RECONCILIATION OF PROCESS FLOW RATES FOR A

STEADY STATE MASS BALANCE ON A CENTRIFUGAL

G HUB BARD''^ AND D LOVE^,^

' ~ o n ~ a a t - ~ uSugar
l e t t Limited, PO Box 5, Maidstone, 4380
2 ~ a v i~d o v Process
e Engineering, 45 Kelvin Place, Durban North, 4051
3
Tongaat-Hulett Sugar Limited, Private Bag 3, Glenashley, 4022

Abstract justify this deeper investigation into an apparently simple

flow measurement problem.
When evaluating the performance of a centrifugal, the flows
and compositions of the streams entering and leaving the unit Evaluation of a continuous A-centrifugal a t Darnall
must be known. It is not necessary (nor easy) to measure all
flows and compositions but the unknown quantities can be Background
calculated by steady state mass balance from a minimum Continuous centrifugation of high grade massecuites is not a
number of measurements. If more than the minimum number new concept but has not been widely implemented in raw or
of measurements are available, a means must be found to refined sugar processing due to unacceptably high levels of
reconcile the inevitable discrepancies between different ways crystal breakage, high purity rise, high sugar moisture and
of calculating unmeasured quantities. Published mathematical
lump formation. Recently, several companies claim to have
techniques have described how to achieve the 'best fit' to the
reduced the above problems to the extent that high grade
available data and provide an adjusted data set which recon-
ciles the probable measurement errors with the requirements continuous centrifugals now offer an economically feasible
of mass balance. These methods were adapted to allow them alternative to batch machines. The economic advantages are
to be applied using a simple spreadsheet calculation. mainly due to reduced maintenance costs, lower capital cost
and reduced operator attention required. In order to evaluate
this emerging technology, Tongaat-Hulett decided to install a
Introduction prototype continuous A-centrifugal at Darnall Mill.
Sometimes the apparently mundane prompts a question of un- Installation details
expected depth. Answering the question can provide equally
The continuous A-centrifugal was installed parallel to the
deep insights with wider application. This was the situation
existing bank of batch A-centrifugals but with a separate feed
encountered in testing a new design of continuous A- line direct from the final A-crystalliser. Sugar from the
centrifugal at Darnall. The capacity of the centrifugal needed continuous A-centrifugal was discharged directly into the
to be quantified in terms of the quantity of massecuite that it
existing A-sugar screw and was mixed with the product from
was able to process. Massecuite flow is particularly difficult the batch A-centrifugals. Molasses from the continuous A-
to measure directly on a factory scale but is easily estimated
centrifugal was discharged into a separate molasses gutter to
from other flow measurements and laboratory analyses
allow flow rate to be measured.
(which are in any case required to analyse the centrifugal's
performance) by applying a simple mass balance. The continuous A-centrifugal was supplied with its own
programmable logic controller (PLC) and 'touch screen'
- -

The attraction of this approach is rapidly dispelled when a control panel. variables able to be controlled were massecuite
few sample calculations demonstrate that significantly dif- flow rate (based on motor current), centrifugal speed and
ferent answers for the estimated massecuite flow rate are wash water rates.
obtained, depending on the type of mass balance used and the
consequent sub-set of the available measurements which are MolassesJlow measurement
used in the calculation. This is a situation where a surplus of Molasses flow rate was measured by diverting the flow from
information highlights uncertainty rather than providing a the continuous A-centrifugal molasses gutter into a 200 litre
confirmation of results. A survey of the literature shows that tank supported on an electronic mass balance. The outlet,
this type of problem (known as the reconciliation of process located at the base of the tank, was connected to the molasses
flows) has been the subject of significant academic study. The pump via a flexible hose so as to minimise external forces
challenge was thus to find an appropriate way of applying the that would otherwise have been imposed on the tank by the
solutions of the generalised problems presented in the litera- pump connection. High and low level conductivity probes in
ture to the specific requirements of evaluating a continuous the tank were installed to start and stop the pump to achieve
A-centrifugal. A good practical solution to this problem could batch filling and emptying of the tank. The electronic balance
have much wider application in the sugar industry and would was connected to a computer via an RS232 connection

Proc S A@ Sug Technol Ass (1 998) 72

Reconciliation ofprocessJIow rates for steady state mass balance on centrifugal G Hubbard & DJLove

allowing mass and correspbnding time information to be Crowe (1988b) and by Howat in Perry and Green (1997).
continuously captured and recorded on a computer disk. Much of this work appears to have a strong academic rather
Information on the rate of filling of the tank for specified than practical bias and to understand the theory requires a
periods could be obtained fiom the stored data and hence familiarity with matrix algebra, multivariate statistics,
accurate mass flow rates could be determined. calculus and optimisation techniques, making it not easily
accessible to the average sugar technologist.
Test procedures
The theory is further complicated by the distinction of
The purpose of the investigation was to evaluate the problems into two different types, viz. linear and nonlinear
performance of the continuous A-centrifugal with regard to (Crowe, et a1 1983; Crowe, 1986). In this context, problems
capacity, sugar quality and molasses losses. Capacity was are only linear if both the concentration and the flow rate of a
quoted in terms of massecuite throughput rate, while both stream are known. A limitation to solving only linear prob-
sugar quality and molasses losses would be expected to vary lems would thus be a severe restriction on the usefulness of
as a function of massecuite rate. Hence it was essential to the technique of reconciliation, making the more advanced
determine the massecuite flow accurately to evaluate the techniques for nonlinear problems essential.
performance of the machine. Despite the complexity of the mathematics, the principles on
An example of a set of tests would be investigation of the which the reconciliation of process flow rates are based can
effect of varying massecuite throughput rate on sugar and be stated quite simply as follows:
molasses quality for fixed steam addition, massecuite quality,
wash water rates and wash position. Each test would require a) Identify and discard any measurements with gross errors.
analyses of massecuite, massecuite nutsch, molasses and b) Create an adjusted set of measurements which:
sugar in order to observe the effect of a particular variable on meets the requirements of a full mass balance
the quality and composition of the product streams, and to
solve mass balances for massecuite flow rate determination. takes into account the known accuracies of the
measurements (i.e. less accurate measurements are
Massecuiteflow rate determination adjusted more than the fairly accurate measurements)
Due to the practical difficulties and cost associated with W are optimal in terms of making the minimum
measuring massecuite and sugar flow rate directly, it was adjustments to the measurements.
decided to measure wash water and molasses flow rates and The issue of detecting gross errors although closely allied, is a
to determine massecuite and sugar rates by overall and subject on its own (Crowe, 1988a) and will not be addressed
component mass balances on the centrifugal. Some of the directly here.
balances which can be used to determine massecuite flow rate
are presented in Appendix 1. There will be an infinite number of sets of adjusted measure-
ments which will meet the requirements of full mass balance,
However, the mass balance problem was over-specified, due and to select the optimum set, the sum of squares of the
to the number of measurements and analyses available, adjustments must be minimised. To take account of the
resulting in there being several solution methods to the known accuracies of the measurements, each squared
problem. When applied to actual trial data, the massecuite adjustment must be multiplied by a weighting factor which
flow rate determined by the various conventional mass has been shown to be the inverse of the variance, C T ~of, the
balances differed by up to 25%. This resulted in significant measurement. Even approximate values are adequate
uncertainty when quoting centrifugal capacity. Reasons for (Beckman, 1982) and variances can be estimated from the
disagreement between the balances, and hence apparent simpler concept of a confidence interval (Kneile, 1995). This
violation of the conservation laws, are that: addresses the issue of allowing- larger adjustments to be made
-
the simplifying assumptions necessary for each mass to less accurate data. For example brix measurements with a
balance introduce a degree of error which is different for confidence interval of *0,45 percentage units should be
each component balance adjusted less than molasses flow measurements with a
the individual variables are subject to random analytical confidence interval of *5% in the process of achieving a set
and measurement error which differ in magnitude for of adjusted data which meets the requirements of a full mass
different analyses and instruments. balance.
It was this uncertainty in the estimation of the massecuite This reconciliation problem has been summarised mathe-
flow, because of the known limitations of using conventional matically by Beckman (1 982) as detailed in Appendix 2.
mass balance techniques, that prompted the investigation into Expressed in its mathematical form, the problem is termed a
a better method. constrained optimisation (maximisation or minimisation). In
general there are three methods of solving this type of
Theory of 'reconciliation of process flows' problem (Boas, 1966), viz. elimination, implicit differen-
tiation or the use of Lagrange multipliers. Beckman (1982)
The problem of reconciling the inconsistencies in process has used the technique of elimination in his approach to the
measurements when evaluated in terms of mass balances has reconciliation of process data whilst others (Kneile, 1995;
been addressed by a number of authors as summarised by Crowe, et al. 1983 and Crowe, 1986) have used Lagrange

Proc S Afi. Sug Technol Ass (1998) 72

Reconciliation ofprocess flow rates for steady state mass balance on centrifugal G Hubbard & DJ Love

multipliers. Despite the significant algebraic effort of these will make the sum of squares function insensitive to the
approaches, iterative numerical methods are still required by adjustments in the measurements, achieving poor, or no,
the method of Beckman (1982) and the nonlinear case of convergence in this respect.
Crowe (1986).
Given the complexity of these methods, the question arises as Application of the numerical reconciliation of process
to whether a simpler numerical technique, using the nonlinear flows (NRPF)technique to a continuous A-centrifugal
optimising capabilities of a modem spreadsheet, can be
The numerical reconciliation of process flows (NRPF)
devised.
technique was developed for the continuous A-centrifugal
A numerical technique for the reconciliation of trials due to uncertainty resulting fkom the wide range of
process flows on a spreadsheet massecuite flow rates obtained £tom the different
conventional mass balances (Appendix 1) using the same
Modem computer spreadsheets have the capabilities to per- measured and analytical data.
form nonlinear optimisation. The Core1 Quattro Pro for
Prior to applying the NRPF technique to normal test data, the
Windows spreadsheet (versions 6 and S), which was used in
accuracy and robustness of the technique was first evaluated
this work, provides such a function named 'Optimizer'. The
by the following methods:
user need only identify the cell which is to be maximised or
minimised and indicate which cells are to be altered to The NRPF and conventional mass balance solutions were
achieve this optimisation. There are also a number of compared against a simulated set of ideal data which
convergence parameters for the optimisation which can be satisfied the material balance constraints for a massecuite
altered to improve the speed and accuracy of the optimisation flow rate of 30 th.
if desired. The 'Optimizer' function also has the ability to The NRPF technique was compared against the con-
handle constraints if necessary. ventional mass balances using six simulated data sets
The spreadsheet documentation provides limited information containing known random errors added to ideal data.
on the mathematical methods used to achieve the nonlinear The NRPF and conventional mass balance solutions were
optimisation. In particular, there is no indication of how the compared against solutions from an overall mass balance
constraints are handled. The user thus needs to treat the which included measurement of sugar flow rate (which
'Optimizer' function much like a 'black box' which will give would not be available in normal test work). ,
the correct output if the input is correctly specified.
As part of this work, a technique was developed which Application ofthe NRPF Spreadsheet
provides a simple way of handling constraints within the The application of the NRPF spreadsheet program for flow
standard nonlinear optimisation without requiring the rate reconciliation is relatively simple and the method could
spreadsheet's ability to handle constraints. This 'engineering' be applied universally to any ,unit operation where steady
method of handling constraints is simply to add the weighted state mass balances apply. The program adjusts the measured
sum of squares of the adjustments (described above) to the and analytical data to satisfy the overall mass balance, dry
weighted sum of squares of the imbalances in the constraint solids balance and p01 balance for the continuous A-
equations (which should be zero). Minimisation of this new, centrifugal trials.
augmented, sum of squares function will thus force the Balances:
constraints to be met to an accuracy defined by the weighting
function. Appendix 2 shows how this concept can be ex- Overall: Tonswc+TonsH20-Tons,,,-Tonsmol = 0 Constraint (2)
pressed in the mathematical format used by Beckman (1982). Pol: PollwcTwc - Polsu,Tsug- P O ~ ~ , ~=, T
0 , ~Constraint
~ (3)
Although this method of handling constraints is not Dry Solids: DSwcTwc - DSsugTSug - DSmoITmoI =0 Constraint (3)
mathematically rigorous, it meets the engineering requirement
Note: Dry solids were used instead of brix due to the errors
of being 'good enough for all practical purposes'. For
associated with the optical nature of brix measurement.
example, the user can decide whether to force the overall
However, brix could just as well have been used. An
mass balance to converge to within 1 kg/s or 1 g/s, whichever
empirical relationship derived within Tongaat-Hulett Sugar
is accurate enough in terms of the precision and accuracy of
(lprivate communication) was used to estimate 'dry solids'
the measurements being used.
from brix and p01 measurements. The relationship is:
This 'engineering method' is much simpler than the rigorous
mathematical methods used by Beckman (1982), Kneile Dry Solids = Brix * (1 - 0,00066*(Brix - Pol)).
(1995) and Crowe (1986). It also has an advantage over using A Quattro Pro spreadsheet (Table l) was designed to apply
the built-in constraint handling of the Quattro spreadsheet in the NRPF technique to the continuous A-centrifugal. The
that the way that the constraints are handled is more trans- optirnizer (Tools-Numeric Tools-Optimizer menu option) is
parent to the user and can be manipulated if necessary by set up to adjust the Measured Values (Gj) (Spreadsheet cells
altering the weighting function. Specifically, the user should D4 .. D13) in order to minimise the Total Weighted Squared
select the smallest weighting function necessary to force the
constraints to converge to the required accuracy. Large
weighting functions will force the constraints to converge but 'RG Hoekstra, Tongaat-Hulett Sugar Ltd, PIBag 3, Glenashley, 4022

292 Proc S Afr Sug Technol Ass (1998) 72

Reconciliation ofprocessJIow rates for steady state mass balance on centrifugal G Hubbard & DJ Love

Error (SSQ2, per Eq. (5)) (Spreadsheet cell 521). The The Total Weighted Squared Error (SSQ2) (Spreadsheet cell
Weighting Factors (W4 and WxV)(Spreadsheet cells I4..113) J2 1) comprises the sum of the Weighted Squared Adjustments
were determined from the Standard Deviation values (SSQl, per Eq. (l)) (Spreadsheet cells J4..J13) and the
(Spreadsheet cells H4..H13) of the various analyses and Weighted Squared Imbalance of each mass balance
measurements. The standard deviations are estimated from (Spreadsheet cells J18..J20), per Eq. (5). In order to ensure
the confidence intervals, namely: that the material balance constraints were met, a Mass
Brix analytical error = It0,45% Balance Weighting Factor (WB) of 1 (Spreadsheet cell C16)
Pol analytical error = +0,20% was applied to each of the squared mass balances. This can be
Molasses flow error = +5% of actual measurement varied by the user to achieve the required convergence of the
Water flow error = *2% of actual measurement. balances on the adjusted data. As mentioned previously, this
value should be the minimum necessary to achieve the
Thus the weighting factors reflect these confidence intervals,
required convergence.
with large factors indicating high confidence in the data.

Table l. Example of the NRPF spreadsheet using an ideal data set with a massecuite flow rate of 30 tlh.

Comparison of the NRPF technique with conventional mass completely satisfied and has a maximum imbalance (error) of
balances using simulated ideal data approximately 0,5%.
The purpose of comparing the NRPF technique and the The ideal data set was used to calculate massecuite flows
conventional balances with a known massecuite flow rate was using the conventional mass balances and the NRPF
to select those balances which produced the most accurate technique. The results are presented in Table 2.
results and also to check the relative accuracy of the flow Five of the 11 solutions above have absolute errors of less
rates with the conventional balance flow rates. than 1%. This represents the difference between the calcu-
A set of simulated ideal data satisfying the overall, p01 and lated massecuite flow rate and the ideal flow of 30 t/h. A
dry solids material balance constraints, was generated based further three balances have solutions with absolute errors
on a massecuite flow rate of 30 tonlhour (Appendix 3, Table ranging from 1,78 to 4,03%, while the remaining three
5). Molasses brix was determined from dry solids using the balances all have errors greater than 5%. The reasons for any
correlation described above and hence the brix balance is not deviation from the ideal massecuite flow of 30 t/h are due to

Proc S A j - Sug Technol Ass (1998) 72 293

Reconciliation ofprocessfIow ratesfor steady state mass balance on centrzfugal G Hubbard & DJ Love
the simplifying assumptions made in solving each of the Comparison of the NRPF technique with conventional mass
balances. The NRPF solution ranked third in terms of balances using simulated data with known random errors
accuracy and had an absolute error of only 0,07%. However,
at this very low level of error the relative ranking is not If the degree of accuracy of measured data in a real situation
significant and is probably affected by round-off errors. was known, it would be a trivial task to select the mass
Based on the above comparison of the various mass balances, balance which would provide the 'best' solution. However, in
using ideal data, it was decided to investigate further only the case of the continuous A-centrifugal trial there will be
those which resulted in solutions with an absolute error of less random and unknown errors in the measured data. It was
l%, i.e. the balances which were greater than 99% accurate. therefore decided to investigate how robust each of the
These balances are ranked 1 to 5 in Table 2 and comprised accepted alternative calculation methods were when used on
four conventional mass balances and the NRPF technique. simulated data with known random errors.
In Appendix 4, six sets of simulated data containing normally
It is interesting to note that three of the best four conventional distributed measurement errors were calculated from the ideal
balances were all based on data which were not directly data of Appendix 3, and are shown in Table 6. Massecuite
analysed for, namely, non-p01 or non-crystal. This is most flow rates were then calculated (Table 7) using the accepted
likely due to the fact that non-p01 (%) and non-crystal (%) balances and the results were compared. The six random data
were determined by difference of pol and moisture (100-brix) sets as well as detailed solutions are presented in Appendix 4.
from 100. This could have an 'averaging' effect on the errors A summary of the results is presented in Table 3.
resulting from the simplifying assumptions in the
calculations. In Table 3, the mean values represent the average massecuite
flow rate calculated over the six random data sets while the
standard deviation indicates the variation of data about the
Table 2. Comparison of mass balances using an ideal data set.
mean. The absolute mean error is the difference between the
massecuite flow rate calculated by a particular type of balance
for the ideal data set (i.e. without errors), and the mean
massecuite flow rate of the six data sets containing random
errors. Each of the balance types were assigned rankings
based on absolute mean errors. Previous rankings determined
on ideal data are repeated for comparison.
It is evident from the rankings that the two most robust
massecuite flow determination methods, both in terms of
absolute mean error and standard deviation, are the NRPF
technique and Balance A2 (dry solids balance using molasses
and wash water rates). However all the balance types selected
show good agreement with massecuite flows calculated for
the ideal data.
Comparison of the NRPF technique and conventional mass
balances with an overall balance using actual experimental
data
In order to c o n f m the accuracy of the conventional mass
balances and the NRPF technique when applied to real data,
Note: Absolute error is the % difference between the calculated flow Sugar which would not be available during normal tests
rate and the ideal flow rate of 30 tonsthour. to evaluate the centrifugal, was measured in two special tests.
Table 3. Summary of comparison between mass balances using ideal data with known random errors.
Ideal data Six data sets containing random errors
Conventional mass
balances W C flow (th)
Standard Absolute mean
Ranking Mean Ranking
deviation error (%)
Balance A2 30,Ol 2 30,39 1,37 1,26 1
Balance D2 30,OO 1 29.09 2.12 3.03 5
Balance El 29,73 5 28,98 1,74 2,53 4
Balance E2 29,86 4 29,11 1,76 2,5 1 3
NRPF M/C flow rate 29,98 3 29,45 1,34 1,76 2
Mean 29,92
Standard deviation 0,11
Note: Absolute mean error is the % difference between the mean M/C flow rate, determined from the six random data sets, and the ideal data flow rate
calculated by that particular balance.

294 Proc S A) Sug Technol Ass (1998) 72

Reconciliation ofprocessjlow rates for steady state mass balance on centrzfugal G Hubbard & DJLove

This allowed an overall mass balance to be performed about 1 5 kg per tip. Hence, a 1% accuracy was assumed for recon-
the centrifugal enabling massecuite flow rate to be calculated. ciliation purposes.
The two runs were performed at different massecuite feed
Tonslwc = Tonsmol+ Tons,,, - TonsHlo.
rates over a period of 30 minutes each. Molasses and sugar
Overall mass balance does not require any laboratory analyses samples were taken and molasses flow rate was checked at
and hence eliminates any random analytical error. Sugar flow the start, middle and end of each run. This allowed the
rate from the continuous A-centrifugal was determined by massecuite flow rate to be calculated by the conventional
using a sugar servo-balance located immediately after the balances and the NRPF technique. These were then compared
dryer at Darnall and stopping all the batch centrifugals. The with the overall balance flow rates in Table 4.
servo balance sums batches of about 500 kg and is accurate to
I

Table 4. Comparison of mass balance types with an overall mass balance

using actual experimental data (with extra measurements).

Conventional balances Run l Run 2 Average error

Ranking
(flow = tlh) W C flow Error (Yo) MIC flow Error (%)
Overall balance 18,35 26,89
Balance A2 19,28 5,07 26,20 2,57 3,82 2
Balance D2 17,31 5,67 25,5 1 5,13 5,40 3
Balance El 17,68 3,65 24,49 8,93 6,29 5
Balance E2 17,78 3,11 24,62 8,44 5,77 4
NRPF MIC flow rate 18,18 0,93 25,85 3,87 ' 2,40 1

The reconciled flow rates using the NRPF method resulted in and those determined by the NRPF technique range between
the lowest average error of 2,4% from the massecuite flow 0,39 and 5,3 l%, the maximum difference is as high as 16,7%
rates determined using the overall balance. Each mass balance (Balance A2, Experiment 5 in Figure 1). This indicates the
has been ranked according to the average errors. Errors for significant level of error which could result from randomly
the conventional mass balances range from 3,82 to 5,77%. selecting a particular conventional balance and ignoring some
The above results confirm the robustness and hence the measurements.
reliability of the NRPF technique for application to actual Figure 1 also gives an indication of the relative bias of the
plant data. conventional balances relative to the NRPF technique.
Balance A2 flow rates have a positive offset increasing at
Reconciliation of actual data from the continuous higher flow rates. Balance D2 flow rates have a negative
A-centrifugal trials offset increasing (-ve) at higher flow rates. Balance El and
E2 are very similar with positive offsets at low flow rates and
The NRPF technique has demonstrated its accuracy and increasingly negative offsets at higher flow rates.
robustness in its application to simulated ideal data, simulated
data with known random errors and actual plant data with
extra measurements. Given this 'proof' in application and its
sound theoretical basis, it was selected as the best method for
estimating massecuite flow in evaluating the continuous A-
centrifugal. It is of interest to compare the results of this
method with what would have been estimated if the more
conventional methods had been used.
The data from a trial comprising six individual experimental
runs were used to calculate massecuite flow rates. Raw
experimental data is presented as Table 8 in Appendix 5. The -15
aim of this particular trial was to investigate the effect of 10 15 20 25 30 35
NRPF MIC Flow Rate (Tonshour)
increasing massecuite throughput on sugar and molasses
quality. The relative percentage difference between the
conventional balance massecuite flows and corresponding
-- Balance A2
-- Balance E l
+-

-o
Balance D2
Balance E2
NRPF flows for the six experiments are presented in Figure 1.
While average differences between massecuite flows deter- Figure 1. Comparison of mass balance types using actual plant data
mined by each of the accepted conventional mass balances from the continuous A-centrifugal.

Proc S Afi Sug Technol Ass (1998) 72

Reconciliation ofprocessflow rates for steady state mass balance on centrifugal G Hubbard & DJ Love

Conclusions APPENDIX 1
Mass balances for determining the massecuite feed rate to
Determination of massecuite flow rates, using the various
a continuous centrifugal.
possible conventional mass balances on a continuous A-
centrifugal, resulted in a significant degree of uncertainty due
Balance AI : Brix balance using molasses and wash water
to the differing solutions obtained using the same measured
flow rates
and analytical data. To overcome this, a numerical method,
with a sound theoretical base, was developed for the
reconciliation of the process flows.
Before applying this technique to the evaluation of the
Tonswc =
Bxmo~
(100 - moisture)1- TonsH20

performance of a continuous A-centrifugal, the technique was

evaluated as follows:
The accuracy of the NRPF techniques was tested against
conventional mass balances using a simulated ideal data
set with a known flow rate and proved to be 99,98 % Balance B1 : Combined brix and non-p01 solids balance
accurate. using wash water rate
The robustness of the NRPF technique was tested against TonsMjc =
simulated ideal data containing random errors and proved
to be as robust as the best of the conventional mass
balances. 100-moisture 100- moisture
The NRPF technique was tested against conventional mass
balances using actual plant data with extra measurements
and compared better with the overall mass balance Balance C l :Non-p01 solids balance using molasses rate
massecuite flow rate than any of the conventional balances.
The NRPF technique is easily applied to actual experimental
data and provides a more reliable estimate of massecuite flow.
The technique has been shown to give results which are at
times significantly different from more conventional mass
balance methods. Balance D1 : Non-p01 solids balance including non-p01 in
The numerical reconciliation of process flows technique is sugar
simple and can be universally applied to any unit operation Tons~lc
where steady state mass balances apply. -

REFERENCES TonsHl0 - PoEug)+Tonsmol(Bxrn0l- Polrno~

- Bxsubsug+ P O E ~ ~ )
(BXW - P01w - Bxsug + P0Gug)
Beckman, JR (1982). Data adjustment for non-reactive batch or
steady-state processes. Chem Eng Commun 15: 357-365.
(4)
Boas, ML (1966). Mathematical Methods in the Physical Sciences.
John Wiley, New York: 142-143. Balance E1 :Non-crystal balance
Crowe, CM (1986). Reconciliation of process flow rates by matrix
projection, Part 11: The nonlinear case. AIChE Journal 32(4):
616-623
Crowe, CM (1988a). Recursive identification of gross errors in
linear data reconciliation. AIChE Journal 34(4): 541-550.
Crowe, CM (1988b). Progress in the reconciliation of process flow
measurements. Proc SA Inst Chem Engrs, 5th National Meeting,
August 1988. The above balance equations can be converted to dry solids
Crowe, CM, Garcia Campos, YA, Hrymak, A (1 983). Reconciliation balances rather than brix balances by replacing the brix terms
of process flow rates by matrix projection, Part I: Linear case. with an empirical relationship developed within Tongaat-
AIChE Journal 29(6): 881-888. Hulett Sugar (Balances A2 to E2 respectively). The relation-
Kneile, R (1995). Wring more information out of plant data. Chem ship is:
EngVol 102(3): 110-116.
Perry, RH and Green, DW (1997). Perty's Chemical Engineers' Dry Solids (DS) = Brix (1 - 0,00066 (Brix - Pol)).
Handbook, Seventh Edition, McGraw-Hill, New York, 30-1 - 30-
36.

Proc S Afi Sug Technol Ass (1 998) 72

Reconciliation ofprocessflow rates for steady state mass balance on centrljhgal G Hubbard & DJLove

APPENDIX 2.
Mathematics of process flow reconciliation.

Beckman (1982) has summarised the reconciliation of process by assuming that &a is the 95% confidence interval of a
flow problem mathematically as minimising the weighted normal probability distribution. Generally, for the full
sum of squares function, SSQl, which is given by: mathematical analysis of the problem, it is assumed that the

s s ~=i
M
1-(cj
GJf
j-1
WF.
- F,r+X M N

j=l i=1
WX..
+jYji
yji
- xji)2
measurement errors are independent of each other.
The numerical technique for dealing with the constraints
using conventional, unconstrained, optimisation can be
expressed mathematically in the format used by Beckrnan
Where: .. (1982) as:
SSQl = Sum of squares function to be minimised
W< = 1/o; =
Weighting factor for stream j flow rate
Wxji = l/o:, = Where:
Weighting factor for stream j composition
Gj = Measured flow rate of stream j SS@ = augmented sum of squares function to be
minimised
yji = Measured composition of component i in stream j
WB = weighting factor for the squared imbalance of
F, = Adjusted flow rate of stream j constraints
xji = Adjusted composition of component i in stream j and the rest of the variables are as previously defined.
M = Total number of flow streams The constraint for forcing all the individual component
N = Total number of components fractions of a stream to sum to unity can be neglected in most
o = Standard deviation of measurement sugar factory applications as in the equation above. This is
because there is usually a pseudo component called non-p01
o2 = Variance of measurement.
or non-sucrose which is never measured but is simply
calculated by difference. The component fractions can thus be
Subject to the following constraints: forced to sum to unity by leaving non-p01 out of the
optimisation and simply calculating it by difference once the
optimisation has been done.
The mathematical descriptions of the reconciliation problem
i.e. the sum of the mass flows in (taken as positive) equals the can probably be more easily understood by referring to the
sum of the mass flows out (taken as negative). spreadsheet example given in Figure 1 where the variables
and subscripts can be interpreted as follows.
Subscript 'j' refers to the different streams with
j =l massecuite
j = 2 water
i.e. for each .i-th. component of all N components the sum of j =3 molasses
the mass flows in equals the sum of the mass flows out. j =4 sugar.
N
C X,, =l Subscript 'i' refers to the different constituents of a stream
i=S(j )
1ljlM i= l p01
i = 2 brix or (100-moisture).
where SG) is the index of the first non-zero valued component
in stream j. Thus, for example
i.e. the adjusted compositions of all the components in each G2 is the measured mass flow of water
stream must sum to unity (with the adjusted values of all ~ 3 2 is the measured brix of molasses
compositions measured as zero being forced to zero). is the adjusted mass flow of water
F2
Kneile (1995) described how the simpler concept of a confi- is the adjusted brix of molasses
~ 3 2
dence interval of *a can be used to estimate the variance, 02,
as: WF3 is the weighting factor for the adjustments to the
mass flow of molasses
Wx32 is the weighting factor for the adjustments to the brix
of molasses.

Proc S Afr Sug Technol Ass (1998) 72

Reconciliation ofprocessJlow ratesfor steady state mass balance on centrzfigal G Hubbard & DJ Love

APPENDIX 3.
Table 5. Ideal data generation based on fixed massecuite flow rate.
FEED to continuous centrifugal PRODUCT from continuous centrifugal
Massequite Sugar
Assume flow rate (th) 30,OO Assume 'wet' Pol % (ex centrifugal) 98,60
Assume massecuite brix % 92,50 Assume moisture % (ex centrifugal) 0,70
Assume massecuite Pol % 82,OO Sugar brix % 99,30
Massecuite DS % 91,86 Sugar DS % 99,25
Massecuite purity % 88,65
Crystal content % 60,61 Crystal balance
Crystal flow in M/C (th) 18,18 Assume crystal loss % 3,OO
Hence crystal flow in sugar ( t h ) 17,64 .
Assume sugar Nutsch purity %
Nutsch 72,07
(= W C Nutsch purity + 5 units)
Assume Nutsch brix % 82,OO Sugar crystal content % 96,79
Nutsch DS % 80,54 Hence sugar flow rate ( t h ) 18,22
Assume Nutsch PO]% 55,OO
Nutsch purity % 67,07 Molasses properties
Molasses brix % 67.26
Water Molasses DS % 66,34
Assume flow rate (t/h) 2,50 Molasses Pol % 46,46
Wash % MIC 8,33 Molasses purity % 69,07
Purity rise % 2,98

Balance check WC Water Molasses Sugar Imbalance

Overall mass balance 30,OO 2,50 14,28 18,22 0,0000
DS balance 27,56 0,OO 9,47 18,09 0,0000
Pol balance 24,60 0,OO 6,63 17,97 0,0000

APPENDIX 4.
Random data sets and mass balance solutions based on ideal data.
Table 6. Simulated ideal data with random known errors (actual massecuite flow = 30 tlh).

Standard Normal distribution of random data with ideal data as mean value
Ideal data Error
Data 1 Data 2 Data 3 Data 4 Data 5 Data 6
Moisture flow rate (th) 14,28 5% 0,7 1 13,80 14,55 13,60 12,64 14,80 14,14
Wash water rate (tlh) 2,60 2% 0,05 2,56 2,52 2,5 1 2,54 2,50 2,55
Massecuite Pol 82,OO 0,20 0,20 8 1,97 82,Ol 82,06 82,20 82,16 81,68
Massecuite brix 92,60 0,45 0,45 91,97 92,95 92,64 92,80 9 1,94 92,36
Nutsch Pol 66,OO 0,20 0,20 55,24 55,12 54,66 55,20 55,03 55,03
Nutsch brix 82,OO 0,45 0,45 81,89 8 1,70 8 1,92 81,73 82,15 82,89
Molasses Pol 46,46 0,20 0,20 46,3 1 46,5 1 46,35 46,55 46,68 46,56
Molasses brix 67,26 0,45 0,45 67,08 67,41 67,25 66,76 66,64 66,80
Wet sugar Pol (ex centrifugal) 98,60 0,03 0,03 98,59 98,61 98,61 98,58 98,55 98,61
Sugar moisture (ex centrifugal) 0,70 0,04 0,04 0,67 0,75 0,71 0,77 0,67 0,70
Calculated M/C dry solids 91,86 91,36 92,28 9 1,99 92,15 9 1,34 91,71
Calculated Nutsch dry solids 80,54 80,45 80,27 80,45 80,30 80,68 81,36
Calculated moisture dry solids 66,34 66,16 66,48 66,32 65,87 65,76 65,90
Calculated sugar dry solids 99,25 99,28 99,2 1 99,24 99,19 99,27 99,25

298 Proc S Aj? Sug Technol Ass (1998) 72

Reconciliation ofprocessJlow rates for steady state mass balance on centrzfigal G Hubbard & DJ Love

APPENDIX 4 (continued)
Table 7. Massecuite flow rate determination for simulated ideal data with random known errors.

APPENDIX 5.
Table 8. Raw experimental data.

Note: Measured and analytical data are presented in columns 2,3,4, 6 , 9 and l l .

Molasses Sugar
EXP Purity rise
No. Brix DS Pol Purity Pol Moisture DS Corr Pol
1 62,25 6 1,49 43,75 70,28 5,65 98,48 0,95 99,Ol 99,36
2 64,90 64,06 45,25 69,72 5,09 98,58 0,87 99,09 99,38
3 67,85 66,91 46,90 69,12 4,49 98,26 1,23 98,74 99,42
4 68,85 67,87 47,25 68,63 4,OO 98,06 1,18 98,77 99,18
5 68,25 67,28 46,75 68,50 3,87 97,89 1,46 98,50 99,28
6 69,20 68,19 47,OO 67,92 3,29 97,55 1,43 98,50 98,93

Note: Measured and analytical data are presented in columns 2,4, 7 and 9.
The duration of each experiment run was four minutes, with one minute between runs.
DS = Dry solids

Proc S AJi. Sug Technol Ass (1998) 72