Minor Losses in Pipes: K-Factor vs Equivalent Length

"Minor losses" is the worst-named quantity in pipe hydraulics. The term is a category label inherited from the early textbooks — losses at fittings, as opposed to major losses along the pipe wall — and it says nothing whatsoever about magnitude. On plenty of real systems the fittings are the larger half.

Here is the example worked at the end of this article: 12 metres of 2-inch Schedule 40 line carrying 10 m³/h of water, with six 90° elbows, two gate valves, a swing check valve, and an entrance and an exit. The pipe friction comes to 0.44 m of head. The fittings come to 0.60 m — 58% of the total. Nothing about that system is unusual, and nothing about those losses is minor.

This guide is about applying fitting losses correctly rather than about looking the numbers up. It covers the two methods in general use, why the same elbow gets different values in different references, what happens to all of it at low Reynolds number, and where in a model the loss actually belongs. If what you need is the values themselves, the Crane TP-410 K-factor table has them.

The velocity head both methods share

Every method of handling fitting losses expresses the loss as some multiple of the velocity head — the kinetic energy the flowing fluid carries, written as a height of fluid:

• $h_f$ — head loss across the fitting, in metres of the flowing fluid
• $\Delta p$ — the same loss expressed as a pressure (Pa)
• $K$ — resistance coefficient for the fitting (dimensionless)
• $V$ — reference velocity through the fitting (m/s)
• $\rho$ — fluid density (kg/m³)
• $g$ — gravitational acceleration (9.81 m/s²)

That K is dimensionless is the whole point of the formulation: a fitting with K = 1 destroys exactly one velocity head, whatever the fluid and whatever the pipe. Every disagreement between methods, references and software comes down to two questions — what number K takes, and whether it is allowed to change with flow.

One thing to fix now, because it causes more errors than any coefficient ever will: $V$ is a reference velocity, and the reference is part of the K value. A reducer, an expander and a tee each have two velocities to choose from, and a K quoted against the small bore means something very different from the same number quoted against the large bore. Always check which velocity a published K belongs to.

Method 1: the resistance coefficient

The dominant approach in process and building services is the Crane TP-410 method, which builds K from two pieces:

• $n$ — the fitting's equivalent length-to-diameter ratio, a purely geometric property (30 for a standard 90° elbow, 16 for a long-radius one, 340 for a globe valve)
• $f_T$ — the fully turbulent friction factor for clean commercial steel at that pipe size

The part most people miss is what $f_T$ depends on. It is keyed to pipe size, not to your Reynolds number. Two consequences follow, and both surprise people:

• The same fitting has a different K in different pipe sizes. A standard 90° elbow is K = 0.81 in ½-inch pipe and K = 0.39 in 12–16 inch. A long-radius elbow runs from 0.43 down to about 0.21. That is not a rounding difference — it is a factor of two across the size range, and it is why a single "K = 0.75 for an elbow" quoted with no size attached is not usable.
• K does not respond to flow rate at all. Double the flow through the elbow and the Crane K is unchanged; only $V^2$ moves. That is a deliberate simplification, it is a very good one in fully turbulent flow, and it falls apart at low Reynolds number — which is what the 2K and 3K methods further down exist to fix.

Not every loss is built as $n \cdot f_T$. Entrances and exits are pure geometry and carry fixed values regardless of size: a sharp-edged entrance is 0.5, a well-rounded one 0.04, an inward-projecting one 0.78. A pipe exit is K = 1.0 — the full velocity head, because the fluid carries its kinetic energy out into the tank or the atmosphere and the system never gets it back. Exits are the single most commonly omitted loss in hand calculations.

One caution on sources. Successive editions of TP-410 have revised parts of the $f_T$ ladder, and other compilations — Idelchik, the Hydraulic Institute, manufacturers' own test data — differ again. The differences are mostly small, but they are real. Pick one reference and stay inside it for a given calculation; a spreadsheet that takes elbows from one edition and valves from another is worse than one that uses either consistently.

Method 2: equivalent length

The older approach replaces each fitting with the length of straight pipe that would lose the same head, adds those lengths to the physical pipe length, and runs a single Darcy-Weisbach calculation:

• $L_e$ — equivalent length of straight pipe, added to the physical length (m)
• $K$ — resistance coefficient for the fitting
• $D$ — internal diameter of the pipe (m)
• $f$ — Darcy friction factor, and the crux of the whole method (see below)

It is a genuinely useful device. One number goes into one equation, it is easy to check by hand, and a great deal of manufacturer data is still published this way.

It also contains a trap, and the trap is the $f$ in the denominator. An equivalent length is only valid at the friction factor it was derived at. Crane's $n$ is an equivalent length ratio — it is $K / f_T$, expressed on the fully turbulent basis. If you take that $n$ and convert it to metres using your operating friction factor, you have reintroduced exactly the Reynolds dependence that $f_T$ was constructed to exclude, and the two do not cancel.

Concretely, using the worked example below. Its operating friction factor is 0.0227, while $f_T$ for 2-inch pipe is 0.019. For one standard 90° elbow:

• Straight from the L/D ratio: $30 \times 0.0525 = 1.58$ m.
• Via $K D / f$: $0.57 \times 0.0525 / 0.0227 = 1.32$ m.

Same fitting, same table, two answers about 20% apart. Neither calculation is wrong in itself — they are answers to different questions — but silently mixing them is. The rule is simple: use whichever basis your reference states, and never convert between the two with a different friction factor.

Equivalent length still earns its place for hand calculations and spreadsheets where you want a single pipe length, for manufacturer data published as $L_e$, and for reproducing a legacy calculation on its own stated basis. For anything solved numerically, the K form is cleaner, because the solver already knows $f$ and does not need it twice.

When "minor" stops being minor

There is a one-line test, and it falls straight out of the equations above. Both the pipe and the fittings are multipliers on the same velocity head, so compare them directly:

• $\sum K$ — the fittings term: every fitting K on the run, added up
• $f L / D$ — the pipe term: friction factor × length ÷ bore

In the worked example, $\sum K = 7.12$ against $f L / D = 5.19$, so the fittings carry 58% of the loss. Put the identical fitting list on a 600 m run of the same pipe and $f L / D$ becomes 259 — the fittings drop to under 3% and you could reasonably ignore them.

That comparison is the whole of it, and it gives a usable rule of thumb: when fittings are spaced closer than roughly 50 pipe diameters apart, they dominate. Pump skids, meter runs, compressor packages, valve manifolds, plant room pipework and anything on a module — model every fitting. Cross-country transmission lines, long distribution mains, borehole risers — round them off and spend the effort elsewhere.

The corollary is the part that bites: the same fitting list is negligible in one design and controlling in another. "We always ignore fittings" is a habit that travels between projects and is right about half the time.

The Reynolds problem: the 2K and 3K methods

A constant K is a fully turbulent value. Real fitting losses climb as Reynolds number falls, steeply once the flow goes laminar, and a number that cannot see Reynolds number cannot follow that.

Two correlations are in general use. Hooper's 2K method (1981):

• $K_1$ — the laminar coefficient, tabulated per fitting (800 for a standard 90° elbow)
• $K_\infty$ — the fully turbulent coefficient, the value K tends to at high Re
• $Re$ — Reynolds number of the pipe the fitting sits on
• $D_{\text{in}}$ — the actual inside diameter, in inches

Darby's 3K method (2001) adds a third coefficient and scales better on large bore:

• $K_1$ — laminar coefficient, as above (800 for a standard 90° elbow)
• $K_i$ — turbulent coefficient (0.14 for that elbow)
• $K_d$ — size-scaling coefficient, in inches (4.0 for that elbow)
• $D_{\text{NPS}}$ — the nominal pipe size, in inches

Note that the two methods take different diameters — the actual bore for Hooper, the nominal size for Darby. It is another quiet way to be 10% out.

Both put the Reynolds dependence in a $K_1 / Re$ term that is negligible in turbulent flow and dominant in laminar flow. For a standard 90° elbow in 2-inch pipe, with Crane's constant K = 0.57 for comparison:

Read it from the bottom up. Above about 10⁵ all three agree within a few percent — the constant-K method is doing exactly what it was built to do, and there is no reason to reach for anything more complicated. By the laminar boundary the constant K is roughly 40% below the correlations. At Re = 100 it is low by an order of magnitude.

So the practical guidance is narrow rather than sweeping. For water, air, steam and gas at design flow — the overwhelming majority of pipework — the constant-K method is appropriate and the extra coefficients buy nothing. Reach for 2K or 3K when you are working with viscous liquids (oils, glycol, polymer solutions, sludges), small bore, or a deeply turned-down flow, and check where you actually are with a Reynolds number calculator before deciding. The coefficient tables are published in Hooper's original paper, in Darby's Chemical Engineering Fluid Mechanics, and in Perry's handbook.

SimuPipe works on the Crane turbulent basis, and flags a run where auto-K fittings sit on a pipe below Re 2000 so that you know the losses there are approximate rather than discovering it later. Where you have measured or manufacturer data for a fitting, entering the K directly overrides the correlation.

Where you put the loss changes the answer

There are two ways to place a fitting loss in a model, and the choice is not cosmetic. Either the fitting is a discrete point loss at its own node, or its K is folded into the adjacent pipe's resistance so the pipe carries a single lumped coefficient:

• $f L / D$ — the pipe's own friction term
• $\sum K$ — every fitting on that pipe, summed into the same bracket
• $V$ — the pipe velocity, shared by both terms, which is why they can be added at all

For an incompressible liquid in a constant-bore pipe, the two are algebraically identical. Same velocity, same density, same total — the distinction genuinely does not matter, which is why it is rarely discussed.

For a compressible fluid it matters a great deal. Gas density falls as pressure falls along the pipe, so velocity rises along the pipe, so where a loss is taken decides the density and velocity at which it is taken. The same ΣK lumped at the inlet, lumped at the outlet, or distributed along the run gives three different answers, and the gap widens with the pressure ratio.

SimuPipe folds a pipe's fittings into that pipe's own lumped resistance, which is the basis Crane's tables were built on and keeps the compressible case consistent. The practical takeaway is for anyone comparing tools: when two programs disagree on a gas line that carries fittings, check the placement convention before you start arguing about K values.

Worked example

A 12 m run of 2-inch Schedule 40 carbon steel (inside diameter 52.5 mm) carrying 10 m³/h of water at 20 °C — density 998.2 kg/m³, viscosity 1.002 mPa·s, absolute roughness 0.045 mm. Fittings: six 90° standard elbows, two full-open gate valves, one swing check valve, a sharp-edged entrance and an exit to atmosphere.

Step 1 — velocity and velocity head. $V = Q/A = 1.283$ m/s, so $V^2/2g = 0.0840$ m.

Step 2 — Reynolds number and friction factor. $Re = \rho V D / \mu = 67{,}100$ — comfortably turbulent — and Colebrook-White gives $f = 0.0227$.

Step 3 — the pipe term. $f L / D = 0.0227 \times 12 / 0.0525 = 5.19$.

Step 4 — the fittings term. Crane K at $f_T = 0.019$ for 2-inch pipe:

Step 5 — total. Both terms multiply the same velocity head:

Of which the pipe wall accounts for 0.44 m and the fittings for 0.60 m. As a sanity check on the other method, the equivalent length of that fitting set is $\sum K \cdot D / f = 16.5$ m — there is more equivalent pipe in the fittings than there is real pipe in the run.

Two details worth noticing. The swing check valve alone contributes more loss than all six elbows would in a long-radius pattern, which is the kind of thing that only becomes visible once the fittings are itemised. And the entrance and exit together contribute 1.5 — more than four elbows — despite being the two items most often left out entirely.

You can build the same run in SimuPipe and get the itemised breakdown without the arithmetic: add the pipe, open its fittings list, pick the quantities, and the ΣK is derived from the bore and folded into that pipe's resistance automatically. Any row's K can be overridden with a measured value, and the report prints the provenance of each one.

Common mistakes, quickly

• Mixing bases. Taking K from one reference and L/D from another, or converting Crane's L/D to metres with your operating friction factor.
• The wrong reference velocity. On a reducer, expander or tee, a K quoted against the small bore is a very different number from the same K against the large bore.
• Using a turbulent K in laminar flow. Below about Re 10⁴ the constant-K method progressively understates the loss; on a viscous line it is not close.
• Charging a tee's branch K to the run leg. A tee has two K values, they differ by a factor of three, and which applies depends on the flow split.
• Double-counting a control valve. If the valve is sized by its Cv or Kv, it already carries its full loss. Adding a "globe valve" K on top counts it twice.
• Forgetting the exit. K = 1.0, on every discharge to a tank or to atmosphere, in every system.
• Ignoring fittings by habit. Run the ΣK against f·L/D ratio once per system rather than assuming the answer from the last one.

Where a network simulator fits in

None of this needs software. A single line with a known flow is a spreadsheet problem, and the friction loss calculator will do it faster than a spreadsheet will.

What changes with a network is that the flow split is no longer known. As soon as there are loops, parallel branches or several supply points, the fittings influence how the flow distributes, the distribution sets each pipe's velocity, and the velocities set the losses — all of which have to be solved together rather than in sequence. A pump skid where the fittings carry 58% of the loss will move its own operating point when a check valve is added, and there is no order in which to do that by hand.

That is the case for modelling it: not that the fitting equations are difficult, but that once they are coupled you want the coefficients itemised, consistent and visible on every pipe at once. You can try that in the browser sandbox without an account, or create a free account to save the model.

For the pipe-friction half of the same calculation, see how to calculate pressure drop with Darcy-Weisbach.