Thermal Verification, Sensitivity & Test Correlation

What Is It?

Thermal verification, sensitivity and test correlation are the processes that establish confidence in a thermal prediction. Verification checks that the numerical model has solved the equations correctly — energy balance, mesh convergence, time-step convergence. Sensitivity analysis checks how the result changes when uncertain inputs (convection coefficient, emissivity, contact conductance) are varied. Test correlation compares the predicted temperatures with measured temperatures from physical testing. Together, these processes distinguish a credible thermal analysis from a calculation.

Why It Matters

A thermal model is often dominated by uncertain boundary conditions rather than numerical solver error. The convection coefficient, the emissivity, the contact conductance and the heat input may each be uncertain by 20–50% or more. The solver may converge to a tight tolerance, but if the boundary conditions are wrong, the result is wrong. Verification, sensitivity and correlation are the tools that address this — verification confirms the numerics, sensitivity reveals the effect of uncertain inputs, and correlation confirms (or reveals discrepancies in) the physical model. Without these, a thermal prediction is an unvalidated calculation.

A thermal model is often dominated by uncertain boundary conditions rather than numerical solver error. The solver may converge beautifully, but if the convection coefficient, the emissivity or the contact conductance is wrong, the result is wrong. Verification, sensitivity and correlation are essential.

Energy Balance Check

The energy balance is the most fundamental verification check. At steady state, the total heat entering the system must equal the total heat leaving. At transient, the heat in minus the heat out equals the rate of stored energy. The energy balance should be checked for the whole system and, where possible, for sub-regions. A significant imbalance indicates a problem: a missing boundary condition, an unintended heat leak, a mesh error or a convergence problem. The energy balance is the thermal equivalent of the force balance in structural analysis — it must be satisfied.

Mesh and Time-Step Convergence

Mesh convergence checks that the temperature field does not change with further mesh refinement. The analysis is run on progressively finer meshes and the critical temperatures (peak, interface, surface) are compared. If they stabilise, the mesh is adequate. Time-step convergence checks that the transient result does not change with smaller time steps. Both are necessary verifications — a result that has not been shown to be mesh- or time-step-independent is not verified.

Heat-Source Verification

The heat sources must be verified — is the total heat input correct? Is the spatial distribution correct? Is the temporal profile correct? A common error is a unit conversion mistake (W vs kW, W vs W/m³) or a factor-of-area error (total power vs power per unit area). The total heat input should be checked against the known power dissipation. The spatial distribution should be checked against the actual component layout. The temporal profile should be checked against the actual duty cycle.

Boundary-Condition Sensitivity

The boundary conditions — particularly the convection coefficient, the emissivity and the contact conductance — are often uncertain. A sensitivity study varies these parameters within their plausible range and checks the effect on the critical temperatures. If the temperature is insensitive (changes by a small amount), the uncertainty is not a concern. If the temperature is highly sensitive (changes by a large amount), the boundary condition must be better defined — through more accurate correlations, CHT or testing. The sensitivity study reveals which boundary conditions matter and which do not.

Analytical Checks

Before relying on the FEA result, analytical checks should be performed to verify the order of magnitude. A thermal resistance network — a one-dimensional hand calculation using series and parallel resistances — gives a quick estimate of the temperature. A one-dimensional conduction calculation gives the through-thickness gradient. A lumped-capacitance calculation gives the transient response time. These checks do not replace the FEA but they provide a sanity check — if the FEA result is wildly different from the analytical estimate, something is wrong.

Temperature Sanity Checks

Beyond the formal checks, simple sanity checks can catch errors. The peak temperature should not exceed physically plausible values (e.g. above the material melting point). The temperature distribution should make physical sense — hot near heat sources, cool near heat sinks, gradients in the expected direction. A temperature that is higher than the heat source temperature (without a reason) indicates a problem. A region that should be cool but is hot in the model may have a missing heat path or an incorrect boundary condition.

THERMAL CHECK: Correlate the physics, not just one temperature point. Matching a single thermocouple does not validate the model. The energy balance, the heat-source magnitude, the boundary conditions and the temperature distribution should all be consistent with the physics.

Test Correlation

Test correlation compares the predicted temperatures with measured temperatures from physical testing. The comparison reveals whether the model represents the real physics. Common temperature measurement methods include thermocouples (point measurements, robust, wide range), RTDs (point measurements, high accuracy), thermal imaging (surface temperature field, non-contact) and heat-flux sensors (direct heat-flux measurement). Each method has its own limitations — thermocouples disturb the local temperature, thermal imaging requires emissivity correction, heat-flux sensors are sensitive to mounting. The comparison should consider the sensor location, the sensor disturbance, the thermal lag and the measurement uncertainty.

Sensor Disturbance and Location

Temperature sensors can disturb the local temperature. A thermocouple attached to a surface may act as a heat fin — conducting heat away from the surface and lowering the local temperature. The sensor location must be precisely known — a thermocouple 5 mm from the intended location may read a significantly different temperature if the gradient is steep. The sensor should be placed at the critical location — the expected hot spot, the interface, the point of maximum gradient. Multiple sensors provide redundancy and help validate the temperature distribution, not just a single point.

Calibration

When test data is available, the model can be calibrated — the uncertain parameters (convection coefficient, emissivity, contact conductance) are adjusted to match the measured temperatures. Calibration should be done carefully — adjusting parameters to match one temperature point may produce a model that matches that point but is wrong elsewhere. The calibration should match multiple points and should preserve the physical plausibility of the parameters. A calibrated model is more credible than an uncalibrated one, but it is still a model — it should be validated against additional data that was not used for calibration.

Key Takeaways

• Thermal models are often dominated by uncertain boundary conditions, not solver error
• Energy balance, mesh convergence and time-step convergence are the fundamental verification checks
• Sensitivity studies on h, ε, hc and heat input reveal which uncertainties matter
• Analytical checks (resistance networks, 1D conduction) provide sanity checks on the FEA
• Test correlation should validate the physics — energy balance, distribution and multiple points — not just one temperature