Pipe Network Analysis: Hardy Cross Method

In hydraulic engineering, pipe flow is commonly considered as a full-pipe pressure-driven

flow occurring in closed conduits of a circular cross-section. The two basic equations that
describe pipe flow are the continuity (or mass conservation) equation and the energy
equation.
Assuming incompressible fluid and considering a pipe with varying cross-sectional area, the
continuity equation can be written as:

𝑄 = 𝐴1 𝑢1 = 𝐴2 𝑢2 = 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (1)
or
𝜋𝐷1 2 𝜋𝐷2 2
𝑄= 𝑢1 = 𝑢2 = 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (2)
4 4

where Q is the volumetric discharge, ui is the uniform cross-sectional velocity, and Ai and Di
are the cross-sectional area and pipe diameter, respectively, at point (i).
Written in terms of ‘head’, the energy equation reads:

𝑢1 2 𝑝1 𝑢2 2 𝑝2
+ + 𝑧1 = + + 𝑧2 + ℎ𝑒 (3)
2𝑔 𝛾 2𝑔 𝛾

where pi (= γyi) is the hydrostatic pressure at point (i), zi is the elevation of the pipe from
some reference datum, and he is the energy loss (head loss) between the two sections (Figure
1).

Figure 1: Pipe Flow Characteristics

The term he includes both the frictional (hf) and the local (minor; hL) energy losses. The
frictional losses are quantified by the well-known DarcyWeisbach equation:

1
 𝑓𝐿𝑢2 8𝑓𝐿𝑄 2
ℎ𝑓 = = 𝑔𝜋2 𝐷5 = 𝑅𝑄 2 (4)
2𝑔𝐷

where f is the friction coefficient and R is the resistance coefficient. The friction coefficient
𝜌𝑢𝐷 𝑢𝐷 𝜀
depends on the Reynolds number 𝑅𝑒 = = and the relative roughness𝐷, where 𝜌 is the
𝜇 𝜈
fluid density, μ is the dynamic (or absolute viscosity), ν is the kinematic viscosity, and ε is the
roughness of the pipe’s interior surface (available from the manufacturer’s specifications).
The Darcy-Weisbach formula is valid for both turbulent and laminar flows. For laminar
flow (Re < 4000) the friction factor is independent of the relative roughness:

64
𝑓=𝑅 (5)
𝑒

For intermediate turbulent flow (4000 < Re < 108 and 0 < ε/D < 0.05) the friction factor is
quantified by the Colebrook equation as:

1 𝜀 9.30
= 1.14 − 2𝑙𝑜𝑔 (𝐷 + ) (6)
√𝑓 𝑅𝑒 √𝑓

For hydraulically smooth (ε = 0) turbulent pipe flow, the Colebrook equation reduces to the
Prandtl equation:

1
= 2𝑙𝑜𝑔(𝑅𝑒 √𝑓) − 0.8 (7)
√𝑓

For high Reynolds numbers (Re > 108) (hydraulically rough flow) the factor f depends only on
the relative roughness and is expressed by the von Karman equation as

𝜀 −2
𝑓 = [1.14 − 0.869𝑙𝑛 (𝐷)] (8)

Colebrook and Prandtl equations are implicit and require an iteration process for the
estimation of the friction factor (f).
Minor losses occur locally for reasons such as pipe expansion or contraction, pipe bends and
valves. In general those losses are expressed as

𝐾𝑢2
ℎ𝐿 = (9)
2𝑔

where K is some experimental parameter.

Pipe networks are comprised of a large number of interconnected pipes forming loops and
branches. The pipes (branches) are connected on the junctions (nodes) of the network, and in
groups they form closed loops. For simplicity it is assumed that pipe flow is steady and that
water delivery to consumers occurs only from the junctions. Given the discharges inflowing

2
or outflowing from the network junctions, the unknown quantities are the discharges of all
pipes connected at the junctions. For this purpose the continuity and energy equations are
employed.

Continuity equation at the junction

Due to the incompressibility of water, at every junction (j) the sum of all discharges –
inflowing or outflowing and those from the connected pipes (k-number) – should be zero:

∑𝑘(𝜎𝑘 𝑄𝑘 ) ± 𝑄𝑗 = 0 (10)

where Qk are the flows of all connecting pipes at joint j, σk = (±) accounts for the sign of each
pipe flow (positive for inflowing and negative for outflowing) and Qj is the inflowing (+) or
outflowing (–) discharge at the junction (j = 1 to J-number of junctions; Figure 2).

Figure 2: Continuity of flows at a junction

Since both the direction and magnitude of the discharge of the connecting pipes are
unknowns, it is important that the discharges selected during the first approximation satisfy
the continuity equation (Equation 10), and also the convention of positive and negative flows
should be applied consistently throughout the network.

Energy equation for a loop

For each branch of the network (i), the energy loss hfi depends on the resistance coefficient Ri
and 𝑄𝑖 2 (where i = 1,….., N-number of branches; Equation 4). Thus, an equation based on the
energy conservation principle can be written for each loop of the network, stating that the
algebraic sum of the head losses along the branches of a certain loop is zero. This is true
when no pump or turbine is placed along a pipe that would add or subtract energy from the
loop. The total head loss between any two junctions is independent of the path followed
(Figure 3).

3
 Figure 3: Energy balance within a loop

For each loop (l) the energy equation takes the form:

∑𝑚(𝜎𝑚 ∗ 𝑅𝑚 𝑄𝑚 2 ) = 0 (11)

where 𝜎𝑚 ∗ is the directional sign of the flow in each branch forming the loop (m is the
number of branches comprising loop l). The clockwise direction is taken as positive and the
convention is retained for all the loops of the network. The total number of unknown
discharges equals the number of N-branches in the network. The total number of equations
that can be written is (J – 1) number of junction equations (Equation 10) and M-number of
loop equations (Equation 11). Since N = M + J – 1, the number of unknowns equals the
number of equations. The resulting non-linear system can also be solved numerically by
using some successive iteration algorithm.

Hardy Cross Method

The Hardy Cross method solves the system of nonlinear algebraic equations
by using a successive iteration algorithm and by neglecting some high-order terms during the
analysis. Due to its quick convergence to the analytic solution, the method is widely applied
in computer simulation software packages related to hydraulics and structural analysis. The
sequential steps of the Hardy Cross method are as follows:

Step 1— Some initial discharge values (by magnitude and direction) are selected for all the
branches of the network. The selection criterion is simply the satisfaction of the continuity of
volumetric discharges on all junctions of the network, something that can be easily
accomplished. However at this point, the selected discharges do not satisfy the energy
conservation principle within the loops of the network.

4
Step 2— Next, a successive correction of those selected discharges over all the branches for
each loop is pursued. This is accomplished by estimating a correction amount ΔQl, common
to all the branches of a loop (l), assigned according to the sign 𝜎𝑚 ∗ of each branch (relatively
to the loop it belongs). If the number of iterations is (n) then the next iteration (n + 1)
produces a corrected discharge of the branch (m) according to the relation:

𝑄𝑚 𝑛+1 = 𝑄𝑚 𝑛 + 𝜎𝑚 ∗𝑛 ∆𝑄𝑙 (12)

The correction ΔQl is found using the energy conservation equation along each loop as

∑𝑚 𝜎𝑚 ∗ 𝑅𝑚 (𝑄𝑚 + 𝜎𝑚 ∗ ∆𝑄𝑙 )2 = 0 (13)

where m is the number of branches that form the loop. Assuming that ΔQl is small as
compared to Qm and neglecting the second-order term ∆𝑄𝑙 2 , then Equation 13 yields

∑𝑚 𝜎𝑚 ∗ 𝑅𝑚 𝑄𝑚 2
∆𝑄𝑙 = ∑𝑚 2𝑅𝑚 𝑄𝑚
(14)

This correction, given the proper sign, is added algebraically to Qm of the branch estimated
during the previous iteration step. Of course, after the correction along one loop, the
corrected discharges on all neighbouring loops (having common branches with the recently
‘corrected’ one) no longer satisfy the energy balance. Thus a new correction must be made
for all the loops.

Step 3—The procedure is repeated over all loops as many times as needed until it satisfies a
convergence criterion, that is, the maximum value of correction ΔQl for all loops of the
network is less than a predefined small number.

During the solution process it should be emphasized that (1) after each correction along a
loop the volumetric continuity principle is maintained on all junctions of the network, and (2)
in the case that a discharge of a certain branch changes sign due to the imposed correction,
then care should be taken for changing the sign of the discharge in the subsequent
corrections. This iterative procedure can be conveniently coded for computer programming,
provided that appropriate information is given describing the structure of the network. This is
accomplished by the synthesis of the connectivity matrix having (M) rows and (N) columns
equal to the M-number of loops and the N-number of branches, respectively. Each element of
the matrix is either 0, +1 or –1. If the branch is not part of the loop, then the element is zero
(0); if the branch is part of the loop with a positive flow direction, then it is positive one (+1);
and if the branch participates to the loop with a negative flow direction, then it is negative
one (–1).