loading...
running...
How to use
Pick a preset or build a column: choose 1 to 3 layers and give each a thickness, grain size and porosity. Water flows upward from the reservoir on the left, through the sand, and over the overflow rim into the measuring cylinder. The four tubes on the board show the water pressure inside the column: their levels step down across the layers, and the steepest step marks the tightest layer. Press Start test to collect water (constant head) or to let the standpipe drain (falling head); the time speed-up is automatic so even silt finishes in seconds. Inject dye sends a blue front up through the layers at the seepage velocity.
Try: (1) Uniform medium sand, double the head and watch Q double; (2) Silt sandwich, then make the silt layer thinner and see how fast the flow recovers; (3) Loose vs dense; (4) Cold water against 40 °C; (5) Quicksand, then reduce the head just below the critical value. Drag to orbit, wheel to zoom, Space pauses, R resets.
The acrylic column, sand layers, porous stones, overflow collar, reservoir, manometer board with ruler, valve, measuring cylinder, retort stand and the three jars of reference sand were modelled in Blender with Cycles-baked sand, metal and wood textures. Three.js r160 moves the water levels, sand layers, dye and piezometer hoses from the same model as the 2D companion.
The science
Water does not flow through soil as it does through a pipe. It threads through the gaps between grains, and the drag of all those thin pores makes the average flow proportional to the push. A permeameter is the standard laboratory way to measure how easily a soil lets water through: a column of sand is held between two water levels and the discharge is timed. This page models that experiment. Water enters at the bottom, climbs through up to three layers of sand and spills over the rim of the column; the head difference Δh is the gap between the reservoir level and that rim.
Darcy's law. In 1856 Henry Darcy found, from the fountains of Dijon, that the discharge through a sand filter is proportional to its area and to the head loss per unit length:
Q = K A (Δh / L) = K A i, q = Q / A = K i, vseepage = q / n
K is the hydraulic conductivity (m/s), i the hydraulic gradient, q the Darcy flux (a discharge per area, not a real water speed) and n the porosity. Water really moves at the seepage velocity q/n because only the pore fraction is open. The head h = z + p/(ρ g) is the sum of elevation and pressure head; it is what the four piezometer tubes on the board measure, and it falls linearly with distance through a uniform layer.
From grains to conductivity: Kozeny-Carman. Treating the pores as a bundle of tortuous capillaries gives the permeability k (a property of the solid skeleton) and then K (which also depends on the fluid):
k = d² n³ / [180 (1 − n)²], K = k ρ g / μ, μ(T) = 2.414×10−5 · 10247.8/(T+133.15) Pa s
Grain size enters squared, so ten times finer grains cut K a hundred-fold, and a loose packing (n = 0.46) passes several times more water than a dense one (n = 0.30). Colder water is more viscous, so the same sand drains more slowly in winter. The dashed grey curve in the third graph is the empirical Hazen rule K = C d10² with C = 100 (d in cm, K in cm/s); the two agree to within a factor of about two for clean sands, which is the usual scatter of such estimates. Kozeny-Carman is reliable for fairly uniform sands and silts; it fails for clays (surface forces, fabric) and for very poorly sorted soils, and the grain size here is a single effective diameter.
Layers. For layers stacked along the flow, the same flux q passes through every one, so the head losses add: Δh = Σ q Li / Ki. The effective conductivity is the thickness-weighted harmonic mean:
Keff = L / Σ (Li / Ki), Δhi = Δh (Li/Ki) / Σ(Lj/Kj)
The harmonic mean is dominated by the tightest layer: a thin silt layer between two sands takes most of the head (watch the manometer levels step down across it) and sets the flow for the whole column. If the layers were instead side by side along the flow, the conductivity would be the arithmetic mean Σ Ki Li/L, dominated by the most permeable layer. The readout shows both; the 2D companion page lets you flip between the two geometries and see the flow net.
Two ways to measure K. Constant head: collect a volume V in time t and compute K = V L / (A Δh t). Falling head, used for tight soils, lets a standpipe of area a drain through the sample: dh/dt = −(K A / a L) h, so K = (a L / A t) ln(h0/h1). The experiment buttons run both on the model; the result is compared with the K of the stack in the check list.
Quicksand. Flow is upward here, so the water pushes against the weight of the grains. Where the head lost above a level equals the submerged weight of the sand above it, Δh = Σ (Gs − 1)(1 − n) L with Gs = 2.65, the contact stress between grains drops to zero and the sand loses its strength. The column then shakes and the top stone lifts. Darcy's law does not describe the fluidised state, so the numbers in that regime describe only the state just before failure.
Dye travel. The dye front moves at the seepage velocity in each layer, so the breakthrough time is t = Σ Li ni / q. Real fronts smear because water speeds differ from pore to pore (dispersion); the in-page Monte Carlo adds a small dispersion to confirm that the mean arrival time is unchanged.
Numerical checks run in the page. (1) A finite-volume solution of steady flow across the stack (1,200 cells, harmonic face conductances, tridiagonal solve) against the closed-form harmonic mean, with layers snapped to whole cells; (2) flux in against flux out; (3) the sum of layer head losses against the applied head; (4) a falling-head test integrated with RK4 against the logarithmic closed form, and the K recovered from it; (5) 1,500 random-walk particles advected through the layers against Σ L n / q; (6) Kozeny-Carman against Hazen for a reference medium sand, and the viscosity of water at 20 °C against 1.002 mPa s; (7) a 2D finite-volume solver (conjugate gradients) for layers across the flow against the harmonic mean, for layers along the flow against the arithmetic mean, and a mass balance for a low-conductivity lens. Your own timed run is added to the list when it finishes.
Simplifications. One effective grain size and one porosity per layer; no wall effects (real columns have faster flow along the tube wall), no air bubbles or clogging, steady saturated flow only (the manometers answer instantly), no compaction of the sand under the flow, and water properties from the temperature alone. Darcy's law is linear only for slow flow: above a grain Reynolds number of about 1 inertial losses appear (Forchheimer or Ergun terms), which this page flags but does not model. Sizes are drawn at 0.09 display units per cm, and time can be sped up so that a slow silt test finishes in seconds. Values of K for real soils span more than ten orders of magnitude, from about 10−1 m/s for clean gravel to below 10−10 m/s for clay.