280409 isaret presentation slides

8/6/2019 280409 ISARET Presentation Slides

1/45

ISARET 2009

April 28th - 29th 2009

Data Monitoring and

Reconciliation for RefineryHydrogen Networks

by

Siti Rafidah Ab. RashidFaculty of Chemical Engineering

Universiti Teknologi MARA

[email protected]


2/45

ISARET 2009

April 28th -29th 2009

Definition of Terminologies

Data Reconciliation

an adjusting process of data to satisfy

process model constraints (material and

energy balances)

it estimates process variables by adjusting the

measurements and improves the accuracy of

the process variables


3/45

ISARET 2009



Systematic Errors

the undetected mistakes that cause a

measurement to be very much farther from

mean measurement

caused by fouling of the sensors, wear and

tear, solid deposition on the probe, corrosion

on the sensors, miscalibration and instrumentmalfunction


4/45

ISARET 2009



Refinery Hydrogen Network

the distribution path of the supply and demand of

hydrogen in a refinery

F-304

F-304 D-304

D-304

D-407

D-407 D-404

D-404D-409

D-409D-408

D-408

C-301A

C-402

C-401

FT903

FT004

FT035

FT025

FT025

D-305

D-305

PV361

PV361

FT011

FT028

FT028

FT049

FT

014

FT

014

N2Not Compensated

Not Compensated

Not Compensated

Not CompensatedNot CompensatedT & P

Compensated

Not Compensated

D-410

D-410 D-418

D-418

38oC 38oC 38oC

From E -408

To H -403

Vent

Flare

To D -2102

To D -302

From D -302

To D -408

38oC

73oC

38oC 38oC 44oC

Chloride

absorbedFrom

U1400 To

D-808

From

E-185

T & P

Compensated

38oC

73oC

38oC

44oC

38oC 38oC

70oC

38oC40oC

FI001

Vent

FromHydrogenHeader


5/45

ISARET 2009


Background

The ISSUES are

Imbalance & Inconsistent Overall Material Balance

- Measuredstreams are corrupted with errors duringmeasurement, processing and transmission of themeasured signals

- Some streams are not been measuredbecause high

accuracy flowmeters are usually expensive


6/45

ISARET 2009


Background (cont.)

FT

FT

FT

FTFT

FT

FT

Contains Error

Contains Error

Unmeasured

Unmeasured

HYDROGEN

NETWORK

Imbalance &

InconsistentOverall

Material

Balance


7/45

ISARET 2009


HOW to Overcome?

Data monitoring and reconciliation

(cheaper & simpler)

improve the accuracy of measurements

improve plants profitability and flexibility

Background (cont.)


8/45

ISARET 2009


Problem Statement

Strict environmental regulations & heavier crude oilsupply increases hydrogen demand

Hydrogen Network Management identifies the best routeto an optimised hydrogen network

However, not all process data are reliable or measured Therefore, there is a need to perform Data Monitoringand Reconciliation which will help the refiners toexecute hydrogen network mgmt effectively

In this work

A linear data reconciliation program is developed

&

an existing oil refinery hydrogen network is selected as a case study


9/45

ISARET 2009


Objectives of the Study

The objectives of this project are to:

develop a systematic approach of data

reconciliation

apply the developed technique to a case

study of a refinery hydrogen networks


10/45

ISARET 2009


Scope of the Study

This work only considers:

reconciliation of data for steady state

process single variable or linear system

(flowrate)

material balance as constraint


11/45

ISARET 2009


Previous Works

Sanchez & Romagnoli (1996), utilized the QRfactorization to solve linear & bilinear datareconciliation problems. It allows the problem todecompose into lower dimension sub-problems.

Ripps (1965) continued the work of Reilly and

Carpani (1963) in systematic error detection.They defined a statistical method namely asGlobal Test.


12/45

ISARET 2009


Previous Works (cont.)

Tariq (2006) has done a linear and

bilinear steady state data reconciliation on

a refinery hydrogen network using QR

Decomposition technique. He usedMATLAB as the calculations tool.


13/45

ISARET 2009


Methodology

S4: Construction

of matrix Ax & Au

S1: Steady state,linear system

S3: Measured vs

Unmeasured etc.

S7: Method by Ripps

(1965)

S5: Method by

Sanchez & Romagnoli

(1996)

S2: Flowrates and

variance are data

available

S6: Method by

Narasimhan &

Jordache (2000)

ISARET 2009


14/45

ISARET 2009


The Case Study:


A refinery hydrogen network encountered a problem ofimbalance or inconsistent in material balance around thesystem. The study was initiated in 2004 and datareconciliation was advised.

This is crucial for future projects such as future plantdebottlenecking, hydrogen recovery and optimisation inits consumption and production.

The error-free flows are desired to give true picture ofthe network.

This particular hydrogen network is just like any othermodern refineries hydrogen networks. Is has 2production units and 7 units that consume hydrogen.

ISARET 2009


15/45

ISARET 2009


The Case Study:


In this study, only one producer and oneconsumer are considered.

The producer chosen for this exercise isnaphtha catalytic reformer denoted as Unit 400.

The hydrogen produced from this unit has thepurity of 85 to 88 mole%. This gas goes to thehydrogen header and then distributed to itsconsumers.

The selected consumer in this case is dieseldesulphurisation unit, which consumes hydrogento remove sulphur compound impurities.

ISARET 2009


16/45

ISARET 2009


Refinery Hydrogen Networks

F-304F-304 D-304D-304

D-404D-404

C-301A

C-401

FT025FT025

D-305D-305

PV361PV361

FT011FT011

FT028FT028

FT049

FT

014

FT

014

Not Compensated

Not CompensatedNot Compensated

T & P

Compensated

Not Compensated

D-410D-410 D-418D-418

From E-408

To H-403

To D-2102

To D-302

From D-302

To D-408

73oC

38oC 38oC 44oC

Absorbed FromU1400 To

D-808

From

E-185

T & P

Compensated

73oC

38oC

44oC

38oC 38oC

70oC

38oC 4

0oC

From Storage

FI001

FT035

FT004

Not Compensated

FT903

Not Compensated

From

Hydrogen

Header

ISARET 2009


17/45

ISARET 2009


Methodology

S4: Construction

of matrix Ax & Au

S1: Steady state,linear system

S3: Measured vs

Unmeasured etc.

S7: Method by Ripps

(1965)

S5: Method by

Sanchez & Romagnoli

(1996)

S2: Flowrates and

variance are data

available

S6: Method by

Narasimhan &

Jordache (2000)

ISARET 2009


18/45

ISARET 2009


Step 1: Identify & Define ProblemSimplified Block Diagram of H2 Networks

M01

M04

M28M07C06S05C27 M10 C11 S13 M14

M12

P02

C03

S08 S09

S25M24C23M22S21

P26

S01

S11

S08

S07

S05S04

S02

S10

S09

P29

S24

S23

S26

S25

S28

S50

S49

S48S47S45S43S41

S39

S44

S42

S21S18S53

S15

S12 S13 S52 S16 S17 S19 S20

S22

S27

S51

S46

M01

M04

M28M07C06S05C27 M10 C11 S13 M14

M12

C03

S08 S09

S25M24C23M22S21

P26

S01

S11

S08

S07

S05S04

S02

S10

S09

P29

S24

S23

S26

S25

S28

S50

S49

S48S47S45S43S41

S39

S44

S42

S21S18S53

S15

S12 S13 S52 S16 S17 S19 S20

S27

S51

S46

FT

028

FT

014

FT

049

FT

035

FT

004 FT

903

FT

011

FT

025

FT

001

ISARET 2009


19/45

ISARET 2009

April 28th -29th 2009Step 2 & 3 : Analyse Available Plant

Data and Classify VariablesMEASURED VARIANCE ASSIGNED

VALUES VALUES

1.219 0.000016729 027.63 0.009089739 0

1.219 0.000016729 0

1.219 0.000016729 0.00701

0.01 0.000354655

1.209 0.000354655

26.412 0.009210576

26.412 0.009210576

2.476 0.0006212972.228 0.000621297

2.228 0.000621297

0.2476 0.000621297

3.976 0.000621297

3.933 0.000354655

0.0437 0.000035466

2.724 0.000354655

28.55 0.00035465525.416 0.000354655

1.728 0.000160437

0.06998 0.000002093

1.658 0.000146515

15.332 0.007737293

0.317 0.000003675

16.91 0.010339947

1.895 0.000621297

2.476 0.000621297

1.457 0.000621297

y = Measured

Values

unr =assigned

values

var = variances

ISARET 2009


20/45

ISARET 2009


Step 4 (a): Process Model as ConstraintsElements of Measured Variable Matrix,Ax

Ax = S01 S02 S04 S05 S07 S08 S09 S10 S11 S12 S13 S15 S16 S17 S18 S19 S20 S21 S22 S23 S26 S42 S48 S49 S50 S52 S53

M01 1 -1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

P02 -1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

C03 0 0 -1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

M04 0 0 0 -1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

S05 0 0 0 0 0 0 0 0 1 -1 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

C06 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

M07 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0

S08 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 -1 0 0 0 0 0 0 0 0 1 0 0 0

S09 0 0 0 0 0 -1 0 0 0 0 0 0 0 1 0 -1 0 0 0 0 0 0 0 0 0 0 0

M10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 0 0

C11 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 1 0 -1 0 0 0 0 0 0 0 0

M12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0

C13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 -1 0 0 0 0 0 0

M14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0

S21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0

M22 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

C23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

M24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

S25 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0

P26 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 1 1 -1 0 0

C27 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

M28 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 1 1

P29 0 0 0 0 0 0 -1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

23 units and 27 measured streams

ISARET 2009


21/45

ISARET 2009


Example of process model

construction

For example, for the second unit P02, the

element of matrixAxcan be written as:

Ax= [-1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

Input flow: S04

Output flow: S01

The other 25 streams are not associated to P02.

ISARET 2009


22/45

ISARET 2009


Step 4 (b): Process Model as ConstraintsElements of Unmeasured Variable Matrix,Au

Au = S24 S25 S27 S28 S39 S41 S43 S44 S45 S46 S37 S51

M01 0 0 0 0 0 0 0 0 0 0 0 0

P02 0 0 0 0 0 0 0 0 0 0 0 0

C03 0 0 0 0 0 0 0 0 0 0 0 0

M04 0 0 0 0 0 0 0 0 0 0 0 0

S05 0 0 0 0 0 0 0 0 0 0 0 0

C06 0 0 0 0 0 0 0 0 0 0 0 0

M07 0 0 0 0 0 0 0 0 0 0 0 0

S08 0 0 0 0 0 0 0 0 0 0 0 0

S09 0 0 0 0 0 0 0 0 0 0 0 0

M10 0 0 0 0 0 0 0 0 0 0 0 0

C11 0 0 0 0 0 0 0 0 0 0 0 0

M12 -1 1 0 0 0 0 0 0 0 0 0 0

C13 0 0 0 0 0 0 0 0 0 0 0 0

M14 0 0 -1 1 0 0 0 0 0 0 0 0

S21 0 0 0 0 1 -1 0 0 0 0 0 0

M22 0 0 0 0 0 1 -1 1 0 0 0 0

C23 0 0 0 0 0 0 0 0 -1 -1 0 0M24 0 0 0 0 0 0 0 0 1 0 -1 0

S25 0 0 0 -1 0 0 0 0 0 0 1 0

P26 0 0 0 0 0 0 0 0 0 0 0 0

C27 0 0 0 0 0 0 0 0 0 1 0 -1

M28 0 0 0 0 0 0 0 0 0 0 0 0

P29 0 0 0 0 0 0 0 0 0 0 0 0

12 X 27 (12 units and 27 measured streams).

ISARET 2009


23/45

ISARET 2009


Step 5: QR Decomposition

A decomposition of general rectangular matrixA, defined as:

AP = QR

where

Q is an orthogonal matrix

Ris an upper-triangular Pis a permutation matrix. Permutation matrix is a matrix that has

exactly one entry 1 in each row and each column and 0's elsewhere

In this work, Householder transformation technique is chosen

Alternatives are Givens transformations and Gram-Schmidt

orthogonalisation method

It is chosen because it is stable and easy to code in Visual Basic

Programming (Gunter and Van De Geijn, 2001) & QR decomposition was

noted as the most computationally efficient (Kelly, 1999)

ISARET 2009


24/45

ISARET 2009


Step 5: Linear Data Reconciliation

Method by Sanchez and Romagnoli

(1996) is applied.

Where,

is reconciled values

Pis permutation matrix

y is the measured flowrate

x

ISARET 2009


25/45

ISARET 2009


Step 6: Estimate the Unmeasured

Flowrates

Method by Narasimhan and Jordache (2000) isapplied.

Where,

R1, R2are subsets of matrix R

Q1 is subset of matrix Q

is reconciled measured variable

unris assigned valuesx

x

ISARET 2009


26/45

ISARET 2009


Step 7: Systematic Error Detection

Method by Ripps (1965) is applied namely as GlobalTest.

Where, ris the vector of balance residuals, which is given by

r= Ay c

Ais the linear constraint matrix,Ax

ccontains known coefficients and for linear flowprocesses, c = 0

Vis variance-covariance matrix

ISARET 2009


27/45

ISARET 2009



(cont.)

is equal to the sum square of the

differences between the reconciled and

measured values. (Narasimhan and

Jordache, 2000).

= ( - y)2x

ISARET 2009


28/45

ISARET 2009



(cont.)

Under Ho, the above statistic follows a distribution with degrees of freedom,

where is the rank of matrixA. If the test criterion chosen is (where it is the

critical value of chi-square distribution at the chosen level of significance) then

Ho is rejected and a systematic error is detected if . also be

written as critical value of global testing, c.

2

2

In this work, theDOF = 23

(no. of units in the

system)

v = 10%

(commonly used)

c = 32.01Systematic

error is

detected

as = 225.5

ISARET 2009


29/45


Results and Discussions

There are some differences between the results obtained and Tariqs (2006) The average differences is 0.9%. This shows the VBA Excel program

developed is acceptable and resulting very small differences only.

Tariq (2006) used MATLAB as his calculations tool

Full QR Decomposition which produces an upper triangular matrix R of the samedimension as A, and a unitary matrix Q so that A=Q*R.

This work uses VBA Excel as the calculations tool

Economy-Size QR Decomposition is applied which it only computes the first ncolumns ofQ and R is n-by-n for m n matrix A.

Similar to Tariqs, this program detected the presence of systematic error butdid not locate the error, which requires other techniques, beyond the scope ofthis project, to improve the data.

ISARET 2009


30/45


Results and Discussions (cont.)

Comparison of Linear Data Reconciliation Results

0

5

10

15

20

25

30

1 3 5 7 9 11 13 15 17 19 21 23 25 27

Stream No.

Flowrates(ton/hr

)

VBA Excel

MATLAB

Possibly contains systematic error asdetected by the program developed.

Due to the scope of this work, the

program is detecting the presence of

systematic error of the system but

unable to identify the causes of the

error.

ISARET 2009

R lt & Di i ( t )


31/45

April 28th -29th 2009Results & Discussions (cont.)

Stream No. StreamName

Measured Flow Rates(ton/hr)

Reconciled Flow Rates (ton/hr) Variance(ton/hr)

1 S01 1.2190 1.2190 0.00002 S02 27.6300 27.6300 0.0000

3 S04 1.2190 1.2206 -0.0016

4 S05 1.2190 1.2021 0.0169

5 S07 0.0100 0.03324 -0.0232

6 S08 1.2109 1.5314 -0.3205

7 S09 26.4120 26.4120 0.0000

8 S10 26.4120 26.4120 0.0000

9 S11 2.4760 2.7048 -0.2288

10 S12 2.2280 2.4570 -0.2290

11 S13 2.2280 1.9992 0.2288

12 S15 0.2476 0.2478 -0.0002

13 S16 3.9760 3.0867 0.8893

14 S17 3.9330 4.4408 -0.5078

15 S18 0.0437 0.0945 -0.0508

16 S19 2.7240 2.7240 0.0000

17 S20 28.5500 26.9830 1.5670

ISARET 2009


32/45


Results & Discussions (cont.)

18 S21 25.4160 26.9830 -1.5670

19 S22 1.7280 1.7280 0.0000

20 S23 0.0700 0.0700 0.0000

21 S26 1.6580 1.6580 0.0000

22 S42 15.3320 15.3320 0.0000

23 S48 0.3170 0.3170 0.0000

24 S49 16.9100 2.1060 14.804

25 S50 1.8950 1.8950 0.0000

26 S52 2.4760 2.2470 0.2290

27 S53 1.4570 1.4570 0.0000

Possibly contains systematic error

as detected by the program

developed. Due to the scope of

this work, the program is detecting

the presence of systematic error of

the system but unable to identify

the location of the error.

ISARET 2009


33/45


Conclusions

As per objective of this study, a systematicapproach of linear data reconciliation isdeveloped.

Linear data reconciliation was done in the VBA

Excel program using QR decomposition method. The program is able to reconcile measured

variables, estimate unmeasured variables anddetect the presence of gross error.

The results of the program were compared withMATLAB. This is essential to ensure themathematics is correct through out the work.

ISARET 2009


34/45


Future Works

This work can be extended by: Developing data reconciliation for bilinear and nonlinear

systems. Nonlinear system involves more than one variable, for instance,

combinations of two or more variables ie. flowrate, composition and

temperature. Energy and component balance can be considered for non-linear

system as constraints as addition to material balance.

Establishment of a program to identify the location of gross erroris also essential. The information can be used by engineers forscheduling instrumentation calibration and/or maintenance.

Current trend of data reconciliation also focuses on dynamicreconciliation. This can also be considered for future work.

ISARET 2009


35/45


End of Presentation

Thank You

ISARET 2009


36/45


ISARET 2009


37/45


Attachments

ISARET 2009


38/45


Attachment: GUI of the program

developed

ISARET 2009

A il 28th 29th 2009


39/45


Attachment: Algorithm of Data

Reconciliation Program

Start

Stop

Input is

unmeasured flow

matrix, Au

Solve QR decomposition of

unmeasured matrix, Au,

using equations (3.11)

through (3.14)

Output are Q and

R matrices of

decomposed Au

ISARET 2009

A il 28th 29th 2009


40/45


Attachment:Algorithm of

Data

ReconciliationProgram

Start

Input are:

measured flowrates matrix, Ax

unmeasured flowrates matrix, Au

Q and R matricesvalues of measurements, Y

variance of measurements, var

assigned values, unr

Reconcile the flowrates using the equations

(3.18) to (3.23)

Output are:

Reconciled flowrates, x

Estimation of unmeasured flowrates, u

Global test parameter, ?

Systematic error

is detected

No systematic

error is detected

? > ? ?2

Stop


41/45

ISARET 2009

A il 28th 29th 2009


42/45


Attachment: Variable Classifications

Redundant variable is a measured process variable that is over-determined if it can also be

computed from the balance equations and the rest of the measured variables.

Unmeasured variables can be grouped as observable or non-observable variables. Observable

variable is unmeasured variable that is determinable if it can be evaluated from the available

measurements using the balance equations.

ISARET 2009

April 28th 29th 2009Attachment:


43/45

April 28th -29th 2009Attachment:

Previous Works on QR Decomposition in

Data Reconciliation Swartz (1989) used the QR decomposition in the matrix projection toeliminate the unmeasured variables.

Madron (1992)proposed classifying measured & unmeasuredvariables of linear systems according to pre-established criteria ofrequired & nonrequired.

Sanchez & Romagnoli, (1996), utilized the QR factorization to solvelinear & bilinear data reconciliation problems. It allows the problem todecompose into lower dimension sub-problems.

Kelly (1998) used different matrix projection techniques & highlightedtwo simpler approaches to determine the matrix projection(introduced by Crowe et al. (1983)).

ISARET 2009

April 28th 29th 2009Att h t P i W k


44/45

April 28th -29th 2009Attachment: Previous Works on

Gross Error Detection

Reilly and Carpani (1963) formulated Global Test (detection test)which is the collective chi-square test of all the data and the univariate test for constraints, based on the normal distribution. Ripps (1965) proposed a method which eliminates the measurement that renders the largest

reduction in a test statistics until no test fails. Romagnoli and Sephanopolous (1981) developed a systematic strategy to locate the source of

gross error. The method also rectifies the gross and biased measurement errors in a chemicalprocess.

Almasy and Sztano (1975) presented a global test and measurement test that possessesmaximum power test when there is only one gross error in the measurements, and is called the

MPT. Mah et.al (1976) presentedThe Constraint and Nodal Test (detection and identification tests).

This method requires linear constraints and measured variables. Mah and Tamhane (1982) proposed the univariate measurement test, which examines each

measurement adjustment. The statistical test based on the adjustment distribution by which firstprocess data are reconciled.

Narasimhan and Mah (1987)proposed a test named Generalised Likelihood Ratios. GLR hasthe capability to identify the location of the error and differentiate the types of the error such asinstrument related error or process model related error.

Rollins and Davis(1992) introduced UBET (Unbiased Estimation Technique). UBET is limited tonormally distributed errors, steady state and linear constraints.

Tong and Crowe (1996) introduced Principal Component Analysis (PCA). This technique is a setof correlated variables is transform into a new set of uncorrelated variables. PCA is a veryeffective method for multivariate data analysis.


45/45

280409 isaret presentation slides

Documents