What is thermal analysis?

Thermal analysis computes temperature: where heat is generated, how it travels, and what temperature everything settles at — or how long it takes to get there. The output is a temperature field, plus the heat flux crossing surfaces.

Heat moves by three mechanisms, and knowing which dominates tells you what kind of analysis you need.

• Conduction — through solids, driven by temperature gradient and thermal conductivity. This is the part FEA-style solvers handle directly.
• Convection — heat carried away by a moving fluid. Either modelled cheaply as a coefficient on a surface, or solved properly by computing the flow.
• Radiation — exchange by electromagnetic emission, proportional to the fourth power of absolute temperature. Ignorable at low temperatures, dominant at high ones.

A steady-state analysis answers “what temperature does this reach and hold”. It is cheap, and it is the right question for continuous operation.

A transient analysis answers “how long until it gets there”, and needs one extra material property — specific heat — because you are now tracking stored energy, not just its flow. Use it for warm-up and cool-down, duty cycles, thermal shock, or any case where the peak is a spike rather than a plateau. A component that survives steady operation can still fail during a transient it never reaches equilibrium in.

This is the choice that determines both cost and credibility, and it is where thermal analysis touches CFD.

Solid-only with a convection coefficient. You apply a heat transfer coefficient h and an ambient temperature to each cooled surface. Fast, and entirely adequate when a reliable h is available — from a correlation, a handbook or measurement. Its weakness is that you supplied the answer to the hardest part of the question. A guessed coefficient is the single biggest source of error in thermal work, and the result inherits its uncertainty without showing it.

Conjugate heat transfer. The fluid and the solid are solved together, so the coefficient is computed rather than assumed. This is what you need when the flow pattern is the point — a heat sink whose fins might stall the air, an enclosure with recirculation, natural convection where the flow is driven by the heating itself. It costs a CFD run, and it is the only honest option when nobody can tell you what h should be.

A useful middle path: run conjugate heat transfer once for a representative case, extract the effective coefficients, then use the cheap solid-only model for the many design variants that follow.

Radiation is nonlinear in temperature and depends on what each surface can “see” of every other surface. Those view factors are geometric and expensive to compute for a complicated assembly. It also depends on emissivity, a surface property that varies with finish and degrades with age — a polished metal and the same metal oxidised are different thermal problems.

Rules of thumb worth carrying: below roughly 100 °C with any air movement, radiation is usually a minor correction. In a vacuum it is the only mechanism there is. Above a few hundred degrees it dominates.

Heat generation is often the coupling to another physics rather than a number you type in. Resistive losses in a conductor, iron and copper loss in a motor, induction heating and eddy-current losses all come out of an electromagnetic analysis. Friction and plastic work come from a structural one. Combustion comes from a reacting flow solution.

And temperature usually feeds back out again: constrained thermal expansion produces stress and distortion, which is a coupled problem and one of the most common reasons people need more than one solver.

Thermal analysis is unusual in that it rarely gets a dedicated code. Conduction with convection and radiation boundaries is solved by structural codes such as code_aster and by multiphysics codes such as Elmer; conjugate heat transfer is solved by CFD codes such as OpenFOAM; induction and Joule heating are solved by the electromagnetic code that computed the losses in the first place.