How fast does heat move?
The first law tells you how much heat it takes to warm a tank of water; the second law tells you which way heat flows. Neither tells you how fast. Yet nearly every thermal design question is about rates and temperatures. How thick must the insulation on a house in Ulaanbaatar be, where winter nights reach −30 °C? How large a heat sink keeps a 300 W processor below 90 °C? How long does a casting take to cool? How many square metres of tubes does a power-station condenser need? How deep does winter frost reach into the ground, and so how deep must a water main be buried?
Heat transfer answers these. Heat moves in three ways, by conduction, convection and radiation, each with its own rate law. This lesson builds the three laws, combines them into networks of thermal resistances, solves conduction problems that formulas cannot by finite differences, and finishes with the two most common pieces of heat-transfer hardware: fins and heat exchangers.
Three ways heat travels
Stand by a wood stove. The iron handle of its door gets hot because heat conducts through solid metal, passed from atom to atom by vibrations and, in metals, by free electrons. The warm air rising off the stove carries heat across the room by convection: a moving fluid picks up energy at a hot surface and takes it elsewhere. And your face feels warm even through still air, because the stove glows in the infrared, sending energy by radiation, which needs no material at all and is how the Sun's energy crosses 150 million kilometres of vacuum.
Most real problems mix all three. Heat leaving a heated room passes by convection (and radiation) to the inner face of the wall, by conduction through the layers of the wall, and by convection and radiation from the outer face to the cold air and sky.
Analogy, not a formal construction
Heat flow through a chain of materials behaves like electric current through a chain of resistors, the analogy behind Ohm's law: temperature difference plays the part of voltage, heat flow the part of current, and each layer or surface has a thermal resistance. Resistances in series add; paths in parallel share the flow. The analogy is exact for steady conduction, and it is how engineers reason about every thermal design before they compute anything.
The three rate laws
Definition— Fourier's law of conduction
The heat flux (heat flow per unit area, W/m²) conducted through a material is proportional to the temperature gradient and flows down it:
where is the thermal conductivity, in W/(m·K).
Typical values near room temperature, rounded: copper about 400 W/(m·K), aluminium alloys 120–240, carbon steel about 50, stainless steel about 15, concrete about 1.4, glass about 1, brick 0.6–1, wood 0.1–0.2, water 0.6, mineral wool and polymer foams 0.03–0.04, still air 0.026. Good insulators are mostly trapped air; the solid is there to stop the air moving.
Definition— Newton's law of cooling
The heat flux from a surface at to a fluid at is
where is the convection heat-transfer coefficient, in W/(m²·K).
Newton's law is really a definition of , which is not a property of the fluid but of the whole flow: speed, geometry, fluid properties and whether the fluid boils or condenses. Orders of magnitude are worth memorising: a few to about 25 W/(m²·K) for air moving by itself (natural convection); tens to a few hundred for air blown by a fan; hundreds to tens of thousands for pumped water; and up to around for boiling and condensation.
Definition— Stefan–Boltzmann law
A surface at absolute temperature with emissivity () emits a heat flux , where W/(m²·K⁴). A small surface surrounded by large surroundings at exchanges net
Combining Fourier's law with an energy balance on a small volume gives the heat equation, the partial differential equation every conduction solver solves:
where is any heat generated per unit volume (by an electric current, say) and is the thermal diffusivity, in m²/s. Conductivity says how much heat flows for a given gradient; diffusivity says how fast a temperature change spreads. Its mathematics is in partial differential equations and Fourier series.
Thermal resistance networks
For steady one-dimensional heat flow, each rate law becomes "heat flow = temperature difference / resistance":
| Process | Thermal resistance (K/W) |
|---|---|
| Conduction through a slab of thickness , area | |
| Conduction through a cylindrical shell, radii to , length | |
| Convection from area | |
| Radiation from area (linearised) | , |
Building engineers work per square metre and call the reciprocal of the total resistance the U-value, in W/(m²·K).
Example— An insulated cavity wall
A wall is built (inside to outside) of 12.5 mm plasterboard (), 100 mm mineral wool () and 102.5 mm brick (). Take W/(m²·K) inside (convection and radiation together) and 25 outside. Per square metre the resistances are, in m²·K/W:
so W/(m²·K). With 20 °C inside and 0 °C outside the flux is 6.2 W/m², and 40 m² of wall loses about 250 W. Without the mineral wool, and the same wall loses 2.3 kW. The temperatures at each interface follow by subtracting layer by layer: 19.2 °C on the inner face of the plasterboard, 18.9 °C behind it, then a plunge to 1.1 °C across the insulation, and 0.25 °C on the outer face of the brick. Nearly all the temperature drop sits where nearly all the resistance is.
The cylinder resistance hides a surprise. Insulating a pipe or wire adds conduction resistance but also enlarges the outer surface, lowering the convection resistance. For an insulation of conductivity and an outer coefficient , the total is smallest at the critical radius . For mineral wool in still air that is about 4 mm, smaller than almost any pipe, so insulation always helps. For PVC on a thin electrical wire it is around 16 mm, so the insulation on a cable actually increases the heat it sheds, a happy accident for cable ratings.
Conduction by finite differences
Formulas cover slabs, cylinders and a few idealised shapes. Real parts have corners, varying properties and changing boundary temperatures, so engineers solve the heat equation numerically. The simplest method replaces derivatives by differences on a grid of spacing and time step . Write for the temperature at node and time level . In 1-D, the second derivative becomes , and stepping forward in time gives the explicit (forward-time, centred-space) scheme
The grid Fourier number must not exceed . Rewrite the update as : when all three weights are non-negative and add to 1, so the new temperature is an average of old ones and can never overshoot. Above the middle weight goes negative, a node that is hotter than its neighbours becomes colder than them at the next step, and the oscillation grows without limit.
A cold snap in the ground
The ground surface, initially at 10 °C throughout, drops suddenly to −10 °C and stays there for 30 days. How deep does the 0 °C line reach? Treat the ground as a deep slab with a moist soil's diffusivity, m²/s. This problem also has an exact solution, , which lets us check the numbers.
import math
import numpy as np
# A cold snap: the ground surface drops from +10 C to -10 C and stays there for 30 days.
alpha = 0.5e-6 # thermal diffusivity of a moist soil, m^2/s (roughly)
T_i, T_s = 10.0, -10.0
depth, dx = 6.0, 0.05 # model 6 m of ground in 5 cm slices
x = np.arange(0.0, depth + dx / 2, dx)
def simulate(Fo, days):
dt = Fo * dx**2 / alpha
steps = int(round(days * 86400 / dt))
T = np.full(x.size, T_i)
T[0] = T_s # surface held at -10 C
for _ in range(steps): # explicit (FTCS) update of every interior node
T[1:-1] = T[1:-1] + Fo * (T[2:] - 2 * T[1:-1] + T[:-2])
# the deepest node stays at T_i: far enough down to be undisturbed
return T, dt, steps
T, dt, steps = simulate(Fo=0.4, days=30)
print(f"time step {dt:.0f} s, {steps} steps")
print(" depth (m) finite difference exact (erf)")
for d in [0.0, 0.1, 0.25, 0.5, 0.75, 1.0, 1.5, 2.0]:
i = int(round(d / dx))
exact = T_s + (T_i - T_s) * math.erf(d / (2 * math.sqrt(alpha * 30 * 86400)))
print(f" {d:5.2f} {T[i]:8.3f} {exact:8.3f}")
# where does the 0 C isotherm reach? interpolate between nodes
k = np.argmax(T > 0.0)
frost = x[k - 1] + (0.0 - T[k - 1]) / (T[k] - T[k - 1]) * dx
eta = 0.476936 # erf(eta) = 0.5
print(f"\n0 C reaches {frost:.3f} m (finite difference); "
f"exact {2 * eta * math.sqrt(alpha * 30 * 86400):.3f} m")
# The same scheme with too large a time step
for Fo in [0.5, 0.6]:
T, dt, steps = simulate(Fo, days=3)
print(f"Fo = {Fo}: after {steps} steps the largest |T| is {np.abs(T).max():.3g} C")With (time steps of 2000 s, 1296 of them) the finite-difference temperatures agree with the exact solution to within 0.002 K, and both put the 0 °C line at 1.086 m after a month. At the scheme is on the edge but still bounded; at 0.6 the largest "temperature" is °C after 86 steps, a numerical explosion, not physics. Implicit schemes, which solve a linear system at every step, remove the limit, and the same idea, called the CFL condition, governs computational fluid dynamics.
Two dimensions
In 2-D steady conduction without heat generation the heat equation becomes Laplace's equation, , and on a square grid with its finite-difference form says that every interior temperature is the average of its four neighbours. Writing that for every node gives one linear equation per node, solved here directly with numpy.linalg.solve. The problem: a long square bar with one face held at 100 °C and the other three at 0 °C, which has a Fourier-series solution to compare with, and a neat check: by symmetry, the four problems with one hot face each add up to a bar entirely at 100 °C, so the centre must be at exactly 25 °C.
import numpy as np
# Steady 2-D conduction in a long square bar: one face at 100 C, the other three at 0 C.
# Unknowns: the temperatures at the n x n interior nodes of a grid with spacing h = 1/(n+1).
n = 31
N = n * n
idx = lambda i, j: i * n + j # i: row (y), j: column (x)
A = np.zeros((N, N))
b = np.zeros(N)
T_top = 100.0
for i in range(n):
for j in range(n):
k = idx(i, j)
A[k, k] = -4.0 # five-point Laplacian: neighbours - 4 * centre = 0
for di, dj in [(1, 0), (-1, 0), (0, 1), (0, -1)]:
ii, jj = i + di, j + dj
if 0 <= ii < n and 0 <= jj < n:
A[k, idx(ii, jj)] = 1.0
elif ii == n: # the neighbour is on the hot top face
b[k] -= T_top
# neighbours on the other three faces are at 0 C and add nothing
T = np.linalg.solve(A, b).reshape(n, n)
def exact(x, y, terms=199): # Fourier-series solution of the same problem
s = 0.0
for m in range(1, terms + 1, 2):
s += 4 * T_top / (m * np.pi) * np.sin(m * np.pi * x) * np.sinh(m * np.pi * y) / np.sinh(m * np.pi)
return s
h = 1.0 / (n + 1)
print(f"{N} unknowns, grid spacing h = {h:.4f}")
print(" x y finite difference Fourier series")
for x, y in [(0.5, 0.5), (0.5, 0.25), (0.5, 0.75), (0.25, 0.75), (0.25, 0.5), (0.125, 0.875)]:
i, j = int(round(y / h)) - 1, int(round(x / h)) - 1
print(f" {x:5.3f} {y:5.3f} {T[i, j]:8.3f} {exact(x, y):8.3f}")
print("\nisotherms across the bar (top row is next to the 100 C face):")
for i in range(n - 2, -1, -4):
print(" " + " ".join(f"{T[i, j]:3.0f}" for j in range(1, n, 4)))The centre comes out at 25.000 °C, as symmetry demands, and elsewhere the 961-unknown solution agrees with the Fourier series to within about 0.03 K. The printed grid shows the heat spreading down from the hot face, with the steepest temperature gradients at the two top corners, where the hot face meets the cold ones. Finite element programs do the same thing on unstructured meshes of real parts (see the finite element method), with sparse solvers that handle millions of unknowns.
Convection: coefficients and dimensionless numbers
Convection is conduction into a moving fluid. Right at the wall the fluid is at rest (no slip, from viscous flow), so the heat crosses a thin thermal boundary layer by conduction, and the faster the flow, the thinner the layer and the larger . Because depends on so much, it is correlated through dimensionless groups.
Definition— Nusselt, Prandtl and Biot numbers
- The Nusselt number is the convection coefficient made dimensionless with a length and the fluid's conductivity: how much better the flow transfers heat than pure conduction through a fluid layer of thickness would.
- The Prandtl number compares how fast momentum and heat diffuse in the fluid: about 0.7 for air, 7 for water at 20 °C, hundreds for oils.
- The Biot number , with the volume divided by the surface area, compares the resistance inside a solid with the convection resistance outside it.
A correlation gives Nu as a function of Re and Pr (for forced convection) or of the Rayleigh number (for natural convection, driven by buoyancy). Two classic ones for flow inside tubes: fully developed laminar flow has for a constant wall temperature and 4.36 for a constant heat flux; turbulent flow (, ) is often estimated with the Dittus–Boelter correlation
For example, 0.2 kg/s of water at about 40 °C ( Pa·s, W/(m·K), ) heated in a 20 mm tube has , and W/(m²·K).
Lumped bodies
When , conduction inside a body is so easy compared with convection outside that the body stays nearly uniform in temperature as it heats or cools. An energy balance, , then gives an exponential approach to the fluid temperature with time constant . Small, conductive objects in gases qualify; a thick concrete block in water does not, and needs the conduction solution above.
Radiation
Every surface above absolute zero emits electromagnetic radiation, and at the temperatures of everyday engineering almost all of it is in the infrared. A perfect emitter, a black body, emits ; real surfaces emit a fraction of that, and absorb the same fraction of the radiation falling on them (at the same wavelengths). Polished aluminium has ; most paints, whatever their visible colour, human skin, water and oxidised metals are around 0.9 or more in the infrared.
Radiation matters even at room temperature. A clothed person with 1.5 m² of surface at 30 °C and , in a room whose walls are at 20 °C, radiates a net 86 W, comparable with the heat lost by convection. The linearised radiation coefficient is W/(m²·K), about the same as natural convection in still air, which is why cold walls and windows make a room feel chilly even when the air is warm.
Between surfaces that see each other, how much radiation leaving surface 1 lands on surface 2 is a purely geometric quantity, the view factor . Two rules make view factors tractable: reciprocity, , and summation, for an enclosure. For two large parallel plates the net flux is , so two shiny surfaces facing each other barely exchange heat. A thin shiny sheet placed between them, a radiation shield, cuts the exchange further; dozens of aluminised plastic layers make the multilayer insulation that wraps spacecraft. Complex geometries are handled by view-factor tables in handbooks or by Monte Carlo ray tracing in software.
Fins
When the surface's convection resistance is the bottleneck, add surface: fins, the ribs on a motorcycle cylinder, the plates of a computer heat sink, the spines on a heat-pump coil. A fin is not at the base temperature everywhere: heat must conduct along it while convection drains heat from its sides, so its temperature falls towards the tip. An energy balance on a thin slice of a fin with cross-section and perimeter gives, for the excess temperature ,
With an insulated tip, the solution is , the heat the fin sheds is , and its fin efficiency (actual heat over the heat it would shed if it were all at the base temperature) is . The length is the natural scale: a fin much longer than about is mostly at air temperature and wasted.
The program solves the fin equation by finite differences (a tridiagonal linear system, with a mirror node for the insulated tip), checks it against the formula, and compares a 5 mm pin fin in aluminium and in stainless steel, with its base at 80 °C in air at 20 °C and W/(m²·K).
import numpy as np
# A pin fin: diameter 5 mm, base at 80 C, air at 20 C, h = 40 W/(m^2 K), tip insulated.
D, h, T_b, T_inf = 0.005, 40.0, 80.0, 20.0
P, Ac = np.pi * D, np.pi * D**2 / 4 # perimeter and cross-section
def fin_fd(k, L, n=200):
"""Finite differences for theta'' = m^2 theta, theta(0) = theta_b, theta'(L) = 0."""
m2 = h * P / (k * Ac)
dx = L / n
A = np.zeros((n + 1, n + 1))
rhs = np.zeros(n + 1)
A[0, 0], rhs[0] = 1.0, T_b - T_inf
for i in range(1, n):
A[i, i - 1], A[i, i], A[i, i + 1] = 1.0, -2.0 - m2 * dx**2, 1.0
A[n, n - 1], A[n, n] = 2.0, -2.0 - m2 * dx**2 # mirror node: zero gradient at the tip
theta = np.linalg.solve(A, rhs)
# heat leaving the fin = convection from its whole surface (trapezoidal rule)
q = h * P * dx * (theta.sum() - 0.5 * theta[0] - 0.5 * theta[-1])
return theta, q
def fin_exact(k, L):
m = np.sqrt(h * P / (k * Ac))
q = np.sqrt(h * P * k * Ac) * (T_b - T_inf) * np.tanh(m * L)
return q, np.tanh(m * L) / (m * L)
for name, k in [("aluminium", 200.0), ("stainless steel", 15.0)]:
m = np.sqrt(h * P / (k * Ac))
print(f"{name}: k = {k:.0f} W/(m K), m = {m:.1f} 1/m, 1/m = {1000 / m:.0f} mm")
print(" L (mm) q FD (W) q exact (W) tip T (C) efficiency")
for L in [0.01, 0.02, 0.05, 0.10, 0.20]:
theta, q = fin_fd(k, L)
qe, eff = fin_exact(k, L)
print(f" {L * 1000:5.0f} {q:7.3f} {qe:7.3f} {theta[-1] + T_inf:6.1f} {eff:5.2f}")
print()
q_bare = h * Ac * (T_b - T_inf)
print(f"the 5 mm spot of bare base the fin replaces would lose only {q_bare:.3f} W")The finite differences match the formula to the third decimal. An aluminium pin ( mm) keeps 89% efficiency at 50 mm and sheds 1.67 W, against 0.047 W from the bare spot it stands on: an effectiveness of about 35. Doubling it to 100 mm adds only half as much again, and at 200 mm the tip is within 10 K of the air. The stainless-steel pin ( mm) is finished by about 50 mm: doubling it to 100 mm adds 2%. Fin design is a trade between conductivity, length, spacing (fins too close choke the airflow) and mass, which is why heat sinks are aluminium or copper and why their fins are short and many.
Heat exchangers and the log-mean temperature difference
A heat exchanger moves heat from one fluid to another through a wall: the radiator of a car, the condenser of a power station, the evaporator of a heat pump, the plate exchanger that pasteurises milk. The main types are double-pipe (one tube inside another), shell-and-tube (a bundle of tubes in a cylindrical shell, with baffles to steer the shell-side flow), plate (corrugated plates clamped together, compact and easy to clean) and cross-flow finned coils for air. The fluids may run in the same direction (parallel flow), opposite directions (counterflow) or across each other.
The wall's local heat flux is , where the overall coefficient combines the two convection resistances, the wall's conduction and fouling resistances for the scale and deposits that build up in service. Because the temperature difference changes along the exchanger, the total duty needs the right average.
Theorem— Log-mean temperature difference
For a parallel-flow or counterflow exchanger with constant and constant specific heats, the duty is
where and are the temperature differences between the two fluids at the two ends.
Proof
Write and for the heat-capacity rates and . Across a strip of wall of area , heat leaves the hot fluid and enters the cold. Marching from end 1 to end 2, the hot fluid cools by ; in parallel flow the cold fluid warms by , and in counterflow (cold fluid travelling the other way) it cools by in the marching direction. Either way . Divide by and integrate over the whole area: . Integrating itself gives . Dividing the second equation by the first eliminates the bracket: .
∎
Example— Sizing a water-to-water exchanger
Cool 1.0 kg/s of hot water from 90 °C to 50 °C with 1.5 kg/s of water entering at 20 °C, with W/(m²·K) and J/(kg·K). The energy balance fixes the duty and the cold outlet; the LMTD then fixes the area for each arrangement. The program also checks the counterflow answer the way the proof works, by marching along the wall in small steps until the hot water has reached 50 °C.
import numpy as np
# Water-to-water exchanger: cool 1.0 kg/s of hot water from 90 C to 50 C with 1.5 kg/s of water at 20 C.
cp = 4180.0 # J/(kg K)
mh, mc = 1.0, 1.5
Th_in, Th_out, Tc_in = 90.0, 50.0, 20.0
U = 1500.0 # overall coefficient, W/(m^2 K)
Ch, Cc = mh * cp, mc * cp # heat-capacity rates, W/K
Q = Ch * (Th_in - Th_out) # duty from the hot side
Tc_out = Tc_in + Q / Cc # energy balance gives the cold outlet
print(f"duty Q = {Q / 1e3:.1f} kW, cold water leaves at {Tc_out:.2f} C")
def lmtd(dT1, dT2):
return (dT1 - dT2) / np.log(dT1 / dT2)
for name, dT1, dT2 in [("counterflow ", Th_in - Tc_out, Th_out - Tc_in),
("parallel flow", Th_in - Tc_in, Th_out - Tc_out)]:
dTlm = lmtd(dT1, dT2)
print(f"{name}: end differences {dT1:5.2f} and {dT2:5.2f} K, LMTD {dTlm:5.2f} K,"
f" area {Q / (U * dTlm):.2f} m^2")
# Check the counterflow answer by marching along the area, from the hot inlet end.
# Both streams cool in the direction of marching (the cold one flows the other way).
dA = 1e-4
A, Th, Tc = 0.0, Th_in, Tc_out
while Th > Th_out:
dq = U * (Th - Tc) * dA # heat crossing this strip of wall
Th -= dq / Ch
Tc -= dq / Cc
A += dA
print(f"marching: hot side reaches {Th_out:.0f} C after {A:.3f} m^2; cold side there is {Tc:.3f} C")The duty is 167.2 kW and the cold water leaves at 46.67 °C. Counterflow keeps the two streams 30–43 K apart all the way (LMTD 36.3 K) and needs 3.07 m²; parallel flow starts with a 70 K difference but squeezes it to 3.3 K at the outlet (LMTD 21.9 K) and needs 5.09 m², two-thirds more. The marching check lands on 3.074 m² with the cold side at exactly 20 °C, confirming the formula. In counterflow the cold outlet could even be hotter than the hot outlet; in parallel flow it never can.
When the outlet temperatures are unknown (rating an existing exchanger rather than sizing a new one) the LMTD method needs iteration, and engineers use the equivalent effectiveness–NTU method instead: the effectiveness compares the duty with the most that could ever be transferred, (here ), and is a known function of the number of transfer units for each flow arrangement. Multi-pass shell-and-tube and cross-flow exchangers use the LMTD with a correction factor, from charts or formulas in the handbooks.
Heat transfer in professional practice
- Buildings. Building codes in many countries set maximum U-values for walls, roofs and windows, calculated by standard methods (ISO 6946 for building components, via ISO), and the ASHRAE Handbook is a standard reference for the heat gains and losses of buildings. Thermal bridges, where a steel beam or a concrete slab edge short-circuits the insulation, are found with infrared cameras and 2-D or 3-D conduction models like the one above; heating, cooling and buildings takes this further.
- Electronics. Every chip is a heat source in a network of thermal resistances: junction to case, case to heat sink, heat sink to air or liquid. Data centres in 2026 increasingly use liquid cooling for AI accelerators; see thermal management for computing and powering data centres in Lumenary EE.
- Process plant and power. Shell-and-tube exchangers are designed mechanically to the standards of TEMA and the pressure-vessel codes, and thermally with specialist rating software from several vendors, built on decades of research measurements. Fouling allowances, cleaning schedules and vibration of tubes in the shell-side flow are as important as the LMTD.
- Simulation. Conduction in solids is solved by finite elements in open-source codes such as CalculiX and Elmer and in commercial packages from Ansys, Dassault Systèmes (Abaqus), Siemens and COMSOL; conjugate heat transfer (solid and fluid together) uses CFD. Hot-section parts of gas turbines, cooled by internal passages and protected by ceramic thermal-barrier coatings (see ceramics and glasses in Lumenary Materials), are among the hardest heat-transfer designs in engineering.
Heat-transfer equipment can be dangerous: surfaces hot enough to burn, steam and pressurised hot water, and cryogenic liquids cold enough to injure. Industrial plant is guarded, insulated and lagged to keep touchable surfaces safe, and pressure parts are certified and inspected. Safe experiments at home are simple ones: timing two ice cubes as they melt, one on a metal baking tray and one on a wooden chopping board (the metal one wins, for the reason in the first misconception below), or logging a mug of warm water cooling with a kitchen thermometer.
A short history
Isaac Newton published his law of cooling anonymously in 1701, from experiments on hot iron cooling in a draught. Joseph Fourier, prefect of the Isère department under Napoleon, wrote down the heat equation and invented Fourier series to solve it, in work submitted in 1807 and published in 1822. Josef Stefan found the fourth-power law from experiments in 1879, and his former student Ludwig Boltzmann derived it from thermodynamics in 1884. Wilhelm Nusselt, working in Germany around 1915, showed how dimensional analysis organises convection measurements; Ludwig Prandtl's boundary-layer theory explained them. The log-mean temperature difference was already in use in the nineteenth century for designing condensers, and the effectiveness–NTU method was developed in the 1950s for compact heat exchangers. Relaxation methods for grids like the ones above were worked by hand in the 1940s, and grid solutions of partial differential equations were among the first problems given to electronic computers.
Common misconceptions
Misconception
"Metal feels colder than wood because it is colder." In a room both are at the same temperature. Metal conducts heat away from your hand much faster, so your skin cools more quickly where it touches it.
Misconception
"White surfaces radiate less heat." Colour describes visible light. In the infrared, where room-temperature and warm objects radiate, white paint has an emissivity near 0.9, as high as black paint. To radiate less you need a shiny metal surface, not a light colour. (White does reflect more sunlight, which is a different part of the spectrum.)
Misconception
"The convection coefficient is a property of the fluid." It depends on the flow speed, the geometry, the fluid and whether it boils or condenses. The same water gives of a few hundred in a slow, wide channel and many thousands in a fast, narrow one.
Misconception
"A smaller time step always makes a simulation more accurate, so stability is only about accuracy." For the explicit scheme, a time step above the limit is not just inaccurate: the solution explodes. Below it, accuracy also depends on the grid spacing, so shrinking alone cannot remove the spatial error.
Misconception
"More fins, or longer fins, always mean more cooling." Beyond a length of about a fin's tip is at air temperature and adds nothing but weight, and fins packed too closely choke the airflow between them, lowering . Fins also help only on the side whose convection resistance controls the heat flow.
Exercises
1. Heat transfer
Warm-upHumanA 10 m² wall has a total thermal resistance (surfaces included) of 2.5 m²·K/W. It is 20 °C inside and −5 °C outside. What is the steady heat loss through the wall, in watts?
2. Heat transfer
Warm-upHumanA vacuum flask has two glass or steel walls with a vacuum between them, and the facing surfaces are silvered or polished. What does the silvering mainly do?
3. Heat transfer
CoreHumanAn aluminium sphere 20 mm in diameter ( kg/m³, J/(kg·K), W/(m·K)) cools in air with W/(m²·K). Check that it can be treated as a lumped body, then give its thermal time constant , in seconds.
4. Heat transfer
CoreHumanIn a counterflow heat exchanger, oil is cooled from 90 °C to 50 °C by water that is heated from 20 °C to 40 °C. What is the log-mean temperature difference, to the nearest kelvin?
5. Heat transfer
CoreHumanWhich of these statements about thermal radiation are true? Select all that apply.
6. Heat transfer
CoreComputationalWrite wall_heat_flux(layers, h_in, h_out, T_in, T_out) for a plane wall. layers is a list of (thickness, k) pairs from inside to outside (metres, W/(m·K)); h_in and h_out are the surface coefficients in W/(m²·K). Return a tuple (U, q, temps): the U-value, the heat flux in W/m² (positive from inside to outside), and a list of temperatures at the inner surface, at each interface between layers, and at the outer surface (so len(layers) + 1 values). The tests use the lesson's insulated wall.
7. Heat transfer
CoreComputationalWrite explicit_conduction(T, Fo, steps) that advances a 1-D temperature profile by steps steps of the explicit scheme , keeping the first and last values fixed (they are boundary temperatures). Raise ValueError if Fo > 0.5. Do not modify the array passed in; return a new NumPy array.
Further reading and resources
- Théorie analytique de la chaleur · Joseph Fourier · 1822The book that founded the mathematical theory of heat conduction, and Fourier series with it.
- A Heat Transfer Textbook · John H. Lienhard IV and John H. Lienhard VA complete university heat-transfer textbook, free to download from its authors.
- ASHRAEThe heating, refrigeration and air-conditioning engineers' society; its Handbook is a standard source for building heat transfer.
- Tubular Exchanger Manufacturers Association (TEMA)Publishes the standards widely used for the mechanical design of shell-and-tube heat exchangers.
- CalculiXAn open-source finite element program that solves steady and transient heat conduction as well as structural problems.
- NIST Chemistry WebBook: Thermophysical Properties of Fluid Systems · National Institute of Standards and TechnologyFree thermal conductivity, viscosity and heat-capacity data for water, air and many other fluids.