Tools › Fluids and Heat Transfer › Conduction › Thermal Network Solver › Method
Thermal Network Solver: Method
Open the Thermal Network Solver tool
Thermal Network Solver: Method, Node and Conductor Types, Assumptions and Limits
What the tool does
The tool solves a lumped-parameter thermal network of the kind used by SINDA-type thermal analyzers. The object is divided into nodes, each at one temperature, joined by conductors that carry heat between them. Two tables describe the network. The tool finds either the steady-state temperatures or the temperature histories from an initial state to an end time, and reports the temperature of every node, the heat flow in every conductor, the heat taken by every boundary node, and the heat balance of the whole network.
Nodes
Diffusion node. A lump of material with a thermal capacitance C = m cp (mass times specific heat). Its temperature changes gradually: C dT/dt equals the net heat flowing into it. The entered temperature is its initial temperature.
Arithmetic node. A massless node. It is in heat balance at every instant, so its temperature follows its neighbors at once. Use it for surfaces, interfaces, bolted joints and other points whose capacitance is negligible beside the rest. The entered temperature is only a starting guess.
Boundary node. A node held at the entered temperature whatever heat flows into it: ambient air, a cold plate, deep space, a fixed-temperature mounting face. A heat load on a boundary node has no effect.
A constant heat load Q may be put on any diffusion or arithmetic node (dissipation, absorbed solar or infrared flux); a negative value removes heat.
Conductors
Linear conductor. Heat flow from node i to node j is q = G (Ti − Tj). The conductance G is k A / L for conduction through a path of conductivity k, cross-section A and length L; h A for convection from an area A with coefficient h; hc A for a contact of conductance hc per unit area. Conductances in series combine as 1/G = 1/G1 + 1/G2; in parallel they add. The helper in the Advanced inputs computes one of these three and uses it for a chosen conductor row.
Radiation conductor. Heat flow is q = σ (ε A F) (Ti4 − Tj4) with absolute temperatures. The entered value is the product of the area and the gray-body exchange factor between the two surfaces, in m² or ft². For a small convex surface of emittance ε and area A that sees only a large enclosure or space, it is ε A. For two large parallel plates of equal area it is A / (1/ε1 + 1/ε2 − 1). For black surfaces it is Ai Fij with the view factor Fij. The Stefan-Boltzmann constant is σ = 2π5k4/(15 h3c2) = 5.670374 × 10−8 W/(m²⋅K4) from the defining constants of the SI; it can be changed in the Advanced inputs.
Equations
The net heat into node i is
Ri(T) = Qi + Σ Gij (Tj − Ti) + Σ σ (ε A F)ij (Tj4 − Ti4)
summed over the conductors attached to the node.
Steady state. Ri = 0 at every diffusion and arithmetic node. With radiation the equations are nonlinear; they are solved by Newton-Raphson, T ← T − J−1R with the Jacobian J = ∂R/∂T (a radiation conductor contributes 4 σ (ε A F) T3), with a step-halving line search that keeps every absolute temperature positive and the residual falling. Capacitances play no part. A steady state exists only when every node has a conductor path to a boundary node; the tool says which node does not.
Transient. For diffusion nodes Ci dTi/dt = Ri, for arithmetic nodes Ri = 0. The time from zero to the end time is cut into equal steps Δt and each step solves
Ci (Ti′ − Ti) / Δt = θ Ri(T′) + (1 − θ) Ri(T)
for the new temperatures T′ by Newton-Raphson, together with Ri(T′) = 0 at the arithmetic nodes. θ = 1 is the backward Euler scheme (first-order accurate, never oscillates); θ = ½ is the Crank-Nicolson scheme (second-order accurate, but it can oscillate from step to step when Δt is much longer than the shortest time constant in the network). Both are implicit and stable for any step, so the step is chosen for accuracy and not for stability, unlike the explicit forward-difference scheme whose step must stay below the smallest C/ΣG of the network. Arithmetic nodes are first balanced at time zero against the initial temperatures of the diffusion nodes.
Step count. Left at 0, the number of steps is 20 tend/τmin, held between 200 and 2,000, where τmin is the smallest node time constant Ci/ΣGi with radiation conductors linearized at the initial temperatures (4 σ (ε A F) T3). The automatic scheme is Crank-Nicolson when Δt does not exceed τmin and backward Euler otherwise.
Time-step error estimate. The transient is run a second time with half as many steps. The largest difference between the two runs, over every node and every common time, is printed as the estimate of the time-step error (divided by three for Crank-Nicolson, by Richardson's rule for a second-order scheme). If it is larger than the accuracy needed, enter more steps.
Time to a target temperature. The first time at which the chosen node crosses the target, by linear interpolation between steps.
Heat balance. At steady state the heat loads equal the heat taken by the boundary nodes. In a transient the difference is the rate of heat storage Σ C dT/dt at the end time, and over the whole run the energy stored equals the loads less the heat given to the boundary nodes; both identities are printed with their residuals.
Units
Conductance, radiation area and heat load carry units through the site's unit dictionary. Temperatures are plain numbers in °C (SI) or °F (USCS), capacitance in J/K or Btu/°F, conductivity in W/(m⋅K) or Btu/(h⋅ft⋅°F), film coefficient in W/(m²⋅K) or Btu/(h⋅ft²⋅°F); a change of the unit system does not convert these plain numbers, so re-enter them. The Stefan-Boltzmann constant is always entered in 10−8 W/(m²⋅K4).
What it assumes
Each node is isothermal: the temperature difference inside a lump is small beside the differences between lumps (a Biot number of the lump well below one, or enough nodes across a gradient). Conductances, capacitances and heat loads are constant in time and do not depend on temperature. Radiation exchange is between gray, diffuse surfaces and is fully described by the entered ε A F of each pair; the tool does not compute view factors or multiple reflections (use the Radiation View Factors tool, or a gray-body factor from an enclosure analysis). Heat flows only through the listed conductors.
When it misleads
A network is only as good as its nodes: too few nodes across a steep gradient give a wrong peak temperature that no solver setting will repair. Contact conductances, convection coefficients and emittances are often known to no better than a factor of two, and the result is no more accurate than they are. Conductivity, specific heat and emittance that change over a wide temperature range, time-varying heating (orbital flux, duty cycles), thermostats and heaters, fluid loops, phase change and one-way conductors are not modeled: bracket such cases by runs at the extreme constant values, or use a full thermal analyzer. The transient result carries a time-step error, shown by the printed estimate; a long step under Crank-Nicolson can also ring. Radiation makes the steady problem nonlinear but it still has a single solution for positive absolute temperatures; a failure to converge points to nodes joined only by conductances many orders of magnitude apart.
Limits
At most 40 nodes, 120 conductors and 2,000 time steps. Node labels must differ; a conductor names its nodes by label or by row number of the node table.
Checks
The default network is the hollow aluminum box in deep space of the NASA lesson Building Your First SINDA Model (Rickman, 2020): six wall nodes of 6,263.9 J/K, twelve edge conductors of 0.51816 W/K, six radiation conductors of ε A = 0.88 m² to space at −273.15 °C and 370.17 W of absorbed sunlight on one wall. The lesson prints −23.2 °C for the sunlit wall, −120.9 °C for the opposite wall and −107.9 °C for the four side walls, both at steady state and after 50,000 s; the tool gives −23.24, −120.88 and −107.94 °C at steady state. (The lesson uses σ = 5.67 × 10−8; with that value the tool gives −23.24, −120.88 and −107.93 °C.) The solver is also checked against closed forms (walls in series and in parallel, the lumped-capacitance exponential, a node radiating to a sink, a radiation shield, radiative cooling of a lump) and against an independent numerical solution of seeded random networks.