14.3 Pendulums: Linear, Nonlinear, Driven & Double ended on the nonlinear term that broke superposition. Heat transfer is the opposite case: inside a solid the physics is the linear heat equation of 9.3 Solving the Heat Equation: Separation of Variables. What is new is the boundary, where a hot body loses heat to a fluid through a Robin condition with a coefficient . One dimensionless number, the Biot number, says whether the PDE is needed or a first-order ODE like the cooling coffee of 3.5 Modeling with First-Order ODEs will do. The rest uses the course so far: the fin is a boundary-value problem (8.1 Two-Point Boundary Value Problems), the sink a resistor network (14.2 Electrical Circuits), and 2D conduction a relaxation (10.8 Elliptic Solvers: Jacobi, Gauss–Seidel & SOR).
Transcript
- Heat moves three ways. Conduction is diffusion through a solid, Fourier's law. Convection is heat carried off by a moving fluid, which engineers squeeze into Newton's cooling law with a coefficient h. Radiation needs no medium: a surface emits epsilon sigma T to the fourth, with T absolute. Factor T to the fourth minus T infinity to the fourth, and radiation also becomes Newton cooling, with h radiation equal to epsilon sigma times T plus T infinity times T squared plus T infinity squared. For an anodised surface at 80 degrees in 25 degree air, with epsilon 0.9, h radiation is 7.10 watts per metre squared kelvin. That is the size of natural convection in air, so it cannot be ignored.
- A uniform body of volume V loses heat only through its surface, so rho c V theta dot equals minus h A theta. The time constant tau is rho c V over h A, which is R th times C th: the RC circuit again, with no thermal inductor. The assumption is a uniform body. Heat crosses conduction and convection resistances in series, and their ratio is the Biot number, h L over k. Take a steel plate, L equal to 10 millimetres. In still air, h is 25, Bi is 0.0056 and tau is 1529 seconds: lumped is fine. Quenched in water, h is 1000, Bi is 0.222, tau is 38.2 seconds, and the surface is cold while the centre is hot.
- When Bi is not small, solve the slab. Separation of variables and symmetry give a cosine, and the Robin condition becomes zeta tan zeta equals Bi, a transcendental eigenvalue problem. Zeta tan zeta rises from zero to infinity on each interval, so there is exactly one root per interval and bisection finds it. For small Bi the leading mode is exactly the lumped result. At t over tau equal to 1, with Bi equal to 0.1, the centre is 0.3861 and the surface 0.3676, and the lumped curve is 0.3679. At Bi equal to 1, the centre has only dropped to 0.5339. At 10 it is 0.9684, while the surface has nearly reached the fluid. The lumped error against the centre is minus 4.7, minus 31.1 and minus 62.0 percent.
- A fin adds surface. A slice of the rod has conduction in, conduction out, and convection from its sides, so k A theta double prime equals h P theta, which is theta double prime equals m squared theta. With the base fixed and an insulated tip, the solution is cosh of m times L minus x, over cosh of m L, and the heat rate is M theta b times tanh of m L. Take an aluminium fin, k equal to 200, h equal to 50, 2 millimetres thick and 30 long. Then m is 16.12 per metre and m L is 0.4837. The tip is at 74.14 degrees, the fin passes 7.968 watts, and its efficiency, tanh of m L over m L, is 92.87 percent.
- Now a sink with 8 of these fins on a base. Each fin passes 7.968 watts, and the unfinned base passes 6.05. Conductances add, like parallel resistors, so the total is 8 times 7.968 plus 6.05, which is 69.79 watts. The bare base alone passes 8.25 watts, so the fins multiply the heat by 8.46, and the sink resistance, theta b over q, is 0.788 kelvin per watt. Add a fan, h equal to 100, and the heat rises to 131.3 watts, the resistance falls to 0.419. This is Kirchhoff's law from lesson 14.2, with temperature as voltage and watts as current.
- Without time dependence, the heat equation becomes Laplace's equation: each node is the average of its four neighbours, and sweeping that rule until nothing changes is relaxation. The test problem is a hot inner square in a cold outer square. The heat per metre is the shape factor S prime, times k, times the temperature difference, and S prime depends only on the ratio b over a. For b over a equal to 2 it is about 10.23, and for 3, about 6.22. Counting sweeps with 60 cells across, Gauss-Seidel takes 715 and optimal SOR 199. At 120, 2568 against 394. At 240, 9121 against 786. Doubling the grid multiplies Gauss-Seidel by about 3.6, and SOR by 2.
- Fins turned a boundary condition into the whole problem: theta double prime equals m squared theta, with cosh and sinh solutions mixed by the tip condition. The next lesson is vibrating beams. The syllabus lists Euler-Bernoulli theory, the static deflection problem, clamped, pinned and free ends, mode shapes and natural frequencies, numerical beam solvers, and a glimpse of wing bending and flutter. The beam equation is fourth order, and, from memory, to be shown there, its mode shapes mix cosh and sinh with cos and sin. So this lesson's boundary-condition bookkeeping returns, with two conditions at each end.
Three ways heat moves
Conduction is diffusion through a solid, Fourier’s law (W/m) from 9.2 The Heat (Diffusion) Equation. Convection is heat carried off by a moving fluid; engineers lump the flow into Newton’s cooling law
with in W/(m K). Typical ranges (from memory, orders of magnitude): natural convection in air to , forced air to , forced water to . Radiation needs no medium: a surface emits with absolute, emissivity and W/(m K) (from memory). Factoring gives Newton cooling with
For C ( K) in C air ( K) with (anodised aluminium, from memory), W/(m K): the size of natural convection, so it cannot be ignored there.
Newton’s cooling, lumped capacitance and the Biot number
3.5 Modeling with First-Order ODEs fitted a cooling constant and warned that Newton’s law needs a uniform temperature. Now we derive it and quantify the warning. A body of volume , density , specific heat and surface area , at uniform , loses heat only through its surface. With ,
This is the RC charging of 14.2 Electrical Circuits with thermal capacitance (J/K) and convective resistance (K/W). Heat has no inertia, so there is no thermal inductor and no ringing.
The model assumes a uniform body. Heat actually crosses two resistances in series: conduction through the solid, about , and convection at the surface, . Their ratio is the Biot number
If the solid conducts easily compared with the surface carrying heat away, the interior stays uniform, and the lumped model holds. A common rule from memory is ; we measure below what it buys. The is the solid’s.
Worked example: a steel plate. Take steel (from memory: W/(m K), kg/m, J/(kg K)), mm thick, so mm. In still air, : and s, about min. Quenched in water, (an order of magnitude): , s, and the lumped model is doubtful, since the surface is cold while the centre is hot.
The slab with a convective boundary
Take a slab , initially , with both faces cooled by a fluid:
Separation as in 9.3 Solving the Heat Equation: Separation of Variables and symmetry give with . The Robin condition is
This is the Sturm-Liouville eigenvalue problem of 8.2 Eigenvalue Problems & Sturm–Liouville Theory with a Robin end, and it is transcendental. But rises from to infinity on each interval , so there is exactly one root per interval; bisection on finds it with no pole to cross. Time dependence is with the Fourier number , and orthogonality of the cosines gives
Two limits check it. As , , the Dirichlet eigenvalues. As , gives , and the leading mode decays as : exactly the lumped result. The lab’s roots, checked with scipy.optimize.brentq:
The series converges slowly at small times ( needs about terms; the lab uses ), while implicit finite differences (10.4 Implicit Methods & Crank–Nicolson) handle the Robin end as a ghost-point row: use the series to validate such a code.
Lumped against slab: what Bi buys
The lab plots four curves against , so the lumped curve is the same for every (and ). At the lumped value is ; the slab values, matching an independent -term script, are:
| centre | surface | average | lumped | lumped against centre | |
|---|---|---|---|---|---|
At the centre is above the surface and lumped is within of the volume average, which is what the lumped model really computes. At the centre has only dropped to while the surface is at . At the surface has nearly reached while the centre has barely noticed. The error also grows with time, because the rate is wrong: the late-time slope of the average is of the lumped slope, , and . At the lumped time constant is about times too short.
At late times centre over surface is : , , . Since , this is about , so means a spread near . It is a tolerance, not a law.
Cooling fins: a boundary-value problem
A fin adds surface for convection. A rod of cross-section , perimeter and conductivity conducts along and loses heat through its sides. A slice of width has conduction in, conduction out and convection :
with a decay length. Using one coordinate is the Biot idea again: it needs the transverse Biot number small, which is for the fin below. With and an insulated tip ,
A long fin has and , the infinite-fin rate. For a convective tip, with gives . The efficiency is actual heat over the heat if the whole fin sat at ; for an insulated tip
Worked example. An aluminium fin (from memory, W/(m K)), mm thick, mm long, mm deep, in air with W/(m K); base C, air C, so K. mm, mm, so m (the thin-fin is low) and . The tip is at , that is C; W; . The convective tip gives W, only more, because . Both match a solve_bvp run.
How long should it be? The infinite-fin limit is W:
| (mm) | (W) | efficiency | |
|---|---|---|---|
Going from to mm multiplies the metal by but the heat by only . The last row has ; beyond it extra metal is wasted.
The fin BVP also discretises: with a ghost point at the tip and a tridiagonal solve, as in 8.1 Two-Point Boundary Value Problems. The exercise does this against the closed form.
The sink as a network of resistances
A sink has fins on a base. Each passes and the unfinned base passes . Conductances (W/K) add like parallel resistors, so . For of the fins above on a mm base ( mm unfinned, mm deep, insulated tips):
The bare base passes W, so the fins give , and K/W. With a fan, , the total is W and K/W (fin efficiency ): doubling cut the resistance by , not .
Series resistances add too: for a W chip with (hypothetical) K/W junction-to-case and K/W interface, C, or C with the fan. This is Kirchhoff’s law of lesson 14.2 with temperature as voltage and watts as current.
2D steady conduction
With the heat equation becomes Laplace’s equation, , with the neighbour-average property of 9.6 Laplace's Equation: Rectangles & Disks. On a grid each node is the mean of its four neighbours, and sweeping that rule until nothing changes is relaxation, the solvers of 10.8 Elliptic Solvers: Jacobi, Gauss–Seidel & SOR.
The test problem is a square flue: a hot inner square (, side ) in a cold outer square (, side ). The heat per metre is , where the shape factor depends only on , because Laplace’s equation has no length scale. By Green’s identity , so the lab sums squared differences over all grid edges. For cells across the outer square, converged to a largest change of and checked against a sparse direct solve:
| extrapolated | ||||
|---|---|---|---|---|
| about | ||||
| about |
The differences shrink by per halving (order , not : the inner corners are singular). A fit I remember from heat-transfer texts, for , gives and , about and high; I could not check the coefficients against a source. For an assumed W/(m K) and K, gives W per metre.
Sweeps to converge for (with in brackets), using , optimal for the plain square:
| Gauss-Seidel | SOR | |
|---|---|---|
| () | () | |
| () | () | |
| () | () |
Doubling multiplies Gauss-Seidel by about (toward the limit of ) and SOR by , the law. Multigrid (10.9 Multigrid) removes the growth.
Heat exchangers: a glimpse
In a heat exchanger two streams swap heat across a wall. With capacity rates , (W/K) and overall coefficient , parallel flow along area obeys
This linear system (7.2 Linear Systems & the Eigenvalue Method) has eigenvalue (energy conservation) and , so the difference decays as , with , . The effectiveness is
and counterflow has (from memory) . With W/K, , W/K (, ), inlets C and C: parallel (hot outlet C), counterflow (C), both matching a numerical ODE solve.
The lab
The view menu switches between three experiments.
- Fin view, default (, , mm, mm, ): , , tip C, W, efficiency ; sink W against W bare, K/W.
- Convective tip: W, tip C, sink W.
- Press the aluminium-and-fan button (): W, K/W. Reset and set mm: one fin W at , the sink W ( K/W), with cm of metal against the fan’s .
- Slab view, , cursor at : centre , surface , average , lumped . At the curves nearly merge; at the lumped line falls away.
- 2D view, outer/inner , grid : Gauss-Seidel sweeps, SOR , . At , Gauss-Seidel takes sweeps and .
The exercise: fin and Biot eigenvalue
The playgrounds reproduce the fin-length table and the slab-against-lumped comparison. The exercise builds three pieces: the closed form for an insulated tip, a finite-difference solve of the same BVP (the tridiagonal system with a ghost point), and the first Biot eigenvalue by bisection.
Concept checks
Problems
An aluminium plate (from memory: W/(m K), kg/m, J/(kg K)), mm thick, so mm, at C in C air with W/(m K). Check that the lumped model applies, then find the time to reach C.
Show worked solution
, far below . s. The excess falls from K to K, so s, about min.
A circular pin fin, mm, mm, W/(m K), W/(m K), insulated tip, base K above the air. Find , the tip excess temperature, the heat rate and the efficiency.
Show worked solution
, , so , m, . Tip excess K. W/K, so W. Efficiency .
(a) Use in to show and compare with at . (b) Show the leading-mode centre-to-surface ratio is about .
Show worked solution
(a) ; replacing in the bracket by gives at , against the exact . (b) The ratio is ; exactly at .
Per square metre: m brick ( W/(m K)), m foam (), indoor , outdoor W/(m K), air at C inside and C outside (all assumed). Find the heat flux and the brick-foam interface temperature.
Show worked solution
Resistances (mK/W): . Flux W/m. Interface: C. The foam takes K of the K drop.
What comes next
Fins turned a boundary condition into the whole problem: , with and solutions mixed by the tip condition. The next lesson, 14.5 Vibrating Beams & Structures, is vibrating beams. The syllabus lists Euler-Bernoulli theory, the static deflection BVP, clamped, pinned and free ends, mode shapes and natural frequencies, numerical beam solvers, and a glimpse of wing bending and flutter. The beam equation is fourth order, and (from memory, to be shown there) its mode shapes mix and with and , so this lesson’s boundary-condition bookkeeping returns with two conditions at each end.
Key Takeaways
- Heat moves by conduction, convection () and radiation (); the engineering sits in the boundary condition.
- Lumped capacitance is , : an RC circuit with no inductor.
- compares internal and surface resistance: at the centre-surface spread is about , at the lumped time constant is times too short.
- The convective slab has eigenvalues from , one per interval, found by bisection; recovers the lumped model.
- The fin BVP gives , and : diminishing returns in length.
- Thermal resistances add like electrical ones, in series and in parallel.
- 2D conduction is Laplace relaxation: Gauss-Seidel sweeps grow like , SOR like .