Media Filtration Design: Mechanisms, Head Loss and Backwash

Is media filtration cake filtration or depth filtration?

For rapid granular filters the dominant mechanism is depth (deep-bed) filtration, not surface straining. Particles far smaller than the pore openings are captured throughout the bed depth at grain surfaces, so removal is a volumetric process governed by particle-collector collision physics rather than by a sieving cut-off. Straining and cake formation dominate only when the particle-to-grain size ratio is large (above roughly 0.15–0.2), which is the regime of pressure cloth and precoat filters, not sand beds.

The practical consequence is that a granular filter removes turbidity by transport plus attachment. The classic idealisation is Yao's single spherical collector: a grain of diameter d_c intercepts particles from the approaching flow, and the fraction it captures is the product of a transport efficiency and an attachment efficiency. This is why coagulation upstream matters so much — it fixes the attachment term. Condition the water well with the right coagulation and filtration equipment and the same bed removes an order of magnitude more particles per unit depth.

What are the particle transport mechanisms?

Three transport mechanisms carry suspended particles from the bulk flow onto a media grain, each dominating a different particle-size band:

• Interception — a particle following a streamline passes within one particle radius of the grain and touches it. Scales with (d_p/d_c)²; dominant for particles > ~1 µm.
• Sedimentation — denser particles settle across streamlines under gravity onto upward-facing grain surfaces. Scales with the gravity number (Stokes settling velocity / approach velocity); matters for dense particles > ~1 µm.
• Diffusion (Brownian) — sub-micron particles wander off streamlines by thermal motion and strike the grain. Scales with Pe^−2/3; dominant below ~1 µm.

Because interception and diffusion pull in opposite directions with particle size, the single-collector efficiency has a minimum near 1–2 µm — the hardest particles to filter. This is why very fine, poorly coagulated colloids break through first.

How does removal depend on bed depth?

Combining the collector efficiency over all grains in a clean bed gives an exponential (log-linear) removal with depth — the deep-bed filtration equation:

Two design levers follow directly. Finer media (smaller d_c) raises removal but also raises head loss and shortens runs, so there is a trade-off. And because attachment α multiplies the whole exponent, chemical pretreatment is the single most powerful control on filtrate quality. Filters are commonly specified by L/d_c ratio — typically 800–1,200 for mono-media sand and 1,000–1,500 for dual media — to guarantee enough collectors in series.

What media grading should you specify?

Granular media are graded by sieve analysis and specified by two numbers from the cumulative passing curve:

• Effective size (ES = d₁₀) — the sieve size passing 10% by mass. It largely sets both head loss and the finest particle the bed captures. Rapid sand is typically 0.45–0.65 mm; coarse dual-media anthracite 0.9–1.4 mm.
• Uniformity coefficient (UC = d₆₀/d₁₀) — the spread of grain sizes. Lower is better; specify UC ≤ 1.5 (ideally ≤ 1.4). A high UC means fines migrate to the surface after backwash, causing early head loss and short runs.

The table gives representative design ranges. Note the trend: coarser-over-finer dual media allows a higher filtration rate and stores more solids deep in the bed before head loss limits the run.

How do you calculate clean-bed head loss?

Flow through a clean packed bed is laminar (grain Reynolds number typically < 10), so head loss is given by the Carman–Kozeny equation — the low-Reynolds limit of the Ergun equation:

Head loss then develops during the run as captured solids reduce porosity. The filter is run until either the head loss reaches the available driving head (the terminal head loss, ~2.5 m in a gravity filter) or turbidity breakthrough begins — whichever comes first. Good design makes those two events coincide, so the bed is used fully but never passes solids. The moment breakthrough starts to rise is the optimum backwash point.

Worked example: filter area and clean-bed head loss

Design a rapid gravity sand filter for a plant flow of Q = 500 m³/h at a filtration rate of v = 8 m/h, sand d₁₀ = 0.55 mm, bed depth L = 0.7 m, porosity ε = 0.42, sphericity φ = 0.8, water at 15°C (μ = 1.14×10⁻³ Pa·s, ρ = 999 kg/m³).

• Total filter area: A = Q / v = 500 / 8 = 62.5 m². With 4 cells (1 offline for backwash) each cell ≈ 62.5 / 3 ≈ 21 m² at peak, e.g. four 4.0 m × 4.0 m cells (16 m² each, 64 m² total).
• Superficial velocity in SI: v = 8 m/h = 8 / 3600 = 2.22×10⁻³ m/s.
• Grouping (1−ε)²/ε³: (0.58)² / (0.42)³ = 0.3364 / 0.0741 = 4.54.
• Grain term (φ d_c)²: (0.8 × 5.5×10⁻⁴)² = (4.4×10⁻⁴)² = 1.936×10⁻⁷ m².
• Carman–Kozeny (k = 5): h_L = [5 × (1.14×10⁻³) × 4.54 × (2.22×10⁻³) × 0.7] / [999 × 9.81 × 1.936×10⁻⁷].

Numerator = 5 × 1.14×10⁻³ × 4.54 × 2.22×10⁻³ × 0.7 = 4.02×10⁻⁵. Denominator = 999 × 9.81 × 1.936×10⁻⁷ = 1.897×10⁻³. So h_L ≈ 0.021 m ≈ 21 mm of clean-bed head loss. That leaves the bulk of the ~2.5 m available head as the solids-storage budget that fills up over the run — confirming the media is coarse enough for a practical run length at 8 m/h.

How do you design the backwash?

Backwash reverses clean water (often with an air scour) up through the bed to fluidise it and carry away stored solids. The bed must expand enough to release deposits but not so much that media is lost over the weir. The key threshold is the minimum fluidisation velocity v_mf, where upward drag equals the buoyant weight of the media:

Typical backwash rates are 30–50 m/h for sand and 40–60 m/h for dual media, usually with 15–25 min air scour first. Because anthracite is lighter than sand, a dual-media bed re-stratifies coarse-over-fine after each backwash, which is exactly the grading that gives long runs. Under-washing leaves mudballs; over-washing wastes water and can classify the bed. Where a filter feeds a downstream membrane, tight turbidity control here protects that asset — see our guide to ultrafiltration membrane design and the upstream coagulation and flocculation guide that sets the attachment efficiency.

Media filtration design sequence

1. Fix the pretreatment and attachment. Set coagulant/polymer dose so the attachment efficiency alpha approaches 1; without it, no bed depth will produce a low-turbidity filtrate.
2. Choose media grading. Select effective size (d10) and uniformity coefficient (<=1.5) for the target particle size and run length; use dual media for higher rates.
3. Set the filtration rate and area. Pick a design rate (5-12 m/h sand, 10-20 m/h dual media), then A = Q / v, with one cell offline for backwash.
4. Set bed depth from L/dc. Specify depth so L/dc gives 800-1,500 collectors in series, guaranteeing the required log removal.
5. Check the head-loss budget. Compute clean-bed head loss (Carman-Kozeny) and confirm ample storage head remains below terminal head loss (~2.5 m).
6. Design the backwash. Set backwash rate at 1.3-1.5x minimum fluidisation velocity for 20-40% expansion, with air scour; confirm no media carry-over.