What the specifications say and what the browsers actually do are two different documents. These 45 divergences are the ones that cost production teams their weekends — browser asymmetries, silent failures, and the sharp edges that only appear on real networks.
Filter by tag
Every quirk carries a tag (contested: 3, field: 15, quirk: 9, trap: 18). The filter itself requires JavaScript; all 45 quirks are listed below in course order.
Codes give minimum sewer slopes as a table against diameter — 0.40% for 8 in, 0.22% for 12 in, and so on. Those numbers are derived: they are the slopes that produce roughly 2 ft/s (0.6 m/s) at full or half-full flow through Manning. Reading the table without knowing that hides the failure mode. A pipe laid at exactly minimum slope but running at a tenth of its design flow is not self-cleansing, because the velocity that matters is the velocity today, not the velocity at build-out.
On any new sewer serving an area that will fill in over decades, check the velocity at initial flow as well as at design flow. If initial velocity is under about 0.3 m/s you have a deposition problem for years, and the answer is usually a steeper laid slope or a smaller initial pipe, not a bigger one.
A compact station sketch can hide the lift, laydown and isolation space needed to remove a pump or valve. Hydraulic neatness is not operability if routine work requires dismantling the header.
Trace both the water path and the removal path. Show what must be isolated, lifted and kept in service during each credible intervention.
The two liquids constrain velocity from opposite directions, which is why the same force main sizing spreadsheet gives you wrong answers if you carry it between projects. In a sanitary force main the binding constraint is a minimum velocity, around 0.6–0.9 m/s, or grit and rag settle out and the main slowly chokes. In a potable transmission main the binding constraint is usually a maximum, both for energy cost and because surge pressure on shutdown scales directly with velocity.
The consequence: a sewage force main is often deliberately undersized relative to least-cost diameter, and a water main deliberately oversized. Both are correct. Neither is a mistake to be corrected by the other project's rule of thumb.
Write the governing velocity constraint at the top of the sizing sheet, with its direction, before you pick a diameter. If the sheet does not say whether the constraint is a floor or a ceiling, it is not a design, it is arithmetic.
The Harmon peaking factor is 1 + 14/(4 + √P) with P in thousands.
People quote it as "about 2 for a decent-sized town", which is loose enough to hide
that at P = 100 the expression is 1 + 14/(4 + 10) = 2.0 exactly.
Below 100,000 the factor is above 2, and it climbs steeply: at 10,000
people it is 2.94, at 1,000 it is 3.80. A designer who remembers "peaking factor
is about 2" and applies it to a 2,000-person lift station has undersized the peak
by nearly 40%.
Babbitt gives 5/P^0.2 and disagrees with Harmon by a wide margin at
small populations — 3.45 against Harmon's 3.80 at 1,000. Neither is "the" answer.
Never carry a peaking factor as a remembered number. Compute it from the served population, and for anything under about 20,000 people compute both Harmon and Babbitt and state which you used and why. For very small stations both formulas are extrapolations and metered data from a comparable station beats either.
A station with three pumps each rated 50 L/s is not a 150 L/s station. With the largest unit out of service — the standard reliability criterion — it is a 100 L/s station, and even that overstates it, because two pumps running in parallel on the same force main do not deliver twice one pump's flow. The system curve rises as flow increases, so the two-pump operating point sits well below 2 × Q₁, often at 1.3–1.6 × Q₁.
So the honest firm capacity of that station might be near 65 L/s, not 150. That is a factor of more than two between the number on the nameplate sum and the number you can defend to a reviewer.
State firm capacity as the intersection of the N−1 pump combination curve with the system curve, not as a sum of nameplate ratings. §7.3 does this properly; the arithmetic shortcut is always optimistic.
The same route produces different pressures at different wet-well levels, flows, roughness states and pump combinations. A clean line with no condition block can be correct and still mislead a reviewer.
Put flow, boundary levels, pipe condition, pump state and datum beside every profile trace.
Mixing a reservoir surface at one end with a pump flange at the other silently imports pressure and velocity terms that a surface-to-surface calculation would cancel.
Draw the two boundary sections first, then cross out only the terms whose physical value is actually zero or equal.
Profiles drafted as one smooth slope distribute valve and fitting losses along the pipe. That may look tidy, but it hides where pressure is actually spent and makes later changes hard to audit.
Show distributed loss as slope and material local losses as steps, at the station where each element sits.
Wet wells and receiving reservoirs move. Selecting on one convenient pair of levels turns a boundary envelope into a false constant.
Pair the credible suction and discharge levels into the cases required by the project, and state which combinations are simultaneous.
The conversion everyone memorises, 1 psi = 2.31 ft, embeds a
density: it is 144 / 62.37, water at about 60 °F. It is a
specific-weight conversion, not a unit conversion, so it drifts with
temperature and with specific gravity. For water the drift is small enough to
ignore in most municipal work (about 1% over 4–40 °C). For anything else it is not.
The trap bites on sludge and on chemical feed. Pumping 6% sludge at SG 1.03, or sodium hydroxide at SG 1.5, and converting pressure to head with 2.31 gives an error that goes the wrong way: the pump makes a fixed head, so a denser liquid means a higher discharge pressure and a higher power draw, both of which the 2.31 shortcut hides.
Keep head in metres or feet of the pumped liquid throughout, and convert to pressure only at the end using the actual specific weight. When you hand a pressure rating to a mechanical engineer, state the SG you used.
Textbooks drop the velocity-head term from TDH with a wave of the hand, and
for a long force main that is fair: at 1.2 m/s, V²/2g is 0.073 m
against maybe 30 m of TDH. But the term scales with the square of velocity while
friction over a short main scales with length, so on a short, fast
discharge the ratio inverts. At 3 m/s the velocity head is 0.46 m; on a station
lifting 3 m into an adjacent gravity manhole, that single dropped term is 15% of
the total.
Compute V²/2g once and compare it to your loss total before
deciding whether to drop it. It costs one line. Dropping it silently on a low-lift,
high-velocity station is one of the more common ways a pump ends up running off the
right-hand end of its curve.
A calculated Reynolds number near transition does not justify interpolating confidently between laminar and turbulent formulae. Disturbance, fittings and surface condition can change the regime.
Expose the uncertainty and check both applicable bounds, or obtain service-specific evidence for the low-flow condition.
Published roughness values describe materials and conditions, not the installed main after joints, deposits, lining defects and years of service.
Carry a documented condition range and test whether the pump selection survives both the smooth and resistant cases.
Colebrook-White is implicit and needs iteration; Swamee-Jain is the explicit
approximation everyone actually uses. The usual claim is "within 1%", which is
true in the region that matters and false at the edges. Measured across the error
surface, as a fraction of the Colebrook root: for Re ≥ 10⁵ and
relative roughness from 10⁻⁵ to 10⁻² — the whole municipal design range — the
worst disagreement is 0.81% in friction factor, at
Re = 10⁵, ε/D = 0.005. Extend down to Re = 5×10³, just
above the laminar transition, and the worst case grows to 2.83%,
at the double corner of low Re and very rough pipe.
The denominator matters and is worth stating: divide the same disagreement by the approximation instead of by the root and you get 2.75%, not 2.83%. Error is measured against what is true, not against the thing being tested. This course's own verifier had it the wrong way round until an author computing the figure independently disagreed with it by exactly that much.
A 3% error in f is a 3% error in hf, which is usually far inside the uncertainty of your C-factor or roughness estimate. The reason to know the number is not to correct for it, it is to know that the approximation is not where your error is coming from.
Use Swamee-Jain for design and stop apologising for it. If you are near the laminar transition — thick sludge, chemical feed lines, very low flow — iterate Colebrook properly, because you are also near the point where the whole turbulent-flow framework stops applying.
Hazen-Williams is not dimensionally homogeneous, so the leading constant carries the unit system:
SI: hf = 10.67 · L · Q^1.852 / (C^1.852 · D^4.871)
(Q in m³/s, D and L in m)
US: hf = 4.73 · L · Q^1.852 / (C^1.852 · D^4.871)
(Q in cfs, D and L in ft)
The C value is the same number in both. That is what makes the error so durable: a spreadsheet with the wrong constant still looks right, still uses a familiar C = 130, and produces a head loss wrong by a factor of about 2.3. Nothing in the output is dimensionally absurd, so nothing catches it.
Manning has the identical problem with 1.0 (SI) against 1.486 (US), where
1.486 is just 3.2808^(1/3).
Verify any borrowed Hazen-Williams sheet against one hand-worked case in your units before trusting it. Better: work in SI internally and convert only for display, which is what this course's calculation core does — and it has a cross-unit identity test that would fail loudly if the two constants ever disagreed.
Darcy-Weisbach responds to temperature through the Reynolds number. Hazen-Williams has no viscosity anywhere in it — it was fitted to water at around 60 °F in the transitional turbulent range. Pump 4 °C water in February and Darcy-Weisbach will show more friction than in August; Hazen-Williams shows exactly the same number, because it structurally cannot represent the difference.
It is also fitted, not derived, so it drifts outside its fitting range: very smooth pipe, very high velocity, or very small diameter all push it off. Its exponent on velocity is 1.852, which is close to the ~1.9 of real transitional flow but wrong for fully rough flow, where the true exponent is 2.
Use Hazen-Williams for water distribution networks at ordinary temperatures, where its C-factor library is genuinely useful field knowledge. Use Darcy-Weisbach for force mains, for anything cold, for anything that is not water, and any time you need to defend the number rather than just produce it.
"Use C = 100 for the design life" is inherited from unlined cast iron, which tuberculates badly — a main that started near C = 130 can fall below 80 in aggressive water. PVC, HDPE and cement-mortar-lined ductile iron do not behave that way; field measurements on lined and plastic pipe generally show little change over decades, and published guidance treats them as effectively stable.
So the aging allowance is a material question, not a universal one. Applying the cast-iron allowance to a new HDPE force main means designing for roughly 60% more friction than it will ever have, which pushes you to a larger diameter — which, in a sewage force main, then fails the minimum velocity check. The conservative move creates a real failure mode.
Check both ends of the envelope and require the design to survive both: new and smooth (highest flow, highest velocity, worst surge, pump running out on its curve) and old and rough (highest head, lowest flow, check the velocity floor). A single design C-factor answers neither question.
Look up the K for a standard 90° elbow and you will find 0.3, 0.5, 0.75 and 0.9 in four reputable places. The spread is real and has causes: some values are for threaded fittings and some for long-radius welded bends, some are single-K and some come from the two-K or three-K methods that make K a function of Reynolds number and nominal diameter, and some tabulate an equivalent length in pipe diameters instead.
Treating a single-K table as precise is the error. In a station where fittings contribute 4 m of a 35 m TDH, a ±40% spread on K is ±1.6 m — small. In a low-lift station where fittings are 4 m of an 8 m TDH, the same spread moves the operating point enough to change the pump you select.
Compute the fitting total as a range, not a value, and check whether the pump selection survives both ends. When it does not, the fix is to reduce the fitting count and straighten the pipework — not to find a table with friendlier numbers.
An equivalent length is derived from a K value through the straight-pipe friction factor. Copying L/D from one source while using another friction basis can make the conversion look universal when it is not.
Record the source and basis, then use either K or equivalent length once—not both.
A low-K bellmouth assumes a reasonably uniform approach. Swirl, a nearby wall or an asymmetric wet-well inflow can dominate the elegant entrance geometry.
Review the approach volume with the intake detail; treat the tabulated K as one part of the evidence.
A valve shown open may have a disc in the stream, a failed actuator, an incorrect travel stop or a check element hovering partly open. Its real loss can exceed the schedule.
Connect the assumed position to the selected valve, actuator and commissioning test, and provide pressure taps where the consequence matters.
Three identical pumps on a common discharge header are not three identical systems. The pump furthest from the header outlet pushes its flow through more header length and more tees, so it sees a higher system resistance and sits further left on its curve. It delivers less, runs at a different point relative to best efficiency, and wears differently.
With a symmetric header the effect is small. With a header that dead-ends at the far pump, or where the tees are all in one direction, the imbalance between first and last pump can be over 10% of flow — enough that duty rotation intended to equalise wear does not.
Model each pump's branch to the header outlet separately when you have three or more units, and prefer a symmetric header geometry. If the layout is fixed and asymmetric, adjust the run-hour targets rather than pretending the pumps are interchangeable.
Vertical exaggeration helps reveal grade changes but can visually overstate pressure slopes and clearances. A viewer may read geometry from a diagram that was intended only to show energy.
State both scales and put decisive elevations and pressures in text, not only in the drawn gap between lines.
Using the inside crown as the cover reference is conservative by the wall thickness, but only if that simplification is stated. Mixing inside and outside diameters elsewhere can reverse the error.
Label invert, inside crown and outside crown separately, and name the surface used for every cover check.
Values such as 9+500 and 10+000 can be ordered lexically instead of numerically, making a profile double back while every row formula still appears valid.
Store chainage as a number in one base unit and format the station notation only for display.
A positive steady pressure can disappear during trip, restart or check-valve closure. Conversely, an air valve can change the transient that justified its location.
Use the steady profile to identify vulnerable points, then carry those points into the transient and air-management review.
A second sheet that references every cell in the first reproduces the same boundary and unit mistakes with a different colour scheme.
Rebuild at least the governing cases from the source geometry and assumptions, with independent spot calculations.
Pump head is the change in total head between suction and discharge reference sections. One gauge omits suction pressure, elevation difference and velocity-head change.
Define both measurement sections and convert their pressure, elevation and velocity terms to a common head basis.
Curve data belongs to a stated impeller, speed, liquid, test tolerance and sometimes correction basis. A traced image with those conditions cropped off is not a specification.
Keep the complete certified submittal and identify the exact curve revision used in the selection.
A nominal sphere test describes one geometric clearance. Flexible wipes and stringy material interact with leading edges, clearances and recirculation differently.
Pair the stated passage with service references, impeller geometry and the upstream solids strategy.
The same pump has a specific speed of about 2,000 in US practice (rpm, gpm, ft) and about 39 in SI practice (rpm, m³/s, m) — the ratio is roughly 51.6. Despite being called a number, Ns is not dimensionless as normally written, so a value quoted without its unit system is meaningless.
The practical consequence is a mis-selection. Radial, mixed-flow and axial regimes are separated by Ns thresholds, and those thresholds are quoted in whichever convention the source uses. Reading an SI value of 39 against US thresholds puts a perfectly good mixed-flow pump in the radial box.
Always write the unit system next to a specific speed. When comparing against
a published chart, convert your value into the chart's convention first, and
remember suction specific speed Nss is a separate quantity
with the same problem.
Two errors travel together in NPSH work. The first is using gauge pressure: NPSHa is built from absolute pressure, so atmospheric pressure is a term, and it falls with elevation — about 0.9 m of water head lost per 1,000 m of altitude. A station at 1,500 m has roughly 1.3 m less NPSHa than the same station at sea level, for free.
The second is treating vapour pressure as a constant. It rises steeply and non-linearly with temperature: about 0.17 m of head at 15 °C, 0.32 m at 25 °C, 0.76 m at 40 °C. Warm sewage in a summer afternoon, or a recirculation line that has been running against a closed valve, moves this term by more than the margin many designs carry.
Compute NPSHa at the worst combination: highest liquid temperature, lowest wet well level, site altitude, and the suction losses at maximum flow — not at design flow. Then require a margin over NPSHr, typically 0.6 m or 1.3 × NPSHr, whichever is greater. The bare curve crossing is not a design.
Q ∝ N, H ∝ N², P ∝ N³ are correct, and they are the single most misapplied set of relations in pumping. They map one point on the pump curve to a corresponding point at a new speed, along a parabola through the origin. They say nothing about where the pump will actually operate, because that is set by the system curve.
On a system that is pure friction — no static lift — the system curve is also a parabola through the origin, the two coincide, and the affinity laws do predict the operating point. On a system with static lift they do not. Take a station with 20 m of static lift and 10 m of friction at design flow: drop the speed to 70% and the pump can no longer make 20 m at all. Flow does not fall to 70%, it falls to zero. The cube law on power, which is the entire energy-saving argument for a VFD, evaporates along with it.
Never quote VFD savings from the cube law on a station with meaningful static lift. Draw the system curve, walk the pump curve down in speed steps, and read the real operating points off the intersections — then compute power at each. §7.5 does this. On a high-static station the honest answer is often that a VFD saves very little and its real value is soft-start and level control.
Levels, aging, valve position and pump/header route create a family of curves. Calling the nominal trace “the system” hides the cases that often govern capacity or runout.
Plot and label the credible envelope; keep faults such as blockage separate from normal operating cases.
Unusual pump or system shapes can create more than one crossing, while a reduced-speed curve may fail to clear static lift at all. A solver seeded near the expected duty can miss both facts.
Scan the full allowed flow range for every sign change and check endpoints before accepting a numerical root.
Parallel operation adds flow at equal head — you add the curves horizontally. But adding flow raises system friction as the square of flow, so the combined operating point climbs to a higher head where each pump delivers less than it did alone. The steeper the system curve, the worse the return.
On a friction-dominated force main, two identical pumps commonly deliver 1.3–1.5 × the single-pump flow, and three deliver barely more than two. On a static-dominated system the gain is much closer to double, because the system curve is nearly flat. Same pumps, same station, completely different answer — and the difference is a property of the pipe, not the pumps.
Size for firm capacity from the actual N−1 combined curve intersection. If parallel gain is poor, the fix is a larger force main or a different pump, not more pumps — a third pump on a steep system can add almost nothing while adding a full set of valves, controls and failure modes.
Efficiency loss is the visible penalty for operating away from BEP and the least important one. Far enough to the left of the pump's preferred operating region, suction and discharge recirculation can set up inside the impeller, and the pressure pulsations that follow load the bearings and seals cyclically. The failure shows up as bearing life, seal life and cracked impellers, not only as an energy bill. The boundary is pump-specific; it is not a universal percentage of BEP flow.
This is why an oversized pump — the conservative choice — is a durability problem. A pump selected for a build-out flow it will not see for fifteen years spends those fifteen years throttled left of BEP, and the reliability cost arrives long before the capacity is needed.
Keep continuous operation inside the manufacturer-supported preferred operating region and check the allowable operating region for every transient or intermittent duty. Check initial flows as well as design flows. Staging smaller units, or a VFD where the static head permits it, beats one large pump that is right only at the end of the planning horizon.
Reducing speed moves the pump curve, but static lift, minimum transport velocity, motor cooling and control deadbands do not scale with it.
Use affinity laws to generate candidate duty, then re-intersect the real system and recheck every service constraint.
A lifecycle model can reward low velocity, long detention or an operating point outside the preferred region if those consequences were never given a cost or constraint.
Screen alternatives for hydraulic and operational acceptability before ranking the survivors by present value.
Intuition says the shortest cycles happen at peak inflow. The opposite is
true. Cycle time for a single fixed-speed pump is
t = V/(Qp − Qi) + V/Qi, and that
expression is minimised when inflow is exactly half of pump capacity. At
very low inflow the fill time is long; at inflow approaching pump capacity the
pump-down time is long. In between, both are short at once.
So the starts-per-hour limit is not tested at design peak. It is tested at Qi = Qp/2, which for a station sized on peak flow is an entirely ordinary Tuesday morning.
Size the active wet well volume from the cycle time at Qi = Qp/2 against the motor's permitted starts per hour (typically 6–10 for larger units). Checking only at peak flow will pass a wet well that burns out a motor.
Intercepted material needs lifting, drainage, storage, odour control and disposal. A screen that protects the pump but cannot be maintained during peak inflow creates a different station failure.
Draw the screenings path and the bypass/maintenance condition with the same care as the water path.
Minimum submergence is set by vortex formation, not by cavitation or by priming. Well before the water surface reaches the intake, a free-surface vortex can form and draw an air core down into the suction. Even a few percent entrained air collapses head, unbalances the impeller and makes the pump noisy and short-lived — while the intake is still comfortably underwater.
Required submergence grows with intake velocity, and it is a function of the approach geometry as much as of depth: an off-centre inflow, an asymmetric bench, or a wall too close to the suction bell will all produce vortices at depths that a formula-derived minimum said were fine.
Take minimum submergence from the intake design standard for the geometry you actually have (ANSI/HI 9.8 is the reference for pump intake design), not from a single formula, and treat the approach layout as part of the calculation. For large or unusual stations, a physical or CFD model study is cheaper than rebuilding a wet well.
The check valve exists so that flow does not reverse through a stopped pump. It does that by slamming shut, and the slam is a water hammer event generated inside your own station. A heavy swing disc closing on reversing flow can produce a pressure spike comparable to the surge you carefully designed the force main to avoid.
The trade is real and there is no free option: fast closure means a hard slam, slow closure means reverse flow through the pump and possible reverse rotation. Valve type is the lever — spring-assisted, lever-and-weight, tilting disc, or a cushioned closure — and each buys a different part of the curve.
On any force main long enough for surge to matter, select the check valve as part of the surge analysis rather than as a catalogue item after it. The valve's closure characteristic is an input to the transient model, not a detail.
ΔH = a·ΔV/g is taught as the maximum surge, and for a
frictionless line with closure faster than the pipe period 2L/a it is.
Two caveats change how you use it.
First, it only applies to rapid closure. Close slower than 2L/a and the returning pressure wave relieves the spike, so the real surge is less — sometimes much less. Second, in long lines where friction head is a large fraction of the total, "line packing" can drive the pressure at the valve above the Joukowsky value as the flow decelerates against the frictional gradient. Treating Joukowsky as an unconditional ceiling in that case is optimistic.
Neither caveat is a reason to skip the estimate. Joukowsky is the right first calculation — it tells you in one line whether surge is a footnote or a governing load case.
Use Joukowsky to screen. If the result is a meaningful fraction of the pipe's pressure rating, or if the line is long relative to its wave period, escalate to a proper transient analysis rather than adding a safety factor to a formula that is outside its assumptions.
A generator sized only from motor nameplates can miss starting method, simultaneous auxiliaries, fuel duration and the storage consumed before power is restored.
Join the electrical load sequence to wet-well storage, alarm response time and the required pumping state after an outage.