Hydraulic

Multi-branch closed unification loop: supply/return headers and hydraulic balancing

Q: Why can't I size each branch separately?

Because in a closed loop every branch shares the same supply and return headers. The pressure at the take-off point is not a fixed input: it depends on the flow passing through all the other branches. Opening, closing or throttling one branch changes the header loss and redistributes flow among the rest. That is why the network must be solved in a coupled way, with every flow rate as a simultaneous unknown.

Q: What does the Newton-Raphson method solve here?

It solves the nonlinear system formed by two families of equations: mass conservation at each node (inflow equals outflow) and pressure balance (the head computed along different paths to the same node must agree, i.e. the sum of the losses around each loop is zero). Because the loss h_f grows with Q², the system is nonlinear; Newton-Raphson linearizes it with a Jacobian and iterates until the residual drops below 10⁻⁷.

Q: How does this differ from the Hardy-Cross method?

Hardy-Cross corrects one loop at a time, iterating on circulating flows — it converges, but slowly in large or heavily coupled networks. Global Newton-Raphson assembles the Jacobian of the whole network and updates every flow at each step, with quadratic convergence near the solution. For loops with up to 20 branches, supply/return headers and in-line equipment, the global approach is more robust and faster.

Q: What are the orifice plates for in balancing?

An orifice plate adds a controlled head loss (ΔP ∝ Q²) to the branch with surplus pressure — typically the short branches close to the pump. Without it, those branches would swallow more flow than designed. The plate dissipates exactly the surplus head so that the pressure difference consumed between the headers is equal across all branches, equalizing the flows. It is the fixed equivalent of a balancing valve.

Q: Does the return header also count in the calculation?

Yes, and ignoring it is a common mistake. The return is a manifold: flow accumulates along it, and the return-header head loss penalizes the branches that discharge farthest along the collector. The calculation treats supply and return as two separate headers, each with its own segments, take-offs and fittings, closing the loop at the unification node.

Q: What happens if a branch becomes unreachable?

If the pump lacks enough head to overcome the loss to a consumer (high elevation, very long branch, oversized plate), that branch does not receive its design flow. The solver detects the case, flags the branch as unreachable and signals it in the results — indicating that the diameter, the plate, the pump curve or the point elevation must be revised.

In a closed unification loop the flow rate of each branch is not free: it emerges from the coupling of every branch through shared supply and return headers. The method solves mass conservation and pressure balance simultaneously by Newton-Raphson and balances the branches with orifice plates.

When to use

Use this when several consumers (heat exchangers, reactor jackets, bearings, points of use) are fed by a distribution (supply) header and return to a collection (return) header, forming a closed recirculation loop. This is the typical topology of chilled water, cooling water, thermal oil, seal and lubrication systems. Unlike a single branch, here the pressure at the take-off point is shared: opening or throttling one branch redistributes flow to all the others. The sizing exercise ensures each consumer receives its design flow, computes the total flow the loop pump must deliver and sets the orifice plates that balance the short branches against the long ones.

What a closed unification loop is

Many industrial utilities — chilled water, cooling water, thermal oil, seal and lubrication systems — do not feed a single point but several consumers in parallel from a common trunk. The pump delivers into a supply header (distributor), from which branches take off to serve each heat exchanger, jacket or bearing; the returns merge into a return header (collector) that gathers, or unifies, all the flows back to the tank or to the pump suction. The result is a closed recirculation loop.

The feature that makes this problem hard — and different from sizing an isolated branch — is the coupling: the pressure at each branch’s take-off point is not a fixed input, it depends on the flow circulating through all the branches at once. Throttling one consumer reduces the header loss and pushes flow toward the others. There is therefore no branch-by-branch solution: the entire network must be solved at once.

Why flow distributes unevenly

Consider two identical consumers, one near the pump and one at the end of the header. Both see roughly the same pressure difference between supply and return, but the distant branch carries a longer header run — with more accumulated head loss. Since flow adjusts so that each path consumes the available head, the near branch, with less loss, captures more flow, and the distant one is underserved. It is common for the short branch to receive 1.5 to 3 times the flow of the long branch before any correction.

This imbalance is purely hydraulic and unavoidable in branched systems. The fix is balancing: introducing extra loss in the favored branches until every one consumes the same head difference.

The model: junction and loop laws

The network is described as a graph, with one unknown flow per segment. Two families of equations close the system:

Mass conservation (junction law): at every take-off, junction, and equipment inlet and outlet, Σ Q_in = Σ Q_out. Because each segment carries an independent flow, even in-line nodes generate an equation.
Pressure balance (loop law): every node reached by more than one path — a junction at the return header, the loop closure — must have the same head H computed along any path. This is equivalent to requiring that the sum of the losses around each closed loop be zero.

The head at each node is propagated from the pump discharge, H = z_level − h_suction + H_pump(Q_total), subtracting the loss h_f of each downstream segment. The loss follows Darcy-Weisbach, h_f = [f·(L/D) + ΣK]·v²/(2g), with the friction factor from the Colebrook-White method: laminar (f = 64/Re) below Re 2000, Dunlop cubic interpolation in the transition, and the Serghides estimator + Newton on Colebrook-White in the turbulent regime (Re ≥ 4000).

How Newton-Raphson solves the network

Because h_f ∝ Q², the system is nonlinear and needs an iterative method:

Initial guess of flow for each segment (higher in the headers, lower in the branches).
Residual evaluation — how far each mass and pressure equation is from zero.
Numerical Jacobian — the derivatives of the residuals with respect to each flow are estimated by central finite differences (a 0.001 m³/h perturbation), assembling the matrix that describes how the network reacts to small flow changes.
Newton step — the linear system J·ΔQ = −residual is solved by Gaussian elimination with pivoting, yielding the correction to all flows at once.
Line search — a fraction α of the step is applied (halved until the residual actually drops), ensuring stability even far from the solution.
Convergence when the residual norm falls below 10⁻⁷ (typically in fewer than 15 iterations). A 200-iteration cap acts as a safeguard.

The advantage over the classic Hardy-Cross method — which corrects one loop at a time — is global, nearly quadratic convergence: every flow is updated at each step, which matters in loops with many branches and in-line equipment.

Automatic balancing with orifice plates

Once the network is solved, the imbalance appears: some branches above and others below the design flow. Auto-balancing acts on the leaves (the real consumers, never the headers), inserting an orifice plate in the branch with surplus pressure. The plate adds a loss ΔP ∝ Q², sized to dissipate exactly the head difference between that branch and the critical branch (the one with the highest loss):

K_plate = 2g·Δh_target / v_branch²

With the design flow imposed at each leaf, the surplus head of each branch is computed and converted into K. The result is a distribution in which every branch consumes the same pressure difference between headers — hence the same proportion of requested flow. Headers (parent branches) receive no automatic plate: if the designer wants to relieve the leaf plates (staging to avoid cavitation), a manual plate is inserted in the header, which enters the pressure propagation naturally.

Practical design considerations

Treat supply and return as two headers: the return-collector loss penalizes the most distant branches — ignoring it underestimates the imbalance.
Do not overdo the plate: excessive ΔP generates noise, vibration and cavitation risk just downstream; when the surplus is large, split it between a branch plate and a manual header plate.
Check for unreachable branches: high elevation or an aggressive plate can leave a consumer without head; the solver flags it, and the fix is the diameter, plate, elevation or pump curve.
Keep velocities within range: 1.5–2.5 m/s in the headers and 1.0–3.0 m/s in the branches, with the long branches near the lower bound to avoid accumulating loss.

The chain — build the graph, impose the junction and loop laws, solve by Newton-Raphson, and balance with plates — delivers a flow distribution that respects each consumer’s design and is resolved with numerical rigor by mass and energy conservation.

Formulas and fundamentals

Mass conservation at a node (junction law) Σ Q_in − Σ Q_out = 0

At every internal node (take-off on the supply header, junction on the return header, equipment inlet/outlet) the sum of the incoming flows equals the sum of the outgoing flows. Q in m³/h. It yields one equation per junction node and is the backbone of the network balance.

Pressure balance around the loop (loop law) Σ_loop h_f(Q) = 0 ⇔ H_A(path 1) = H_A(path 2)

For a node reached by more than one path (branches merging at the return header, or a loop closure) the head H computed along every path must agree. This is equivalent to stating that the algebraic sum of the head losses around any closed loop is zero. h_f in m of head; H in m of head.

Segment head loss (Darcy-Weisbach) h_f = [ f·(L/D) + ΣK ] · v²/(2g) , v = 4Q/(π·D²)

Total loss of a segment combining distributed friction and fittings. f friction factor [dimensionless], L length [m], D internal diameter [m], ΣK sum of fitting loss coefficients [dimensionless], v mean velocity [m/s], g = 9.81 m/s². It sets the nonlinear relation h_f ∝ Q² that couples the branches.

Friction factor (Serghides explicit / Colebrook-White) 1/√f = −2·log₁₀( ε/(3.7·D) + 2.51/(Re·√f) ) , Re = ρ·v·D/μ

Turbulent regime (Re ≥ 4000): the Serghides estimator + 2 Newton steps on the Colebrook-White equation (machine precision). For Re < 2000 the laminar law f = 64/Re applies, and between 2000 and 4000 the Dunlop cubic interpolation is used. ε absolute roughness [m], μ dynamic viscosity [Pa·s], ρ density [kg/m³]. This is the Colebrook-White (Serghides) friction-factor method.

Balancing orifice plate coefficient K_plate = 2g·Δh_target / v_branch² ⇒ ΔP_plate ∝ Q_branch²

Extra loss imposed on a branch with surplus pressure to match its available head to the loss of the critical branch. Δh_target is the surplus head to be dissipated [m of head], v_branch the branch velocity at design flow [m/s]. Auto-balancing computes K_plate so that every branch consumes the same head difference between headers.

Standards & methods

ABNT NBR 12214 — Design of water pumping systems
Colebrook-White method — Serghides explicit approximation (turbulent), Dunlop transition, laminar 64/Re
Hydraulic Institute (HI) — piping system analysis and system curve
Crane TP-410 — Flow of Fluids (fitting K coefficients and orifice plates)
ISO 5167 — Flow measurement by orifice plates (basis for the plate K)

Typical reference values

Quantity	Typical range	Note
Header velocity (supply/return)	1.5 to 2.5 m/s	A header is a manifold: a slightly lower velocity reduces the accumulated loss seen by distant branches.
Branch velocity	1.0 to 3.0 m/s	Long branches near the lower bound to avoid penalizing the loss; short branches tolerate higher velocity.
Imbalance before balancing	near branch receives 1.5 to 3× the flow of the far branch	The branch closest to the pump has more available pressure and 'steals' flow; hence the need for orifice plates.
Number of branches supported	up to 20 branches (leaves)	Includes intermediate headers (parent sub-branches) that group consumers.
Convergence tolerance	‖residual‖ < 1×10⁻⁷	Newton-Raphson with a numerical Jacobian; it typically converges in fewer than 15 iterations.
Balancing-plate margin	dissipate 10 % to 40 % of the header head	An overly aggressive plate raises noise and downstream cavitation risk; prefer splitting the loss between branch and header.

Worked example

Cooling-water loop with 4 consumers

Inputs

Total design flow: Q_total = 80 m³/h
Number of branches (leaves): n = 4 branches
Target flow per branch: Q_branch = 20 m³/h
Header diameter (supply/return): D_h = 150 mm
Branch diameter: D_r = 80 mm
Short / long branch length: L = 12 / 60 m

Results

Header velocity: v_h ≈ 1.26 m/s
Branch velocity (20 m³/h): v_r ≈ 1.11 m/s
Short-branch flow without balancing: ≈ 27 m³/h
Long-branch flow without balancing: ≈ 13 m³/h
Plate K on the short branch: K ≈ 35 dimensionless

With 80 m³/h in the 150 mm header, v = 4Q/(πD²) ≈ 1.26 m/s, and each 80 mm branch at 20 m³/h gives ≈ 1.11 m/s — within typical ranges. Solving the network without plates, however, the Newton-Raphson solver reveals the imbalance: the short branch (12 m) has less loss and captures ≈ 27 m³/h, while the long one (60 m) settles at ≈ 13 m³/h. The loss difference between the two paths is the short branch's surplus head. Auto-balancing computes K ≈ 35 for a plate on the short branch (and smaller values on the intermediate ones), dissipating that surplus with ΔP ∝ Q² until every branch consumes the same head difference between the headers — restoring 20 m³/h in each. The total flow the loop pump delivers remains 80 m³/h, now split according to the design.

Run your own calculation Create a free account and get this for your own inputs in seconds.

Common mistakes

Sizing each branch in isolation, as if the pressure at the take-off point were fixed — in a closed loop the pressure is shared and flow redistributes across all branches.
Forgetting the return-header loss: the return line also accumulates head loss, and it penalizes precisely the branches that discharge farthest along the collector.
Skipping the balancing step: without orifice plates the branches near the pump 'steal' flow and the distant ones fall below design, even with the correct pump.
Placing a single plate in the header instead of one plate per branch, expecting it to balance — a header plate throttles all the children together and does not correct the imbalance between them.
Oversizing the balancing plate and generating excessive ΔP, causing noise, vibration and cavitation risk just downstream of the orifice.
Ignoring the elevation of the return points: if the return rises to the top of the source tank, the static component enters the balance and shifts the equilibrium flow.

Frequently asked questions

Why can't I size each branch separately?

Because in a closed loop every branch shares the same supply and return headers. The pressure at the take-off point is not a fixed input: it depends on the flow passing through all the other branches. Opening, closing or throttling one branch changes the header loss and redistributes flow among the rest. That is why the network must be solved in a coupled way, with every flow rate as a simultaneous unknown.

What does the Newton-Raphson method solve here?

It solves the nonlinear system formed by two families of equations: mass conservation at each node (inflow equals outflow) and pressure balance (the head computed along different paths to the same node must agree, i.e. the sum of the losses around each loop is zero). Because the loss h_f grows with Q², the system is nonlinear; Newton-Raphson linearizes it with a Jacobian and iterates until the residual drops below 10⁻⁷.

How does this differ from the Hardy-Cross method?

Hardy-Cross corrects one loop at a time, iterating on circulating flows — it converges, but slowly in large or heavily coupled networks. Global Newton-Raphson assembles the Jacobian of the whole network and updates every flow at each step, with quadratic convergence near the solution. For loops with up to 20 branches, supply/return headers and in-line equipment, the global approach is more robust and faster.

What are the orifice plates for in balancing?

An orifice plate adds a controlled head loss (ΔP ∝ Q²) to the branch with surplus pressure — typically the short branches close to the pump. Without it, those branches would swallow more flow than designed. The plate dissipates exactly the surplus head so that the pressure difference consumed between the headers is equal across all branches, equalizing the flows. It is the fixed equivalent of a balancing valve.

Does the return header also count in the calculation?

Yes, and ignoring it is a common mistake. The return is a manifold: flow accumulates along it, and the return-header head loss penalizes the branches that discharge farthest along the collector. The calculation treats supply and return as two separate headers, each with its own segments, take-offs and fittings, closing the loop at the unification node.

What happens if a branch becomes unreachable?

If the pump lacks enough head to overcome the loss to a consumer (high elevation, very long branch, oversized plate), that branch does not receive its design flow. The solver detects the case, flags the branch as unreachable and signals it in the results — indicating that the diameter, the plate, the pump curve or the point elevation must be revised.

Glossary

Header (manifold/distributor): The larger-diameter main pipe from which the branches take off (supply) or into which they merge (return). A closed loop has one supply header and one return header.
Closed unification loop: A ring topology where the pump feeds the supply header, the branches consume the flow, and the return header gathers (unifies) the returns back to the pump or tank.
Hydraulic balancing: The act of matching the actual flow of each branch to its design value, usually by introducing losses (orifice plates or valves) in the branches with surplus pressure.
Numerical Jacobian: The matrix of partial derivatives of the residuals with respect to each flow, estimated by central finite differences. It is what Newton-Raphson inverts at each iteration to update the flows.
Junction and loop laws: Hydraulic analogues of Kirchhoff's laws: mass conservation at the nodes and a zero sum of losses around each closed loop. Together they form the system the network solves.
Orifice plate (balancing): A disc with a calibrated bore that imposes a known head loss (ΔP ∝ Q²) to throttle a branch and balance the flow distribution.