Phase 5: Thermal CFD — FEA for Power Electronics

5.1 When to Use Icepak vs. Mechanical Thermal

Mechanical Steady-State Thermal solves conduction with user-specified convection coefficients at surfaces. It is appropriate when the dominant heat path is conduction, and when convection can be represented as a known, uniform coefficient at specific surfaces — a value from a correlation or measured data.

Icepak solves the Navier-Stokes equations coupled to the energy equation. It computes the convection coefficient rather than requiring it as an input. It is necessary when airflow path, velocity distribution, or enclosure geometry determines which components are adequately cooled.

The decision test: does the answer depend on how much airflow reaches this specific component? If yes — if the thermal result is sensitive to where air flows and at what velocity — use Icepak. If the problem is "how well does heat conduct from the die through the board to a heatsink," and convection can be approximated by a single coefficient, use Mechanical thermal. It is faster and adequate for conduction-dominated problems.

Practical example — battery backup enclosure

Two questions about the same enclosure, requiring different tools. Question 1: “What is FET junction temperature given a known heatsink surface temperature?” The heatsink temperature is measured from hardware. The heat path is FET die → die attach → copper baseplate → TIM → heatsink — a conduction problem with a known boundary condition. Mechanical Steady-State Thermal is correct: apply the measured heatsink temperature, assign conductivities, compute T_junction. Solve time: 4 minutes.

Question 2: “Will the fans provide adequate cooling with the enclosure fully assembled at 40°C ambient?” The answer depends on how air actually flows inside the enclosure — whether the heatsink, PCBs, and battery modules create blockages that reduce airflow at the FETs. Mechanical cannot answer this. Icepak is required: model the fans, vents, and internal obstacles, and compute local air velocity at the heatsink inlet. Solve time: 45 minutes.

5.2 Icepak Model Construction

Ansys Learning Hub: Icepak in AEDT: Getting Started ↗ · 10 hr · E-learning

Icepak uses a library of simplified geometry objects for most model elements:

Cabinet: defines the enclosure boundaries and sets the control volume for the CFD solve. Size it to contain the enclosure with a small clearance margin.
PCBs: flat plate objects with specified in-plane and through-thickness thermal conductivity. FR4 conductivity is highly anisotropic: much higher in-plane than through-thickness.
Components: block or cylinder objects representing heat-generating devices. Assign power dissipation and, where available, junction-to-case and junction-to-board thermal resistance.
Fans: defined by a performance curve (static pressure vs. volumetric flow rate at the operating air temperature).
Vents: openings defined by their free-area fraction (ratio of open area to total face area).
Heatsinks: available as parametric objects with defined fin geometry (height, spacing, thickness).

The power budget is the most important input. Verify power values against the system power budget before running. A 50% error in component power produces approximately a 50% error in junction temperature in a conduction-dominated thermal path — treat this as an order-of-magnitude estimate for convection-dominated enclosures. Use the object library for standard components and import simplified SpaceClaim geometry only for custom or critical components.

An inverter with six power FETs and one transformer, for example, uses block objects for all six FETs (standard packages with datasheet Θ_JC values), a parametric heatsink object, and fan objects with the corrected operating-temperature curve — the entire model constructed in under an hour without any imported CAD geometry.

5.3 CFD vs. Compact Thermal Models

A full CFD model of every component — resolving fins, die, package body, and leads — is expensive and usually unnecessary. Compact thermal models replace a component with a simplified resistance network that characterizes its thermal behavior without modeling the internal geometry.

The JEDEC two-resistor model (JESD51-2) defines two thermal resistances: Θ_JC (junction-to-case) and Θ_JB (junction-to-board). These are tabulated in component datasheets. ANSYS Icepak includes a library of standard packages with JEDEC data.

Use compact models when the component's internal thermal path is well-characterized and the analysis is evaluating system-level temperature, not the component's internal design. Use detailed CFD with imported geometry when the internal heat path is the design variable — for example, when evaluating a new heatsink geometry, a new die attach material, or a new package design.

Example — inverter FET array: JEDEC compact model vs. detailed CFD geometry

Six TO-247 power FETs on a heatsink. Analysis goal: verify FET junction temperatures below 125°C at rated power (50 W per FET, 300 W total).

Compact model (JEDEC two-resistor):

Each FET: Θ_JC = 0.4 K/W, Θ_JB = 5 K/W from datasheet
In Icepak: each FET is a block object at its footprint location; power = 50 W; thermal resistances from datasheet
Solve time: 8 minutes
T_junction,max = 101°C (at the two FETs nearest the transformer, where air temperature is highest after passing over upstream heat sources)

Detailed geometry CFD (imported package body):

FET body: die (5 mm × 5 mm), die attach, copper clip, mold compound modeled explicitly from component cross-section
Solve time: 47 minutes
T_junction,max = 104°C at the same location (3°C higher than compact model)

Conclusion: 3°C difference (3%) between approaches. For a margin check against a 125°C limit with 24°C of margin, compact models are adequate — the model error is far smaller than the design margin. The 47-minute detailed solve is not justified when the question is “are we below the limit?” Detailed geometry would be warranted if the question were “what is the minimum TIM conductivity before the limit is exceeded?” — where the 3°C uncertainty is material to the decision.

5.4 Forced Convection: Fans, Vents, and Airflow Path

Fan curve: enter the static pressure vs. volumetric flow rate curve at the operating air temperature inside the enclosure. Datasheet fan curves are at 25°C ambient. Hot air at 60–80°C is less dense and has different viscosity — the fan delivers less flow at the same static pressure. Correct the curve to the expected operating temperature.

Vent free-area fraction: the effective flow area is the physical area multiplied by the free-area fraction. A louvered vent at 40% open area has 40% of the flow capacity of the same opening fully open. This value is in the vent manufacturer's data. Setting it to 1.0 for a louvered vent overestimates flow by 2.5×.

Airflow path design: identify recirculation zones — regions where hot exhaust air re-enters the intake path — which reduce effective cooling. Identify stagnant zones — regions where airflow velocity approaches zero — which correspond to poorly cooled areas. Both appear clearly in Icepak's velocity vector plots post-solve.

Example — cast-Al heatsink: fan curve correction and recirculation identification

A forced-air heatsink: 120 mm × 80 mm base, 35 mm fins, 5 mm pitch. Two 60 mm fans. Operating inlet air: 60°C. Total dissipation: 150 W.

Fan curve correction: datasheet curve is at 25°C. At 60°C, air density is 10% lower — the fans deliver less mass flow at the same pressure. Correction factor: (273 + 60)/(273 + 25) = 1.117 for temperature, reducing mass flow by approximately 10%. Enter the corrected curve in Icepak. If the uncorrected 25°C curve is used, Icepak overestimates airflow and underpredicts temperatures by 5–8°C.
Vent free-area fraction: outlet grille is 40% open (louvered). Enter 0.40 in the Vent object. Setting this to 1.0 predicts 2.5× the actual exit velocity and dramatically underestimates the pressure drop at the outlet — the fan operates at a much lower pressure point on the curve and delivers much more flow than is physically present.
Velocity vector check after first solve: plot velocity vectors in the fin channel plane (cut plane parallel to airflow, at mid-height of the fins). Uniform flow entering all fin channels is expected. In this case: a recirculation zone is visible behind the transformer body mounted downstream of the heatsink. Hot exhaust air folds back and re-enters the first two fin channels, raising their inlet air temperature by 8°C.
Fix: add a 15 mm deflector plate between the transformer and the heatsink inlet. Re-solve: the inlet temperature elevation at the first two fin channels drops from 8°C to 2°C. FET case temperature drops by 6°C.

The recirculation zone was not visible from the temperature contour plot alone — it appeared only as a localised hotspot on two fins. The velocity vector plot identified the mechanism directly.

5.5 Parametric Heatsink Study

Ansys Learning Hub: Icepak in AEDT: Advanced ↗ · 10 hr · E-learning

SpaceClaim geometry parameters (Design → Parameters) allow heatsink fin dimensions to be defined as named variables that propagate into Workbench as design parameters. A parameter sweep runs each geometry combination as a complete Icepak solve.

Workflow

In SpaceClaim, define fin height, spacing, and thickness as named parameters.
Link the SpaceClaim component to Icepak in the Workbench schematic.
Open the Parameter Set view in Workbench (double-click the Parameter Set bar).
Define the parameter sweep table: combinations of fin dimensions to evaluate.
Run all design points. Workbench solves each as an independent Icepak run.
Review the Design Points table: junction temperature as a function of geometry.

Use a coarser mesh for the parametric sweep than for a final verification run. The goal of the sweep is to identify the optimal geometry region, not to compute a precise result at every point. Once the optimal region is identified, run one fine-mesh confirmation at the optimum.

Practical example — cast-Al heatsink parametric study

Parameter sweep: fin height (30, 40, 50, 60 mm) × fin spacing (3, 5, 7, 10 mm) = 16 design points. Coarse mesh, 2 mm global size: 8 minutes per point, 2 hours 8 minutes total. Key result: minimum T_junction is not at the tallest fins with the narrowest spacing. Narrowing from 5 mm to 3 mm increases pressure drop to the point where the fans deliver 30% less flow, and T_junction rises despite the larger surface area. The optimum is at 40 mm height / 5 mm spacing.

Fine-mesh verification at the optimum: T_junction = 98°C vs 103°C from the coarse sweep. The 5°C coarse-mesh error is acceptable for locating the optimum but not for reporting the final result. Run the fine-mesh solve only at the confirmed optimum — not at all 16 design points.

5.6 Correlating Simulation to Hardware

Hardware measurements provide the check on the simulation model. Plan measurements before running the simulation, not after — knowing what will be measured shapes which model outputs to prioritize.

Thermocouples at component case: measure T_case, not T_junction. Relate them: T_junction = T_case + (P × Θ_JC). Thermocouple contact resistance makes surface readings typically 2–5°C lower than actual surface temperature; account for this in the comparison.

Acceptable correlation: simulation within ±5–10°C of measured case temperature is typical for a well-constructed model. Larger errors trace to: incorrect or missing power assignments, incorrect or missing TIM conductivity, incorrect fan curve, or an unmodeled heat path.

Systematic overprediction: usually a missing conduction path or an overestimated power value. Systematic underprediction: usually a missing thermal resistance (air gap at a component interface, dry solder joint, incorrect TIM conductivity).

Key points

Plan measurement locations before the hardware build, not after. Thermocouple placement that was not planned cannot be added after assembly.
Junction temperature cannot be measured directly. It is estimated from case temperature plus Θ_JC × power. Report the basis for the junction temperature estimate alongside the measurement.
A model within 10°C of hardware is well-correlated for an electronics thermal model. Agreement better than 5°C does not guarantee model fidelity — verify that each major input uncertainty is small. Compensating errors (overestimated component power offset by underestimated thermal resistance) can produce artificially close results.

Example — inverter assembly: planning measurement locations and resolving a correlation failure

Pre-test plan for a 300 W inverter PCB assembly. Thermocouple placement decided before hardware build (simulation identifies the likely hotspot locations):

TC1: FET case surface, hottest FET from Icepak result at rated power
TC2: transformer outer surface, top winding
TC3: heatsink base plate, near FET mounting footprint
TC4: inlet air temperature (reference)

Post-test correlation (all values at rated power, 40°C ambient):

TC3 (heatsink): Sim 52°C, Measured 54°C, Δ = +2°C — well-correlated
TC1 (FET case): Sim 74°C, Measured 81°C, Δ = +7°C — within acceptable range
TC2 (transformer): Sim 68°C, Measured 88°C, Δ = +20°C — significant underprediction

Diagnosis of TC2 error (20°C): the simulation used the published transformer core loss from the datasheet (3.2 W at nominal flux density). The actual transformer dissipation at full load, measured by inlet-outlet calorimetry, is 7.1 W. The core loss input was based on a flux density that differed from the operating point — the transformer is running at higher flux than the datasheet characterisation condition. Correcting the power assignment to 7.1 W brings the transformer prediction to 89°C (1°C from measured).

This is the most common root cause of large correlation errors: incorrect power input, not a mesh or geometry problem. Verify component power before investigating any other source of error.

5.7 Superposition for Complex Assemblies

Ansys Learning Hub: Electro-Thermal and Thermo-Mechanical Analysis using IcepakFEA ↗ · 8 hr · E-learning

When a complete assembly is too expensive to simulate in full, temperature-rise (T-rise) results from sub-assembly simulations can be combined by superposition to estimate system-level temperatures.

The method: simulate sub-assemblies independently to determine T-rise per unit of dissipated power (K/W) for each component. For a change in one component's power: ΔT = ΔP × (T-rise per watt). Add to the baseline temperature.

Validity conditions: the system must be thermally linear (material properties must not vary significantly with temperature over the operating range — typically valid below 100°C for most electronics materials), and adding or changing a component must not significantly alter the airflow distribution. If the modified component is a large obstacle in the airflow path, the original T-rise data was computed under different flow conditions and superposition fails.

Example — FET array and transformer sub-models: computing T-rise by superposition

A power supply: 200 W heatsink-mounted FET array (six FETs, 33 W each) and a 12 W transformer. Full assembly Icepak solve: 90 minutes. Goal: estimate T_junction for design trade studies in 22 minutes using superposition.

All sub-models use 0°C ambient boundary (T-rise model): this lets T-rise results be added directly to any ambient temperature, without re-running the simulation when ambient changes.

Sub-model 1 — FET array only (transformer removed):

FETs at 33 W each, fans modeled, 0°C ambient
ΔT at FET case above ambient: 31 K (per FET, at rated power)
Solve time: 12 minutes

Sub-model 2 — transformer only (FETs removed):

Transformer at 12 W, fans modeled identically, 0°C ambient
ΔT at transformer surface: 38 K above ambient
Cross-coupling: transformer heat raises FET inlet air by +3°C (computed by adding a 12 W dummy source at the transformer location in sub-model 1)
Solve time: 10 minutes

Superposition result at T_amb = 40°C:

T_{FET case} = 40 + 31 + 3 = 74°C
T_junction = T_case + P × Θ_JC = 74 + 33 × 0.4 = 87°C

Full assembly verification: T_junction = 89°C (2°C from superposition). Acceptable for a design trade study. Superposition is valid here because the transformer (20% of the heatsink cross-section) does not significantly redirect airflow to the heatsink. If the transformer were larger, the airflow redistribution assumption would need verification.

5.8 Engineering Calculations for Model Setup

Two hand calculations belong before every Icepak run: an estimate of the expected mass flow rate through the enclosure, and a comparison of convective to radiative heat flux. Both take a few minutes and prevent common setup errors that produce plausible but wrong results.

Mass flow rate through the enclosure

The expected mass flow rate sets the scale for the thermal result. If the simulation predicts temperatures that differ from this estimate by a large factor, a model input is wrong before post-processing begins.

Volumetric flow rate at the system operating point:

Estimate total system pressure drop ΔP across the enclosure. For an electronics box with reasonable vent sizing, 1–5 Pa is a typical range for low-velocity forced air. Higher if vents are small or heavily louvered.
Read the fan curve at that ΔP to find Q (m³/s or CFM).
Correct Q to operating air temperature T_in: Q_hot ≈ Q_25°C × (T_in + 273) / 298. Air at 60°C is about 10% less dense than at 25°C; the fan delivers lower mass flow at the same speed.
Compute mass flow rate: &?; = ρ × Q, where ρ ≈ 1.0 kg/m³ at 60°C. (At 25°C, ρ = 1.2 kg/m³.)
Check average vent velocity: v = Q / A_eff, where A_eff = physical vent area × free-area fraction. If v exceeds 5–7 m/s at a supply vent, the vent is undersized and is the dominant pressure drop in the system.

Convection vs. radiation: when to model radiation

Icepak can include surface-to-surface radiation, but it substantially increases solve time. Determine whether it is significant before turning it on.

At a surface at temperature T_s in an enclosure at T_amb:

Radiative flux: q_rad = ε σ (T_s&sup4; − T_amb&sup4;), where σ = 5.67×10⁻⁸ W/m²K&sup4;
Convective flux: q_conv = h × (T_s − T_amb)

Worked example at typical conditions: T_s = 80°C (353 K), T_amb = 25°C (298 K), ε = 0.9 (PCB or component surface):

q_rad = 0.9 × 5.67×10⁻⁸ × (353⁴ − 298⁴) ≈ 387 W/m²
q_conv, natural convection (h = 5 W/m²K): 5 × 55 = 275 W/m²
q_conv, low forced convection (h = 15 W/m²K): 15 × 55 = 825 W/m²
q_conv, high forced convection (h = 50 W/m²K): 50 × 55 = 2750 W/m²

Conclusions:

Natural convection: radiation is larger than convection. Include radiation. Neglecting it systematically underpredicts temperature.
Low forced convection: radiation is ~47% of convection. Still significant — include it for accurate results.
High forced convection: radiation is ~14% of convection. Usually negligible for engineering accuracy. Safe to omit for most models.

Thermal resistance pre-check

Before running Icepak, estimate the dominant conduction resistances. For a planar layer (TIM, solder, PCB):

R_θ = L / (k × A)

where L is thickness, k is thermal conductivity, and A is contact area. The temperature drop across that layer at power P is ΔT = P × R_θ.

Use this to predict where the largest temperature drops will occur and whether Icepak's result is in the right ballpark. If Icepak shows a 40°C drop across a TIM layer but the hand calculation gives 5°C, check the conductivity assignment and layer thickness in the model before investigating further.

Key points

Estimate mass flow rate before running. If the simulation-implied flow rate disagrees by more than a factor of 2 with the hand calculation, a fan curve or vent free-area input is wrong.
Natural convection and lightly forced enclosures: include radiation. At 80°C surface temperature, radiation exceeds natural convection. Omitting it is a systematic error, not a conservative simplification.
High forced convection: radiation is secondary. Safe to omit and simplify the solve.
Pre-calculate expected ΔT across each major thermal layer before post-processing. Use R_θ = L/(k×A) as a sanity check on the simulation output.

Practical example — heatsink assembly pre-check

Before opening Icepak for a 300 W forced-air enclosure: fan rated 45 CFM at 0 Pa, 0 CFM at 50 Pa. Estimated system pressure drop: 2 Pa. Fan operating point: ~42 CFM (19.8 L/s). At 60°C inlet: ˙m ≈ 1.0 kg/m³ × 19.8×10⁻³ m³/s = 19.8 g/s. Vent area 80 cm², free-area fraction 0.45: effective area 36 cm². Vent velocity: 19.8×10⁻³/0.0036 = 5.5 m/s — below the 7 m/s threshold. Vent sizing is acceptable.

Convection vs radiation at the heatsink base (80°C, bare aluminium ε ≈ 0.1, forced h = 40 W/m²K): q_conv = 40×55 = 2200 W/m²; q_rad = 0.1×5.67×10⁻⁸×(353⁴−298⁴) ≈ 43 W/m². Radiation is 2% of convection — safely neglected. TIM layer pre-check: k = 3 W/mK, t = 0.5 mm, A = 50 cm². R_θ,TIM = 0.0005/(3×0.005) = 0.033 K/W. At 300 W: ΔT_TIM = 9.9°C. Non-negligible — the TIM layer must be included in the Icepak model explicitly.