World of Shareware - Software Farm 2/wosw_2.zip/wosw_2/GENERAL/ALLWET.ZIP/CHAP6.SML

Text File | 1987-05-22 | 22KB | 834 lines

                 6. PERFORMANCE AND INTERPRETATION OF ANALYSES


6.1 Successful Analyses and Their Interpretation

     In batch mode, ALLWET performs the analysis immediately after successfully

processing the input data.  In interactive mode, the analysis begins immediate-

ly after the confirmation of the multiplicative factors (c.f. section 4.3.20).


     ALLWET begins an analysis by summing, at each node without a reservoir or

pressure setting, the products of the node's positive demands (A, B, C, and D)

with the appropriate multiplicative factor.  Negative demands are included in

the sum without application of a multiplicative factor.  ALLWET indicates the

completion of this calculation by printing the algebraic sum of the demands

from all nodes having fixed demands.


     ALLWET next makes initial estimates of flows both at nodes having pressure

settings or reservoirs, and of flows through pipes. If the reservoir  fraction

of supply values sum to between 0.99 and 1.01, initial flow at each reservoir

is based upon these fraction of supply values, while flow at each  pressure

node is set to zero.  Otherwise, the fraction of supply values (and  therefore

the initial flow estimates) are set the same at all reservoirs and pressure

nodes.  The printed line "OUTFLO-INFLO=X" indicates the completion of these

estimates.  The absolute value of X will be less than .01 times the total of

all nodal demands. (Any difference from zero is caused by the reservoir

fraction of demand values not summing to exactly 1.00.)


     ALLWET now forms the system of nonlinear equations which represents the

mathematical model of the water distribution system.


     The conditions at each reservoir, pressure node, and booster pump station

for the initial flows just calculated are now printed.  The sign of the initial



                                  6-1






flow at each reservoir and pressure node is the opposite of the corresponding

fraction of supply value.  If any reservoir or booster station appears to be

operating out of the range of the H-Q curve, and the calculations terminate

abnormally, an error in the H-Q curve or other input data should be suspected.


     ALLWET solves the equations by iteratively adjusting the flows in pipes in

order to achieve a zero loss in the HGL around every loop.  Messages indicate

when PRVs are turned on or off.  If a check valve is activated (or if reverse

flow occurs through a PRV), ALLWET constrains the flow through the correspond-

ing pipe to zero by removing the pipe from the system for the duration of the

calculations.  In this case ALLWET revises the system of equations and restarts

the solution process.  ALLWET calculates either until a sufficiently accurate

solution has been found (the usual case), or until 60 iterations have been

performed (120 if PRVs are present).  If the calculations either terminate

prematurely due to data errors, or fail to converge, the input data and

calculated results (if any) must be reviewed as described in section 6.2.


     ALLWET now checks the calculations.  First, ALLWET calculates and prints

the maximum unbalanced head in any pseudoloop and in any loop.  (Pseudoloop

means a loop of pipes, temporarily created by ALLWET, containing a reservoir or

pressure node.) If the analysis converged and this value is not very small,

either at least one reservoir have been misspecified, the data contains other

errors, or the data describes a physically unrealistic situation.  Second,

ALLWET checks the net flow at each node and indicates those nodes for which the

net flow is greater than EPSILON.  For successful, convergent analyses in

which EPSILON does not exceed 1/10 the tolerable flow inaccuracy in any pipe,

the sum of any printed nodal imbalances will not significantly exceed X of

OUTFLO-INFLO=X.  In such cases, one or two nodes generally will have most of




                                  6-2






the total imbalance, and any imbalances indicated for other nodes will not

exceed ten times EPSILON.


     The nodal elevations have not yet been used in the calculations.  Now,

however, ALLWET computes the pressure at each node using the calculated HGL and

the node's elevation.  For nodes not having type 1 or type 2 reservoirs  the

interpretation of the pressures is straightforward.  However, for any node with

a reservoir, the pressure calculation may not be relevant, since it depends

upon what nodal elevation was used (e.g., at the top vs. the base of a storage

tank; or pump outlet vs. ground elevation).


     To complete the analysis, ALLWET calculates the flow velocities in all

pipes.


     In batch mode ALLWET generates a complete printout of the input data and

analysis results.  Figure 6 shows this for the sample problem.  In interactive

mode, the user is instructed to use the LIST or BATCH commands to review the

results.


     Node flows are positive if out of the system and negative if into the

system.  This is important in determining whether or not pressure nodes and

type 1 reservoirs are drawing flow from the system.


     Calculated pipe flows from the pipe's "from" node to the pipe's "to" node

are considered positive.  Flows in the opposite direction are considered

negative.  Head loss across each pipe usually has the same sign as the pipe's

flow.  However, pipes with booster pump stations will usually show a negative

head loss (i.e., a head gain) due to the pumps.


AN IMPORTANT NOTE ABOUT CHECK VALVES, PRVS: Read section 6.2.7!!!!!




                                  6-3






6.2 Correction of Unsuccessful Analyses

6.2.1 General Advice

     The equations which ALLWET solved represent a mathematical model of a

water distribution system.  If the input data reflects a physically unrealistic

situation (e.g., too large a demand at a node; a missized pump; or impossible

constraints placed upon the system by valves, pumps, or pressure nodes), then

either unrealistic output (e.g. pressures, flows, pipe head losses or HGL

values) may result, or the pressure-flow calculations may either terminate

prematurely or not converge.


     The cause of any unsuccessful analysis should be discernible with the help

of (i) this section and sections 6.2.2 through 6.2.7, (ii) the information and

diagnostics printed during the calculations, and (iii) calculation  results

generated either during batch use of ALLWET or by the LIST and BATCH interac-

tive commands.


     When performing the flow-pressure calculations, during some iterations

ALLWET may print a line of four numbers, each line often preceded by the

message

          ERROR  CONDITION  ENCOUNTERED  IN  ROW XXX

in which an integer replaces XXX.  Each line of four numbers means ALLWET is

having trouble solving the physical model of the problem.  The first of the

four numbers indicates the iteration number. (For the numerical analysis

oriented, ALLWET found, at this iteration, that the matrix representing the

linearization of the simultaneous nonlinear equations was not positive

definite.) ALLWET tries to repair the problem during successive iterations,

and may or may not eventually achieve convergence within the maximum permitted

number of iterations.  If a valid solution is found, disregard any rows of




                                  6-4






numbers and ERROR CONDITION ENCOUNTERED messages.  If not, other diagnostics

will appear either during or at the end of the flow-pressure calculations.


     If other error or warning message(s) appear during the flow-pressure

calculations, consult the corresponding section (one of 6.2.2 to 6.2.7) sug-

gested by the message.  During interactive sessions, the data base can be

corrected (and saved) using the ALLWET interactive command language, and the

analysis repeated using the RUN command.  If the cause of an unsuccessful

analysis is not discerned within several minutes during an interactive ses-

sion, the user should save the data set if desired, issue the BATCH command,

and then terminate the session via the HALT command.  Batch output from an

interactive or batch job can be carefully studied.  The data base can later be

corrected and saved by either (1) an interactive terminal session, or (2) a

file editor.


     The most usual cause of unsuccessful analyses is incorrect or misplaced

input data.  A quick review of the input data often reveals problems such as

missing or erroneous input values.  The units of all input should be checked.

Make sure the HGL constant and the head values of H-Q curves are in units of

head and not of elevation. (If English units are used, these will both be

feet.) If the calculations seemed to terminate normally, but the maximum loop

misbalance is high and/or pipe flows appear unrealistic, then an HGL constant

may be bad.  Reread section 2.3 to insure that the HGL constants are being used

correctly.  The correctness of H-Q curves and valve parameters should also be

confirmed.  If the first iteration conditions at a reservoir or a booster pump

appear unreasonable, the H-Q curves are likely incorrect.








                                  6-5






6.2.2 Non-Convergence

     If an accurate solution could not be found in 60 iterations (120 if PRVs

or check valves are present), ALLWET prints a message indicating that unless

too small an EPSILON was chosen, the calculated flows and pressures probably

have little meaning.  Any one of several circumstances may have caused this.


     First, a user specified EPSILON may have been chosen as too small, i.e.,

the user requested too much accuracy in the answers.  If the guidance in

section 3.1.1 was followed, this will probably not be the problem.


     Alternatively, the system may have almost converged, and would have given

a few more iterations.  In general, the specified limits of 60 or 120 itera-

tions is really adequate--problems with any intention of converging usually

will do so within this limit.  So unless the maximum head imbalance in any loop

or pseudoloop is less than .01, the calculated flows and pressure should be

considered invalid.


     The data may have a bad pump curve or a misplaced PRV, or may be simu-

lating an unrealistic set of conditions.  When non-convergence is encountered,

after checking EPSILON, check the pumps, reservoirs, pressure nodes and valves

within the system (c.f. sections 6.2.4 through 6.2.7).


     A final possibility is that the system of simultaneous, nonlinear equa-

tions and the initial conditions (i.e., initial flows) selected by ALLWET may

just defy convergence.  Such a risk always exists when solving such a set of

equations.  In fact, some systems of equations will converge with some sets of

initial conditions, but not with others.  Two techniques can coax a reluctant

problem to converge.






                                  6-6






     The first is to select a different reservoir or pressure node as the

reference reservoir.  For most problems any choice works, so by default ALLWET

uses the first reservoir or pressure node it finds.  However, systems with

multiple reservoirs or pressure nodes will sometimes converge only with spe-

cific choices for the reference reservoir. (This causes ALLWET to create a

different set of equations with which to model the system.) Try a reservoir or

pressure node for which the sum of the distances between itself and each of

the other pressure nodes and reservoirs is minimum.  The measure of distance

should be the number of pipes rather than the length of the pipes.  A common

practice is to model pumps in serial by creating adjacent type 2 reservoirs.

Avoid having the reference reservoir in the middle of such a group.  For some

uncooperative systems, several choices may be necessary to achieve convergence

(assuming no errors in the system description and that the system description

reflects realistic physical conditions).


     The second technique is to specify the fraction of supply values closer to

what the final values actually will be. (A little engineering judgement can

help here.) Several sets of initial conditions may have to be tried.  When

doing this, any nodal pressure settings should first be removed (cf. section

6.2.5).



6.2.3  Fragmented Network

     The distribution system must be totally connected, i.e., not consist of

two or more disconnected groups of nodes.  If ALLWET prints the message

          NETWORK FRAGMENTED -- LOOK FOR ERRORS IN INPUT DATA

and ceases calculations before printing the bandwidth and number of loops in

the system, then this condition has occurred.  ALLWET will indicate the

offending node if a single node is isolated from the system.  Otherwise, the



                                  6-7






user may need to review the nodal connections throughout the whole system.


     Section 6.2.7 explains how check valves and PRVs can fragment the

network.



6.2.4  H-Q Curve Problems

     Section 2.5 and Figure 2 emphasize that H-Q curves must decrease in H as

Q increases (negative slope).  During an analysis, if ALLWET indicates that

the slope of a H-Q curve is not negative, this condition is being violated

somewhere along the H-Q curve.


     This warning can be disregarded if the problem converges.  In this case,

the equation fitted to the points of the offending H-Q curve had a negative

slope within the operating range of the pump.  Figure 7 illustrates how this

might arise.


     However, if the calculations do not converge, then the positive slope may

have caused the problem.  This means that during the course of the iterations,

the calculated flow for the pump was not within the pump's normal operating

region.  First, check the offending pump curve to insure that it is correct.



Then take the remedies for nonconvergence suggested in section 6.2.2.  Respec-

ifying the pump curve with more points on the H-Q curve might also help.  This

will result in a tighter fit, and either eliminate or flatten the portion of

the curve with the positive slope.











                                  6-8







Figure 7   Inadvenently Misshped H-Q Curve
             (Fitted curve to shown poits will have negative slope for
              lower values of Q)

     .
     .
     .                     O       O
     .                                       O
     .
     .
     .
  H  .
     .                                                  O
     .
     .
     .                                                       O
     .
     .
     .
     .                                                           O
     .
     .
     .
     .. . . . . . . . .  . . .  . . . . . . . .  . . .  . . . .  .

                                   Q






























                                  6-9







6.2.5  Pressure Node Problems

     Pressure nodes should be used sparingly.  Fixing the nodal pressure sets

the HGL at the node.  A one psi change in a fixed nodal pressure changes the

HGL by approximately 2.3 feet, so that a five or ten pound change in pressure

at a node greatly affects the flow at that node and, therefore, the flow pat-

tern throughout all, or at least part, of the system.  In contrast, a moderate

variation in the demand at a node generally does not cause radical changes in

pressures throughout the system.  Thus, when having problems, analyses of dis-

tribution systems should be performed with flows rather than pressures fixed at

all nodes without reservoirs.  Once a solution is obtained, pressures can be

changed from the values ALLWET computed in the initial analysis, but large

changes in flow patterns from doing this may sometimes occur.  Having several

nearby pressure nodes sometimes hinders convergence, especially if the pres-

sure settings are physically inconsistent.  Pressure nodes sometimes interact

unfavorably with pumps or valves.  Sections 6.2.6 and 6.2.7 explain this.




6.2.6 Booster Pump Induced Problems

     The deployment of booster pumps sometimes causes difficulties.  Make sure

all pipes having booster pumps are correctly oriented.  If the pressure at a

pressure node is set too high, flow at the node will be negative (into the

system), possibly forcing water backwards through a nearby booster pump

station.  ALLWET resents this and might not converge.  Either the pressure at

the node should be lowered, or a demand, rather than a pressure should be fixed

at that node.  An incorrectly sized or unneeded booster pump station might also

result in reverse flow through the pump station.  In this case, the H-Q curve

should be revised or the pump station removed.  Often a pump station might

become unnecessary during an analysis of the late night operation of a

distribution system.  Nonconvergence and near zero or negative last iteration
                                  6-10






flows in those pipes having booster pumps indicate that booster pump deployment

problems may exist.






















































                                  6-11






6.2.7 Check Valve and PRV Induced Problems

     During flow-pressure calculations, ALLWET temporarily removes from the

system pipes having reversed flow through check valves and PRVs.  If such

removals fragment the system into two or more disconnected sets of nodes,

ALLWET may have trouble finding a correct solution.  (To find out if this

occurred, either (i) note the messages during the calculations indicating which

pipes were deleted, or (ii) print the output conditions of all PRVs and all

check valves, noting the pipes with zero flow.)



     Two scenarios an cause the activation of PRVs or check valves to fragment

the system (assuming the correctness and reasonableness of all input data):


(1) One of the disconnected sets has neither a reservoir or pressure node.  In
this case, ALLWET will usually panic and cease the calculations, claiming that
the network is fragmented. (All the nodes in such a disconnected set have only
flow constraints.  ALLWET cannot determine any heads in such a set, so it
panics.)  This is usually caused by a node with a large negative fixed demand
(i.e, into the system) forcing flow backwards through the valve.  Lowering the
magnitude of the negative demand, or setting the pressure (but not too high!)
rather than the flow at the node should eliminate the problem.

(2) All of the disconnected sets contain at least one reservoir or pressure
node.  One might hope that in this case, ALLWET would be smart enough to solve
each system separately.  Unfortunately, ALLWET's algorithms are designed to
handle only a single connected system.  ALLWET normally tries to connect
reservoirs and pressure nodes with imaginary pipes (dubbed pseudolines) in
order to form pseudoloops which used to balance the heads in the system.  When
reservoirs and pressure nodes end up in disconnected sets, no real pipes exist
to complete the pseudoloops.  ALLWET reacts by converting some reservoir or
pressure node to a fixed demand node.  This converts a head constraint to a
flow constraint, with the result that the nodes in the set containing the
converted node will have proper flows but not proper heads.  All this is really
telling you that given the fixed heads, flows and pump curves specified, the
system is functioning as two separate systems and should be solved as such.

     If during an analysis ALLWET prints that a PRV may be backwards, the

calculated path from the associated reservoir to the upstream node of the

valve's pipe crossed through the valve. (See the discussion of PRVs in section

2.6.) First, check that the valve indeed is correctly oriented.  If not, the




                                  6-12






endnodes of the valve's pipe must be reversed.  If the valve orientation is

correct, and if the analysis converged, the results may be invalid.  Therefore,

the HGL at the endnodes of the valve's pipe (from the node output) should be

checked for consistency with the head loss both across the pipe  (from the pipe

output) and through the valve (from the PRV output).  If the analysis produces

invalid answers, fails to converge, or terminates prematurely, then the

associated reservoir must be corrected to meet the requirement of section 2.6

and the analysis repeated.  If a fragmented network occurred, the advice of the

previous paragraph applies.




6.2.8 If All Else Fails

    Call or write ALLWET's author.  If you write, including as many of the

following which are appropriate will help obtain a quicker resolution of your

problem:

(1) A written description of the problem

(2) A sketch of the water distribution system

(3) A diskette with the offending input data file.

(4) A printout demonstrating what went wrong.
    To print out an interactive ALLWET session,

    (a) While holding down the Ctrl key, depress the PrtSc key.  This toggles
        your computer to duplicate on the printer everything which appears on
        the screen.

    (b) Start ALLWET, and enter the commands which demonstrate your problem.

    (c) Upon completing the session, again depress the PrtSc key while holding
        down the Ctrl key.

(5) If the problem was with the STORE command, a copy of the file which was
    incorrectly stored.








                                  6-13