initial layout (volume: 2.379 × 10−2 cubic units), (b) final optimal layout (volume: 2.646 × 10−4 cubic units), and (c) closer view of the final layout. We observe an overall decrease in bounding box volume by

Case study 3: three-component system with same critical temperatures and heat dissipation rates for all components—(a) initial layout (volume: 2.379 × 10−2 cubic units), (b) final optimal layout (volume: 2.646 × 10−4 cubic units), and (c) closer view of the final layout. We observe an overall decrease in bounding box volume by

Case study 3: eight-component system with same critical temperatures and heat dissipation rates for all components—(a) initial layout (volume: 2.534 × 10−2 cubic units), (b) final optimal layout (volume: 3.802 × 10−3 cubic units), and (c) closer view of the final layout. We observe an overall decrease in bounding box volume by

Case study 3: eight-component system with same critical temperatures and heat dissipation rates for all components—(a) initial layout (volume: 2.534 × 10−2 cubic units), (b) final optimal layout (volume: 3.802 × 10−3 cubic units), and (c) closer view of the final layout. We observe an overall decrease in bounding box volume by
5.4 Case Study 4: Different Critical Temperatures and Heat Dissipation Rates for All Components.
In case study 4, all the components have different critical temperatures and heat dissipation rates as shown in Tables 2 and 3 for the three- and eight-component systems, respectively. Figures 16 and 17 show initial and final optimal layouts of the three- and eight-component systems, respectively. Figures 16(b) and 17(b) help visualize how the components with lower critical temperatures and higher heat dissipation rates in both the three- and eight-component systems have moved toward the colder boundary and away from the hotter boundary in the final layout. The final component temperatures for both the layouts are shown in Tables 2 and 3. It can be observed that all of them have satisfied their respective critical temperature limits. Total reductions in bounding box volumes of and were obtained for the three- and eight-component systems, respectively. The objective function convergence plots for this case study are provided in Fig. 18.

Case study 4: three-component system with different critical temperatures and heat dissipation rates—(a) initial layout (volume: 2.379 × 10−2 cubic units), (b) final optimal layout (volume: 4.897 × 10−4 cubic units), and (c) closer view of the final layout. Here we observe a reduction of the total bounding box volume by

Case study 4: three-component system with different critical temperatures and heat dissipation rates—(a) initial layout (volume: 2.379 × 10−2 cubic units), (b) final optimal layout (volume: 4.897 × 10−4 cubic units), and (c) closer view of the final layout. Here we observe a reduction of the total bounding box volume by

Case study 4: eight-component system with different critical temperatures and heat dissipation rates—(a) initial layout (volume: 2.534 × 10−2 cubic units), (b) final optimal layout (volume: 4.217 × 10−3 cubic units), and (c) closer view of the final layout. Here we observe a reduction of the total bounding box volume by

Case study 4: eight-component system with different critical temperatures and heat dissipation rates—(a) initial layout (volume: 2.534 × 10−2 cubic units), (b) final optimal layout (volume: 4.217 × 10−3 cubic units), and (c) closer view of the final layout. Here we observe a reduction of the total bounding box volume by

(a) Case study 4: three-component layout objective function convergence plot and (b) eight-component layout objective function convergence plot
Critical temperature constraints, heat dissipation rates, and final temperatures for components in case study 4A
Component (color) | Heat dissipation rates (W/m3) | Critical temperatures (°C) | Final temperature (°C) |
---|---|---|---|
1 (blue) | 6000 | 70 | 48.25 |
2 (cyan) | 3000 | 30 | 23.14 |
3 (magenta) | 2000 | 20 | 20 |
Component (color) | Heat dissipation rates (W/m3) | Critical temperatures (°C) | Final temperature (°C) |
---|---|---|---|
1 (blue) | 6000 | 70 | 48.25 |
2 (cyan) | 3000 | 30 | 23.14 |
3 (magenta) | 2000 | 20 | 20 |
Critical temperature constraints, heat dissipation rates, and final temperatures for components in case study 4B
Component (color) | Heat dissipation rates (W/m3) | Critical temperatures (°C) | Final temperature (°C) |
---|---|---|---|
1 (blue) | 5000 | 40 | 13.48 |
2 (light blue) | 4000 | 50 | 36.7 |
3 (green) | 3000 | 30 | 15.38 |
4 (cyan) | 3000 | 30 | 30 |
5 (magenta) | 2000 | 20 | 17.979 |
6 (yellow) | 5000 | 50 | 42.507 |
7 (black) | 3000 | 30 | 28.617 |
8 (orange) | 1000 | 20 | 17.034 |
Component (color) | Heat dissipation rates (W/m3) | Critical temperatures (°C) | Final temperature (°C) |
---|---|---|---|
1 (blue) | 5000 | 40 | 13.48 |
2 (light blue) | 4000 | 50 | 36.7 |
3 (green) | 3000 | 30 | 15.38 |
4 (cyan) | 3000 | 30 | 30 |
5 (magenta) | 2000 | 20 | 17.979 |
6 (yellow) | 5000 | 50 | 42.507 |
7 (black) | 3000 | 30 | 28.617 |
8 (orange) | 1000 | 20 | 17.034 |
5.5 Case Study 5: Pressure Head Loss Constraints.
In this last case study, we have included the pressure head loss as an additional physics-based constraint during optimization. Once again, the three- and eight-component systems were considered here as shown in Figs. 19 and 20. The same critical component temperature constraints and heat dissipation rates are maintained as indicated in case study 4. The maximum head loss constraints are 20 m and 35 m for the three- and the eight-component systems, respectively. Sharper interconnect segment bends create greater pressure head loss in the pipes. Figures 19(b) and 20(b) show the final layouts after optimization. To satisfy the head loss constraint, the interconnects have smoother bends compared to the previous case studies. This suggests that the geometry at the interconnect joints is the dominant contributor to the head loss.

Case study 5: (a) initial layout (volume: 2.379 × 10−2 cubic units) and (b) final optimal layout (volume: 1.558 × 10−3 cubic units) with a 2.5 m active head loss constraint. Here we observe a reduction of the total bounding box volume by , and the pressure head loss is an active constraint at 20 m.

Case study 5: (a) initial layout (volume: 2.379 × 10−2 cubic units) and (b) final optimal layout (volume: 1.558 × 10−3 cubic units) with a 2.5 m active head loss constraint. Here we observe a reduction of the total bounding box volume by , and the pressure head loss is an active constraint at 20 m.

Case study 5: (a) initial layout (volume: 2.534 × 10−2 cubic units) and (b) final optimal layout (volume: 3.612 × 10−3 cubic units) with a 5 m active head loss constraint. Here we observe a reduction of the total bounding box volume by , and the pressure head loss is an active constraint at 35 m.
Total reductions in bounding box volumes of and were obtained for the three- and eight-component systems, respectively. To satisfy device temperature constraints, some of the routing interconnects touch the convection boundary. This conducts heat from the devices through the routing to the boundary where it can be dissipated. The optimization finds a balance between smooth bends and reducing pipe length to reduce head loss in a way that is best for system performance. However, the head loss constraints in this case study are active as there exists a trade-off between pressure head loss and volume minimization. For example, if the volume of the system has to decrease, the components should come closer which then requires that the interconnect elbows are sharper. Hence, the minimum bounding box volume is obtained when the head loss constraint remains active.
6 Conclusions and Future Work
A novel method for simultaneously optimizing the 3D placement of components (or devices) and their corresponding interconnect routing paths was presented in this work. The aim is to improve the spatial packaging density and physics-based system performance of a given 3D interconnected engineering system by minimizing the overall bounding box volume, satisfying spatial, physics-based (thermal, hydraulic, etc.), component orientation, and other geometric constraints that prevent interference between devices and/or the interconnect paths. A geometric projection of the various 3D system elements, as well as required sensitivity calculations, were performed that support optimization based on a 3D finite element mesh. A set of design variables was developed that links directly to the geometric functions of the system that represent the volume objective and the different types of constraints.
The proposed method was demonstrated using several case studies to perform simultaneous component packing and routing optimization with spatial, geometric, and physics-based constraints. In addition, component rotation was also demonstrated during the continuous gradient-based optimization process. This improved the spatial packaging density as components could reorient themselves during the optimization to attain more compact solutions. In case study 1, a simple bounding box volume minimization problem was solved with three-component and eight-component systems. Only the component rotation and the geometric constraints for non-interference between components and interconnects were considered for this case study. Case study 2 was performed to minimize the overall volume of a three-component system subject to geometric, rotational, and angular constraints as well. The angular constraints ensure that a specified minimum angle is subtended at interconnect pipe joints, and could be useful for spatially packaging complex 3D systems that must satisfy specific design and manufacturability constraints for pipe bends. Case studies 3, 4, and 5 include physics-based constraints and boundary conditions as part of the optimization problem. In case study 3, all the components have same critical temperatures and heat dissipation rates. In case study 4, each component has a unique critical temperature constraint and a heat dissipation rate. From this case study, it is evident that components having lower critical temperatures and higher heat dissipation rates move toward the colder boundary to satisfy the temperature constraints during volume minimization. The final case study included pressure head loss as an additional physics constraint along with temperature, and geometric constraints. Smoother pipe bends were obtained in the final optimal layouts from this study as sharper bends cause greater head losses. Hence, there exists a trade-off between system volume and pressure head loss.
Some major advantages of the proposed design optimization method are (1) the GPM method is an explicit design representation approach that helps capture the complex geometry of the components and the interconnects, (2) efficient continuous parameterization with a gradient-based optimization approach can be used with this representation when projected on a finite element mesh, (3) this approach allows integration of multi-physics model into the optimization problem formulation, and (4) the primitive geometric shapes of any 3D engineering system can be represented using this approach. Some of the limitations of the proposed approach are (1) it would be interesting how the proposed method would scale with more than 30 or 40 components and 100s of interconnects and (2) the proposed method is based on a gradient-based optimization method and hence only can perform local search. Hence, there is a need for a method that can help explore multiple discrete topologies and a multi-start approach would be more ideal for a comprehensive search.
In future work, we plan to incorporate additional physics models such as thermoelastic, structural, and electrical models for describing system performance objectives and constraints. The proposed design framework does not consider irregular devices shapes. New design representations and shape descriptors for modeling complex device geometries will be investigated as part of future work. Furthermore, we plan to demonstrate our design method on larger-scale industry-relevant 3D applications that include multi-port components with arbitrarily shaped component geometries, interconnects, and 3D spatial topologies. These features in addition to more comprehensive physics-based optimization approaches will be useful in designing a realistic 3D component packing and routing design optimization framework in the future. The method presented here is expected to lead to faster development of holistic SPI2 design automation methods which are robust, efficient, and provide better design solutions than purely manual or heuristic design methods.
Acknowledgment
This work was supported by the National Science Foundation Engineering Research Center for Power Optimization of Electro-thermal Systems (POETS) with cooperative agreement EEC-1449548. The authors are also thankful to industry partner liaisons from CU Aerospace Ltd., PC Krause and Associates, and Ford Motor Company for their invaluable technical feedback on this work.
Conflict of Interest
There are no conflicts of interest.
Data Availability Statement
The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.
Nomenclature
- m =
meters
- r =
maximum vertex distance from the device center location
- w =
width (or thickness) of the interconnect segment
- x =
design variable vector
- z =
reduced design variable vector
- N =
shape functions of the element where the center lies
- T =
nodal temperatures from the finite element formulation (KT = P)
- V =
bounding box volume
- R =
rotation matrix
- T =
temperature vector (state variables)
- dij =
minimum distance between the device center location and line segment representing the routing bar
- fi =
friction factor
- li =
length of a component
- nd =
total number of devices (or components)
- ns =
total number of interconnect segments
- xc =
X-coordinate of the component center
- yc =
Y-coordinate of the component center
- zc =
Z-coordinate of the component center
- bi =
vector pointing from the reference point (or center) to the vertices of a component
- cd =
component reference (or center) point
- gheadloss =
head loss constraint
- gtemp =
device temperature constraint
- gθ =
angle constraint
- pi =
port locations vector
- x0 =
interconnect start point
- xf =
interconnect end point
- Ai =
area of cross section of the pipe
- Ki =
loss coefficient due to surface friction
- =
temperature at the center of the devices
- Tmax =
specified critical temperature
- z' =
expanded design variable vector
- mm =
millimeters
- f(x, T) =
objective function
- g(x, T) =
constraint function
- gphys(x, T) =
physics-based constraints—they depend both on design and on solutions to physics models
- gdd(x) =
non-interference constraints between two components
- gds(x) =
non-interference constraints between one routing segment and one component
- gss(x) =
non-interference constraints between two routing segments
- gl(x) =
minimum bar length constraint
- EV =
electric vehicle
- GPM =
geometric projection method
- HVACR =
heating, ventilation, air-conditioning, and refrigeration
- MMA =
method of moving asymptotes
- PR =
packing and routing
- SPI2 =
spatial packaging of interconnected systems with physical interactions
- W/m3 =
Watt per cubic meters—used for component heat dissipation rate
- K(x) =
global thermal stiffness matrix
- P (x) =
global load vector
- °C =
degree celsius—temperature units
- α =
angle at the elbow bends between interconnect segments
- θi =
angle of rotation along ith axis direction
- ρw =
weight density