Skip to content
LumenaryAcademy
Thermofluids 75 min

Viscous flow and pipes

Why real fluids need a push to keep moving — viscosity, laminar and turbulent flow, friction in pipes and fittings, pumps and where they settle, and boundary layers and drag in outline.

Code tested · 5AI-draftedAwaiting review by a mechanical engineer

Your guides: Chloe · The Designer, Margaret · The Test Engineer, Gideon · The Checker

MotivationH

The push a pipe needs

In Fluid statics and Bernoulli an ideal fluid traded pressure, speed and height along a streamline and lost nothing. Apply that to a level pipe of constant bore. Continuity says the mean speed is the same at both ends; the height is the same; so Bernoulli says the pressure is the same too, and water should glide along a kilometre of pipe for ever without being pushed.

Every water utility knows otherwise. The mains under a street in Nairobi need pumping stations; a district-heating loop in Copenhagen has circulating pumps working day and night; the oil in a car engine needs an oil pump to reach the bearings; your heart does real work to push blood through your arteries. Pumps and fans are among the largest users of electricity in industry and buildings, and much of that electricity goes to overcoming one thing: viscosity, the internal friction of a real fluid.

This lesson puts viscosity back: why fluid sticks to walls, why one dimensionless number decides between smooth and chaotic flow, how to compute the pressure a pipe and its fittings need, how a pump and a pipe system agree on a flow rate, and, in outline, the boundary layers and drag on everything that moves through air or water. Heat transfer and computational fluid dynamics build on all of it.

IntuitionH

Layers that drag on each other

Stir a jar of honey and a glass of water with the same spoon. The honey resists: it takes force to make one layer slide over the next. That resistance to shearing is viscosity. It is not the same as density: olive oil floats on water, yet it is about eighty times more viscous.

The simplest case is a thin film between two large plates, the lower one fixed and the upper one dragged along at speed UU. The fluid touching each plate moves with that plate, the speed in between varies linearly from zero to UU, and the force needed to keep the top plate moving is proportional to UU and to the plate's area, and inversely proportional to the gap.

hUmoving plate, speed Ufixed plateu(y) = U y / hτ = μ U / h
Shear flow between a fixed plate and a plate moving at speed U. The fluid sticks to both plates (no slip) and its speed rises linearly across the gap h. Keeping the top plate moving takes a shear stress τ = μU/h, where μ is the viscosity.

This no-slip condition is the heart of viscous flow. Fluid at a wall is at rest relative to it and fluid far away is moving, so there is shearing in between, and shearing a viscous fluid takes work. That work becomes internal energy, warming the fluid very slightly; engineers call it a loss, and the pump must make it good.

In 1883 Osborne Reynolds, in Manchester, ran water from a tank through a glass tube with a thin streak of dye at the entrance. At low speeds the streak stayed a clean thread; above a certain speed it wavered and burst into eddies that mixed the dye through the tube. He had made visible the two regimes of pipe flow: laminar, moving in smooth layers, and turbulent, with unsteady eddies of many sizes tumbling across the flow.

Analogy, not a formal construction

Laminar flow is like disciplined motorway traffic: every car keeps its lane, and the lanes slide past one another. Turbulent flow is like a busy market crowd, where people weave across the aisle carrying their momentum into slower regions, so speeds even out, but those next to the stalls are jostled far harder. Turbulence in a pipe likewise flattens the velocity profile and multiplies the friction at the wall.

DefinitionsH

Viscosity and the Reynolds number

Definition— dynamic viscosity (Newton's law of viscosity)

For a Newtonian fluid the shear stress on layers sliding over one another is proportional to the rate of shear, the velocity gradient across the layers:

τ=μ dudy.\tau = \mu\,\frac{\mathrm du}{\mathrm dy}.

The constant μ\mu is the dynamic viscosity, measured in pascal seconds (Pa·s). Water, air, oils and most gases and simple liquids are Newtonian.

Definition— kinematic viscosity

The ratio ν=μ/ρ\nu = \mu/\rho of viscosity to density, in m²/s, is the kinematic viscosity. It measures how fast momentum diffuses through the fluid, and it is the combination that appears in most flow formulas.

Representative values near 20 °C and atmospheric pressure, rounded (take design values from property databases such as the NIST Chemistry WebBook):

Fluidρ\rho (kg/m³)μ\mu (Pa·s)ν\nu (m²/s)
Air1.21.8×10−51.8 \times 10^{-5}1.5×10−51.5 \times 10^{-5}
Water9981.0×10−31.0 \times 10^{-3}1.0×10−61.0 \times 10^{-6}
Olive oil910about 0.08about 9×10−59 \times 10^{-5}
Glycerol1260about 1.4about 1.1×10−31.1 \times 10^{-3}
Honeyabout 14002 to 1010−310^{-3} or more

Viscosity depends strongly on temperature: water's halves between 20 °C and 50 °C, and engine oil can be a hundred times thicker on a frozen morning than in a hot engine. Gases go the other way: their viscosity rises with temperature, because it comes from molecules carrying momentum between layers, and hotter molecules travel faster.

Note

Many everyday fluids are non-Newtonian: their apparent viscosity depends on how fast they are sheared. Paint, ketchup and blood thin as they are sheared harder (shear-thinning); a thick suspension of cornflour in water stiffens suddenly when struck (shear-thickening); toothpaste does not flow at all until the stress passes a threshold (a yield-stress fluid). Their behaviour is the subject of rheology. Everything below assumes a Newtonian fluid.

Definition— Reynolds number

For a flow of speed VV past or through an object of size DD,

Re=ρVDμ=VDν.\mathrm{Re} = \frac{\rho V D}{\mu} = \frac{V D}{\nu}.

It compares the momentum the fluid carries (inertia) with the viscous stresses that smooth it out. For flow in a round pipe, with VV the mean speed and DD the internal diameter, the flow is normally laminar below Re≈2300\mathrm{Re} \approx 2300 and turbulent above about 4000, with an unsteady, intermittent transitional range in between.

The Reynolds number spans an astonishing range: about 10−510^{-5} for a swimming bacterium, which stops dead the instant it stops pushing; 10410^4 for water in a garden hose; 10610^6 for air past a car; 10910^9 for a large ship. Two flows with the same geometry and Reynolds number behave alike whatever their size, which is why models can be tested in towing tanks and wind tunnels.

The number 2300 is a convention, not a law. Laminar pipe flow becomes turbulent only when disturbances are large enough: with very smooth inlets and quiet tanks it has been kept laminar beyond Re=105\mathrm{Re} = 10^5, while below roughly 2000 every turbulent patch eventually decays. In a real plant, with pumps, bends and valves upstream, expect transition between about 2000 and 4000, and in that range design for whichever regime gives the larger pressure drop.

DefinitionsH

Laminar pipe flow, exactly

Laminar flow in a long straight pipe is one of the few viscous flows that can be solved exactly with a pencil, and the answer is worth knowing by heart.

Theorem— Hagen–Poiseuille flow

In steady, fully developed laminar flow of a Newtonian fluid through a round pipe of radius R=D/2R = D/2, with a pressure drop Δp\Delta p over a length LL, the velocity at radius rr is the paraboloid

u(r)=Δp4μL(R2−r2),u(r) = \frac{\Delta p}{4\mu L}\left(R^2 - r^2\right),

the volume flow rate is

Q=πR4 Δp8μL=πD4 Δp128 μL,Q = \frac{\pi R^4\,\Delta p}{8\mu L} = \frac{\pi D^4\,\Delta p}{128\,\mu L},

and the mean velocity V=Q/(πR2)V = Q/(\pi R^2) is exactly half the centreline velocity.

Proof

Take as a free body a cylinder of fluid of radius rr and length LL, coaxial with the pipe. Because the flow is fully developed, the velocity profile is the same at both ends, so momentum flows in and out at the same rate and the net force on the cylinder is zero. The pressure difference pushes on its ends with force Δp πr2\Delta p\,\pi r^2; the slower fluid outside drags back on its curved surface with shear stress τ\tau over area 2πrL2\pi r L. Equilibrium gives τ=Δp r/(2L)\tau = \Delta p\, r/(2L): the shear stress rises linearly from zero on the axis to its maximum at the wall. The velocity falls as rr increases, so Newton's law of viscosity reads τ=−μ du/dr\tau = -\mu\,\mathrm du/\mathrm dr, and

dudr=−Δp2μL r⟹u=Δp4μL(R2−r2),\frac{\mathrm du}{\mathrm dr} = -\frac{\Delta p}{2\mu L}\,r \quad\Longrightarrow\quad u = \frac{\Delta p}{4\mu L}\left(R^2 - r^2\right),

where the constant of integration was fixed by no slip, u(R)=0u(R) = 0. The flow rate is the integral of uu over the cross-section, Q=∫0Ru(r) 2πr dr=πR4Δp/(8μL)Q = \int_0^R u(r)\,2\pi r\,\mathrm dr = \pi R^4 \Delta p/(8\mu L). Dividing by the area πR2\pi R^2 gives V=R2Δp/(8μL)V = R^2\Delta p/(8\mu L), which is half of u(0)=R2Δp/(4μL)u(0) = R^2 \Delta p /(4\mu L).

∎

The fourth power of the radius is the striking part. Halve the diameter of a laminar pipe, at the same pressure difference, and the flow falls by a factor of 16. That is why a small narrowing of a blood vessel matters so much to the heart, and why lubricant films and capillaries are so sensitive to their gap.

laminarturbulentdashed: mean speed Vdashed: mean speed V2V≈1.2V
Velocity profiles in a pipe for the same flow rate. Laminar flow (left) has a parabolic profile whose centreline speed is twice the mean. Turbulent flow (right) is much flatter in the core, about 1.2 times the mean on the axis, with a steep fall in a thin layer next to the wall, which is why its wall friction is so much higher.

Rearranged in terms of the mean velocity, Hagen–Poiseuille reads

Δp=64Re⋅LD⋅ρV22,\Delta p = \frac{64}{\mathrm{Re}}\cdot\frac{L}{D}\cdot\frac{\rho V^2}{2},

which is the Darcy–Weisbach equation below with friction factor f=64/Ref = 64/\mathrm{Re}. In laminar flow the pressure drop is proportional to the flow rate; that is about to change.

DefinitionsH

Head loss: Darcy–Weisbach, Colebrook and fittings

Between two points 1 and 2 of a pipe system, the energy equation for steady incompressible flow is Bernoulli's equation with two new terms, written as heights of fluid ("heads", in metres):

p1ρg+V122g+z1+hpump=p2ρg+V222g+z2+hL.\frac{p_1}{\rho g} + \frac{V_1^2}{2g} + z_1 + h_\text{pump} = \frac{p_2}{\rho g} + \frac{V_2^2}{2g} + z_2 + h_L.

hpumph_\text{pump} is the head a pump adds; hLh_L is the head loss, the mechanical energy per unit weight of fluid turned into internal energy on the way. (It is a per-unit-weight form of the steady-flow energy equation of the first law.) A head of 1 m of water is a pressure of ρg×1 m≈9.8\rho g \times 1\ \text{m} \approx 9.8 kPa.

Definition— Darcy–Weisbach equation

The head loss to friction in a straight pipe of length LL and internal diameter DD, with mean velocity VV, is

hf=f LD V22g,equivalentlyΔp=f LD ρV22,h_f = f\,\frac{L}{D}\,\frac{V^2}{2g}, \qquad\text{equivalently}\qquad \Delta p = f\,\frac{L}{D}\,\frac{\rho V^2}{2},

where the dimensionless Darcy friction factor ff depends on the Reynolds number and the relative roughness ε/D\varepsilon/D of the pipe wall.

In laminar flow, f=64/Ref = 64/\mathrm{Re} and roughness does not matter. In turbulent flow there is no exact theory, but there is an excellent fit to experiment, published by Cyril Colebrook in 1939 from his own and Johann Nikuradse's measurements:

1f=−2log⁡10 ⁣(ε/D3.7+2.51Re f).\frac{1}{\sqrt f} = -2\log_{10}\!\left(\frac{\varepsilon/D}{3.7} + \frac{2.51}{\mathrm{Re}\,\sqrt f}\right).

In a smooth pipe (ε=0\varepsilon = 0) only Re\mathrm{Re} matters, and ff falls slowly as it grows: the viscous layer at the wall buries any small roughness. At very high Re\mathrm{Re} the second term vanishes and the pipe is fully rough: the bumps poke through the thin viscous layer and ff depends only on ε/D\varepsilon/D. The equation is implicit, with ff on both sides and no algebraic solution. Engineers of the 1940s read ff from Lewis Moody's chart; today we solve it numerically in microseconds, or use explicit approximations such as Haaland's (1983), within a couple of per cent:

1f≈−1.8log⁡10 ⁣[(ε/D3.7)1.11+6.9Re].\frac{1}{\sqrt f} \approx -1.8\log_{10}\!\left[\left(\frac{\varepsilon/D}{3.7}\right)^{1.11} + \frac{6.9}{\mathrm{Re}}\right].

Typical roughness heights, for orientation (pipe makers publish values for their products, and roughness grows as pipes age and corrode):

Pipeε\varepsilon (mm)
Drawn copper or plastic (PVC, PE)about 0.0015 to 0.007
Commercial steelabout 0.045
Cast ironabout 0.25
Concreteabout 0.3 to 3

Minor losses

Bends, valves, tees, entrances, exits and changes of section disturb the flow and cause extra losses. Each is described by a loss coefficient KK, so that hm=K V2/(2g)h_m = K\,V^2/(2g), and the system's total loss is

hL=(fLD+∑K)V22g.h_L = \left(f\frac{L}{D} + \sum K\right)\frac{V^2}{2g}.

Indicative values, which vary with size, design and manufacturer: a sharp-edged entrance from a tank, about 0.5; a well-rounded one, below 0.05; any exit into a large tank, 1.0 (all the kinetic energy is dissipated); a 90° elbow, roughly 0.2 to 1; a fully open gate or ball valve, below 0.2; a fully open globe valve, around 5 to 10. In a plant room full of fittings, "minor" losses can be most of the total. One can be derived exactly: a momentum balance on a sudden enlargement from area A1A_1 to A2A_2 gives the Borda–Carnot loss K=(1−A1/A2)2K = (1 - A_1/A_2)^2 on the upstream velocity, which becomes the exit loss K=1K = 1 as A2→∞A_2 \to \infty.

ComputationC

Computing it: Colebrook by iteration

Write x=1/fx = 1/\sqrt f. Then Colebrook's equation is x=−2log⁡10(ε/D3.7+2.51 xRe)x = -2\log_{10}\big(\tfrac{\varepsilon/D}{3.7} + \tfrac{2.51\,x}{\mathrm{Re}}\big), which already has the form x=g(x)x = g(x), ready for fixed-point iteration: guess xx, evaluate the right-hand side, repeat. It converges because ∣g′(x)∣|g'(x)| is small (the logarithm flattens everything), well under 0.1 for turbulent flow, so each iteration gains at least one correct digit.

The program below solves the worked example (water at 20 °C, 3.0 L/s through 100 m of 50 mm commercial-steel pipe), compares Haaland's approximation, and prints a small Moody chart as a table.

python
import numpy as np

def colebrook(Re, rr, tol=1e-12):
    """Darcy friction factor from the Colebrook equation (turbulent flow).
    Re: Reynolds number; rr: relative roughness eps/D."""
    x = 8.0                                   # x = 1/sqrt(f); f = 0.016 is a fair first guess
    for it in range(100):
        x_new = -2.0 * np.log10(rr / 3.7 + 2.51 * x / Re)
        if abs(x_new - x) < tol:
            break
        x = x_new
    return 1.0 / x_new**2, it + 1

def haaland(Re, rr):
    return (-1.8 * np.log10((rr / 3.7) ** 1.11 + 6.9 / Re)) ** -2

# Worked example: water at 20 C, 3.0 L/s through 100 m of 50 mm commercial steel pipe
rho, nu = 998.0, 1.00e-6                      # kg/m^3, m^2/s
Q, D, L, eps = 3.0e-3, 0.050, 100.0, 0.045e-3
V = Q / (np.pi * D**2 / 4)
Re = V * D / nu
f, n_it = colebrook(Re, eps / D)
hf = f * (L / D) * V**2 / (2 * 9.81)
print(f"V = {V:.3f} m/s, Re = {Re:.0f}, eps/D = {eps / D:.4f}")
print(f"Colebrook f = {f:.5f} after {n_it} iterations; Haaland f = {haaland(Re, eps / D):.5f}")
print(f"head loss {hf:.2f} m of water = pressure drop {rho * 9.81 * hf / 1e3:.1f} kPa")

# A text Moody chart: f against Re for several relative roughnesses
rrs = [0.0, 1e-4, 1e-3, 1e-2]
print("\n       Re  laminar" + "".join(f"  eps/D={r:<6g}" for r in rrs))
for Re in [1e3, 2e3, 4e3, 1e4, 1e5, 1e6, 1e7, 1e8]:
    lam = f"{64 / Re:8.4f}" if Re < 2300 else "       -"
    turb = "".join(f"  {colebrook(Re, r)[0]:12.4f}" if Re >= 4000 else "             -" for r in rrs)
    print(f" {Re:8.0e} {lam}{turb}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The mean velocity is 1.53 m/s and the Reynolds number about 76 000, well into turbulence. The iteration settles to f=0.02248f = 0.02248 in a dozen steps, and Haaland's formula gives 0.02223, about 1% lower. The 100 m of pipe costs 5.35 m of head, a pressure drop of 52 kPa.

The table is the Moody chart in numbers. In a smooth pipe ff keeps falling, from 0.040 at Re=4000\mathrm{Re} = 4000 to 0.006 at 10810^8; at ε/D=0.01\varepsilon/D = 0.01 it levels off at 0.038 from about 10510^5 on, where the pipe is fully rough and viscosity has stopped mattering. Notice also the jump at transition: at Re=3000\mathrm{Re} = 3000 a laminar flow would have f=64/3000=0.021f = 64/3000 = 0.021, while Colebrook's smooth-pipe value there is 0.044, so a flow that trips into turbulence can double its pressure drop with no change in flow rate.

Worked examplesH

Pumps, system curves and operating points

A centrifugal pump spins an impeller that flings fluid outwards; its casing turns much of the fluid's speed into pressure. At a fixed speed the head it delivers falls as the flow rises, and the maker publishes this pump curve Hpump(Q)H_\text{pump}(Q), measured to a standard such as ISO 9906, with the efficiency and power at each flow.

The pipework has its own curve. To push QQ through it, the pump must supply the static lift Δz\Delta z (plus any difference in surface pressures) and the losses, which grow roughly as Q2Q^2:

Hsystem(Q)=Δz+(fLD+∑K)V22g,V=QπD2/4.H_\text{system}(Q) = \Delta z + \left(f\frac{L}{D} + \sum K\right)\frac{V^2}{2g}, \qquad V = \frac{Q}{\pi D^2/4}.

Pump and pipes must agree: the flow settles at the operating point, where the curves cross. The shaft power the pump then draws is

P=ρgQHη,P = \frac{\rho g Q H}{\eta},

where η\eta is the pump efficiency at that point.

Example— Lifting water to a header tank

A pump lifts water at 20 °C from a sump to a tank 12 m higher, through 150 m of 80 mm commercial-steel pipe with fittings whose loss coefficients add up to ∑K=8\sum K = 8. The pump, at its rated speed, has (as a model fitted to its published curve) H=32−0.060 Q2H = 32 - 0.060\,Q^2, with HH in metres and QQ in L/s, and its best efficiency, 72%, at 14 L/s. Find the operating point and the power. Then the plant needs only 10 L/s: compare closing a valve (throttling) with slowing the pump.

Because ff depends on QQ, the equation Hpump(Q)=Hsystem(Q)H_\text{pump}(Q) = H_\text{system}(Q) has no closed-form solution, so we solve it by bisection: the difference is positive at Q=0Q = 0 and negative at large QQ, and halving the bracket repeatedly pins down the root. For the slower pump we use the affinity laws: at speed ratio ss, flow scales with ss, head with s2s^2 and power with s3s^3 at corresponding points, so the curve becomes H=32s2−0.060 Q2H = 32 s^2 - 0.060\,Q^2.

python
import numpy as np

rho, nu, g = 998.0, 1.00e-6, 9.81
D, L, eps = 0.080, 150.0, 0.045e-3            # commercial steel pipe
dz, K_sum = 12.0, 8.0                         # lift (m) and sum of minor-loss coefficients
A = np.pi * D**2 / 4

def friction(Re, rr):
    if Re < 2300:
        return 64.0 / Re
    x = 8.0
    for _ in range(50):
        x = -2.0 * np.log10(rr / 3.7 + 2.51 * x / Re)
    return 1.0 / x**2

def system_head(q):                           # q in L/s -> head the pipework needs (m)
    if q <= 0:
        return dz
    V = q / 1000 / A
    f = friction(V * D / nu, eps / D)
    return dz + (f * L / D + K_sum) * V**2 / (2 * g)

def pump_head(q, s=1.0):                      # a model pump curve; s = speed / rated speed
    return 32.0 * s**2 - 0.060 * q**2

def pump_eff(q, s=1.0):                       # best efficiency 72 % at 14 L/s (rated speed)
    return 0.72 * (1 - ((q / s - 14.0) / 14.0) ** 2)

def bisect(fun, a, b, tol=1e-10):
    fa = fun(a)
    while b - a > tol:
        m = 0.5 * (a + b)
        if (fun(m) > 0) == (fa > 0):
            a, fa = m, fun(m)
        else:
            b = m
    return 0.5 * (a + b)

print("  Q (L/s)  pump H (m)  system H (m)")
for q in range(0, 24, 3):
    print(f"  {q:6.1f}  {pump_head(q):9.2f}  {system_head(q):11.2f}")

q = bisect(lambda q: pump_head(q) - system_head(q), 0.0, 23.0)
H, eta = pump_head(q), pump_eff(q)
P = rho * g * q / 1000 * H / eta
print(f"\noperating point: Q = {q:.2f} L/s, H = {H:.2f} m, efficiency {eta:.1%}, shaft power {P / 1e3:.2f} kW")

# The plant now needs only 10 L/s. Option 1: close a valve (throttle).
qt = 10.0
Ht = pump_head(qt)
Pt = rho * g * qt / 1000 * Ht / pump_eff(qt)
print(f"\nthrottled to 10 L/s: pump head {Ht:.2f} m, of which the valve burns {Ht - system_head(qt):.2f} m;"
      f" power {Pt / 1e3:.2f} kW")
# Option 2: slow the pump down until its curve passes through the system curve at 10 L/s.
s = np.sqrt((system_head(qt) + 0.060 * qt**2) / 32.0)
Ps = rho * g * qt / 1000 * system_head(qt) / pump_eff(qt, s)
print(f"variable speed: {s:.3f} x rated speed, head {system_head(qt):.2f} m, power {Ps / 1e3:.2f} kW"
      f" ({1 - Ps / Pt:.0%} less)")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The curves cross at 11.6 L/s and 24.0 m of head, near the pump's best efficiency (70%), and the pump draws 3.9 kW. To get 10 L/s with a valve, the pump must climb back up its curve to 26.0 m, and the valve throws away 5.0 m of it: the power hardly changes (3.85 kW). Slowing the pump to 92% of its rated speed instead delivers the same 10 L/s at the 21.0 m the system actually needs, for 3.0 kW, 22% less. That saving is why so many pumps and fans now run from variable-speed drives, the power electronics of power electronics feeding the induction and permanent-magnet motors of transformers and machines.

operating point (11.6 L/s, 24.0 m)10 L/s, 21.0 m at 92% speedpump, rated speedpump, 92% speedsystem curveflow rate Q (L/s)head H (m)0102030406121824
Pump and system curves for the worked example. The rated pump curve (solid) falls as flow rises; the system curve starts at the 12 m static lift and rises roughly with Q². They cross at the operating point, 11.6 L/s at 24.0 m. Slowing the pump to 92% of its rated speed (dashed) moves the crossing to 10 L/s at 21.0 m without wasting head in a valve.

A pump also needs enough pressure at its inlet. If the pressure in the impeller falls to the liquid's vapour pressure (2.3 kPa for water at 20 °C), vapour bubbles form and collapse violently further on: cavitation, which sounds like gravel, wrecks performance and pits the metal. Makers publish the net positive suction head required (NPSHr), and the designer makes sure the system provides more, with a margin. Nor should a pump run for long against a closed valve, where all its power heats a small volume of trapped liquid; installations have minimum-flow lines and controls for this reason.

ExplanationH

Boundary layers and drag, in outline

Outside pipes, the same no-slip condition shapes the flow around every object. In 1904 Ludwig Prandtl explained why fluids with tiny viscosity still feel friction: at high Reynolds number the viscous effects are confined to a thin boundary layer next to the surface, outside which the flow is nearly ideal and Bernoulli works. On a flat plate the laminar boundary layer grows as the square root of the distance xx from the leading edge, with thickness about 5x/Rex5x/\sqrt{\mathrm{Re}_x}, where Rex=Vx/ν\mathrm{Re}_x = Vx/\nu; for air at 10 m/s, that is about 4 mm at 0.5 m along the plate. Near Rex≈5×105\mathrm{Re}_x \approx 5 \times 10^5 it usually becomes turbulent, thicker and with more friction.

If the outer flow slows down (as it does around the back of a cylinder, where the pressure rises again), the slow fluid deep in the boundary layer can be brought to a halt and pushed backwards. The boundary layer separates from the surface, leaving a broad, eddying wake at low pressure. The force a body feels in a stream is the drag, the sum of skin friction (shear stress on the surface) and pressure or form drag (the low pressure of the wake pulling backwards). It is written with a drag coefficient:

FD=CD 12ρV2A,F_D = C_D\,\tfrac12\rho V^2 A,

with AA normally the frontal area. Streamlined bodies avoid separation and have CDC_D below 0.1; a sphere in the usual range has CD≈0.4C_D \approx 0.4–0.5; most cars of 2026 are between about 0.2 and 0.35; an upright cyclist is around 0.9 to 1.1 on a frontal area of about 0.5 m². At about Re=3×105\mathrm{Re} = 3 \times 10^5 a smooth sphere's boundary layer turns turbulent before it separates; the more energetic turbulent layer clings on further round the back, the wake narrows, and CDC_D falls by a factor of three or more: the drag crisis. The dimples on a golf ball trip the boundary layer early, so the ball enjoys this low drag at the speeds it actually flies.

At the other extreme, for Re<1\mathrm{Re} \lt 1, George Stokes found in 1851 that a sphere of diameter dd feels FD=3πμVdF_D = 3\pi\mu V d, that is, CD=24/ReC_D = 24/\mathrm{Re}: the law of fine dust, mist and silt. Between the regimes engineers use fitted correlations. The program below finds how fast quartz sand settles in still water, balancing weight minus buoyancy against drag with the Schiller–Naumann correlation CD=(24/Re)(1+0.15 Re0.687)C_D = (24/\mathrm{Re})(1 + 0.15\,\mathrm{Re}^{0.687}) up to Re=1000\mathrm{Re} = 1000 and CD≈0.44C_D \approx 0.44 beyond: the calculation behind settling tanks in water treatment.

python
import numpy as np

g = 9.81
rho, mu = 998.0, 1.00e-3                      # water at 20 C
rho_p = 2650.0                                # quartz sand

def drag_coefficient(Re):
    """Sphere drag: Schiller-Naumann up to Re = 1000, then roughly constant."""
    return 24.0 / Re * (1 + 0.15 * Re**0.687) if Re < 1000 else 0.44

def settling_speed(d):
    V = g * d**2 * (rho_p - rho) / (18 * mu)  # start from Stokes' law
    for _ in range(200):
        Re = rho * V * d / mu
        V_new = np.sqrt(4 * g * d * (rho_p - rho) / (3 * rho * drag_coefficient(Re)))
        if abs(V_new - V) < 1e-12:
            break
        V = 0.5 * (V + V_new)                 # damped fixed-point iteration
    return V_new, rho * V_new * d / mu

sig3 = lambda v: f"{v:#.3g}".rstrip(".")       # three significant figures
print(" d (mm)   Re        V (mm/s)   Stokes V (mm/s)   time to sink 1 m")
for d_mm in [0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 5.0]:
    d = d_mm / 1000
    V, Re = settling_speed(d)
    Vs = g * d**2 * (rho_p - rho) / (18 * mu)
    t = 1.0 / V
    when = f"{t / 3600:6.1f} h" if t > 3600 else (f"{t / 60:6.1f} min" if t > 60 else f"{t:6.1f} s")
    print(f" {d_mm:5.2f}  {Re:8.3g}  {sig3(V * 1000):>9}  {sig3(Vs * 1000):>14}      {when}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Fine silt of 0.01 mm sinks at less than 0.1 mm/s and takes about three hours to settle through a metre of water, exactly as Stokes' law predicts. By 0.1 mm the Reynolds number is near 1 and Stokes is already 13% too fast; for a 1 mm grain it is six times too fast, and for 5 mm gravel, at Re≈2500\mathrm{Re} \approx 2500, it would be wrong by a factor of 45. One formula, applied outside its range, would have ruined a settling-tank design.

ExplanationH

Pipes and pumps in professional practice

Piping is designed by codes as much as by equations. Pressure piping in many countries follows the ASME B31 codes (B31.1 for power piping, B31.3 for process piping) or, in Europe, EN 13480 under the Pressure Equipment Directive, which set wall thicknesses, materials, supports, inspection and pressure testing. First sizing uses rules of thumb (pumped water lines are commonly sized for about 1–3 m/s, balancing pipe cost against energy, noise and erosion), checked with the calculations of this lesson. Real systems are networks: a city's water supply has thousands of pipes, loops, tanks, pumps and valves, analysed by solvers that apply mass conservation at every junction and the energy equation round every loop, solved by Newton's method. The best-known open-source tool is the US Environmental Protection Agency's EPANET; commercial packages build on similar methods.

Transients matter as much as steady flow. Close a valve suddenly and the moving column of liquid must stop: a pressure wave runs back along the pipe at the speed of sound in the pipe-and-liquid system (around 1000–1400 m/s in a water main), raising the pressure by roughly ρaΔV\rho a \Delta V, Joukowsky's estimate. Stopping water flowing at 2 m/s can add 20 bar or more. This water hammer bursts pipes, which is why valves on large lines close slowly, pumps are stopped gradually, and surge tanks and relief valves protect long pipelines. Any system containing fluid under pressure is designed, tested and inspected under pressure-equipment rules, with relief devices sized for the worst case, and only trained people work on it.

Pump selection follows the method above: plot the system curve, choose a pump whose best-efficiency point sits near the duty, check NPSH, and consider how the duty varies over a year; the Hydraulic Institute and pump makers publish guidance and selection software. Pumps that move water between reservoirs also store energy for electricity grids (see renewables and storage in Lumenary EE and wind and water turbines later in this course). And when the geometry is too complicated for loss coefficients (a new valve, a pump inlet, an artery with a stent) engineers turn to computational fluid dynamics, and then test the result in a flow rig.

History & global context

A short history

The German engineer Gotthilf Hagen measured water flow in brass tubes in 1839, and the French physician Jean Léonard Marie Poiseuille, studying how blood moves in capillaries, found the same fourth-power law independently in experiments published in the early 1840s. Julius Weisbach in Saxony proposed the form of the friction law in 1845, and Henry Darcy, who had built Dijon's clean water supply, confirmed it with careful measurements on pipes of many materials in 1857. Osborne Reynolds's dye experiment of 1883 identified the two regimes and the number that separates them. Prandtl's boundary layer of 1904 reconciled ideal-flow theory with real drag, and his student Johann Nikuradse glued graded sand inside pipes to measure roughness in the early 1930s. Colebrook combined the smooth and rough laws in 1939, and Moody's chart of 1944 put them on every engineer's desk until computers made iteration free.

Common misconceptionsH

Common misconceptions

Misconception

"The pressure falls along a pipe because the fluid speeds up." In a pipe of constant bore, continuity keeps the mean speed constant. The pressure falls because friction turns mechanical energy into internal energy; the pressure gradient is exactly what balances the wall shear.

Misconception

"Double the flow, double the pressure drop." Only in laminar flow. In turbulent flow the pressure drop rises roughly as Q1.75Q^{1.75} in smooth pipes and as Q2Q^2 in fully rough ones, so doubling the flow multiplies it by about 3.4 to 4, and the pumping power by about 7 to 8.

Misconception

"A bigger pump gives more flow." The flow is set by where the pump curve meets the system curve. A pump much bigger than the duty ends up throttled by a valve, wasting energy, or running far from its best efficiency, where it wears faster. Matching the pump to the system is the job.

Misconception

"A perfectly smooth pipe has no friction." Even a smooth pipe has the viscous shear of the no-slip condition: its friction factor is 64/Re64/\mathrm{Re} if laminar and given by the smooth-pipe limit of Colebrook if turbulent. Roughness adds to friction only once the bumps poke out of the viscous layer at the wall.

Exercises

Exercises

1. Viscous flow and pipes

Warm-upHuman

Water at 20 °C (ν=1.0×10−6\nu = 1.0 \times 10^{-6} m²/s) flows at a mean speed of 0.8 m/s through a pipe of internal diameter 25 mm. What is the Reynolds number? (Then decide for yourself whether the flow is laminar or turbulent.)

2. Viscous flow and pipes

CoreHuman

A hydraulic test rig pushes oil (ρ=870\rho = 870 kg/m³, μ=0.10\mu = 0.10 Pa·s) through a straight tube of internal diameter 10 mm and length 2.0 m at 2.0 mL/s. Assuming fully developed laminar flow, what pressure difference is needed, in pascals, to the nearest 10 Pa? Check that the flow really is laminar.

3. Viscous flow and pipes

CoreHuman

The flow rate through a long, straight pipe is doubled, with the fluid unchanged. Which statements are true? Select all that apply.

Choose all that apply

4. Viscous flow and pipes

CoreHuman

Water (ρ=1000\rho = 1000 kg/m³) flows at 2.0 m/s through 50 m of 100 mm pipe with friction factor f=0.020f = 0.020. The fittings on the line (entrance, bends, two valves and the exit) have loss coefficients adding up to ∑K=5.0\sum K = 5.0. The pipe is level and has the same diameter throughout. What is the total pressure drop, in kPa?

5. Viscous flow and pipes

CoreHuman

A centrifugal pump running at 1450 rpm delivers 20 L/s at a head of 30 m and draws 8.0 kW. A variable-speed drive slows it to 1160 rpm. Using the affinity laws (and assuming the efficiency at the corresponding point is unchanged), what shaft power does it now draw, in watts?

6. Viscous flow and pipes

CoreHuman

A dimpled golf ball flies further than an otherwise identical smooth ball. What is the main reason?

Choose one

7. Viscous flow and pipes

CoreComputational

Write colebrook(Re, rel_rough) that returns the Darcy friction factor ff for turbulent pipe flow by solving the Colebrook equation

1f=−2log⁡10 ⁣(ε/D3.7+2.51Re f)\frac{1}{\sqrt f} = -2\log_{10}\!\left(\frac{\varepsilon/D}{3.7} + \frac{2.51}{\mathrm{Re}\,\sqrt f}\right)

numerically, to at least eight significant figures. The tests check that your ff satisfies the equation for smooth and rough pipes from Re=4000\mathrm{Re} = 4000 to 10810^8, and reproduce the lesson's worked example.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

8. Viscous flow and pipes

StretchComputational

Write operating_point(pump, system, q_max, tol=1e-9) that finds the flow rate where a pump curve meets a system curve. pump and system are functions of the flow rate qq that return heads; assume that if a crossing exists in [0,qmax][0, q_\text{max}], then pump(q)−system(q)\text{pump}(q) - \text{system}(q) is positive before it and negative after it. Use bisection and return the crossing as a float, to within tol. If the pump cannot even overcome the system at zero flow (pump(0)≤system(0)\text{pump}(0) \le \text{system}(0), a pump too weak for the static lift), return None.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources

  • Turbulent flow in pipes, with particular reference to the transition region between the smooth and rough pipe laws · C. F. Colebrook · 1939Journal of the Institution of Civil Engineers 11, 133–156. The implicit friction-factor equation used throughout the industry.
  • Friction factors for pipe flow · L. F. Moody · 1944Transactions of the ASME 66, 671–684. The paper that turned Colebrook's equation into the chart on every hydraulic engineer's wall.
  • EPANET · United States Environmental Protection AgencyFree, open-source software for modelling water distribution networks — pipes, pumps, valves and tanks — used by utilities worldwide.
  • Hydraulic InstituteThe North American pump industry association; publishes pump standards and guidance on pump selection and system efficiency.
  • ASMEPublishes the B31 piping codes (power piping, process piping and others) used for pipework design in many countries.
  • ISO, the International Organization for StandardizationPublishes ISO 9906, the international acceptance test for the hydraulic performance of rotodynamic pumps.
  • MIT OpenCourseWare 2.25, Advanced Fluid Mechanics · Massachusetts Institute of TechnologyLecture notes and problem sets on viscous flow, boundary layers and pipe flow at graduate level.