Tennessee Eastman Challenge Archive

Author

N. Lawrence Ricker
Professor Emeritus, Chemical Engineering
University of Washington
Box 351750
Seattle, WA 98195-1750
USA
ricker@u.washington.edu
 

Links to topics

Updated TE Code
New Simulink models of two decentralized control strategies
MATLAB 7.x Codes
Basic (original) TEC Code
Rivera Group's MATLAB Simulation
Optimal steady states
Nonlinear MPC
Simplified TE process
Decentralized control
Using the original TE code in MATLAB

General comments

  1. Many of the codes were written for Matlab 3.x and Matlab 4.x.  Only a few have been updated to work with newer versions.  I will consider updating on a case-by-case basis if there is sufficient demand.
  2. If you would like me to link to your own TE archive, send the URL and a description of your site.
E-mail  your comments and questions to  N. Lawrence Ricker
 

Updated TE Code (version:  January 23, 2015)

A new version of the c code has been developed to allow the use of variable-step integration methods in MATLAB/Simulink and other simulation codes.  The use of such methods with Downs and Vogel’s original Fortran code (or the c translation provided in Basic TEC Code) leads to inconsistent results.  The updated code has been tested with MATLAB/Simulink 2014b.

The updated code also has the following new features:

·         Documentation of most internal variables.  This violates Downs and Vogel’s intent to keep model details secret, but a careful reading of their original code allows one to decipher it.  As more than 20 years have elapsed since the problem’s release, it no longer seems necessary to keep the details obscure.

·         Additional disturbances and measurements.  These may be of use but are optional.

Note that disturbance switches (IDV values) must be supplied as model input signals, not parameters.  

For more details, download the following archive, and see its README.TXT file.  The archive contains a draft paper submitted to ADCHEM 2015 that explains the rationale for the updated model and other technical details.  The archive also contains several demonstrations of open-loop and closed-loop operation with variable-step-size integration.  These may be compared to similar demonstrations using the original code.

 temexd_mod.zip

New Simulink models of two decentralized control strategies

As of 2 December 2002, two new Simulink models have been added to the temex archive (see below).  For more details, download the archive, and see its README.TXT file.  Additional changes made on 24 February 2005.

 

MATLAB 7.x Code

Two of the key interface routines were revised in December, 1998, to operate under MATLAB 5.2 or greater in Simulink 2 or greater. Click the links below to download the corresponding Zip archive. See the README file therein for details on installation and use.  The following comments supersede the instructions in the README file.

NOTE:  these files were updated 23 February 2002 and tested successfully in MATLAB 6.1 (Release 12).

NOTE:  the temex  archive was updated 2 December 2002, and is compatible with MATLAB 6.5 (Release 13).  It was tested on 24 Feb 2005 and seemed to work with MATLAB 7.04 (Release 14).

NOTE:  the temex archive was updated on 2 February 2005.  The Simulink models for the Multi-loop strategies now require Simulink version 6.2 or higher.

temex

temex_tc

Basic TEC Code

Clicking HERE will download zip archive called tecode.zip.  Use Win Zip or equivalent to extract a directory called tecode containing the following files, all in ASCII format:

teprob.f

  • Fortran code provided by Tennessee Eastman

tecommon.inc

  • An "include" file needed to compile TEPROB.F

te_mex.f

  • Fortran code needed to generate the .MEX interface to Matlab.  FOR MATLAB 3.x ONLY!!

te_mextc.fv4

  • This modification of te_mex.f was written for Matlab 4.2c on a MacIntosh.  It has not been tested on other machines.  It includes PI control of the reactor temperature and separator temperature.  There are also modifications to the output vector.  See comments in the code for more information.  The state vector for this version has 52 elements.  The last 2 are the integrated errors for the 2 PI controllers.  You must initialize these properly for the system to be at steady state.

Back to Topics

Rivera Group's MATLAB Simulation

Marty Braun, working with Prof. Daniel Rivera (ASU, Chem Engr) converted Down and Vogel's code from Fortran to MATLAB.  This allows you to work in the convenient Simulink environment without having to compile a MEX file.  You can also modify the MATLAB code so as to make the simulation more amenable to your specific research goals.  

Here is a link to their site, where you can download the code:

http://www.eas.asu.edu/~csel/Software-TennEast.htm 

The disadvantages of this approach are:

  • It runs much slower than the equivalent MEX code (see "temex" above).  In one test involving a PC running MATLAB Release 12, the MEX version was about 100 times faster.
  • Numerical results will not match those of Downs and Vogel.  Braun was unable to duplicate the random noise and disturbance generation of the original code.  The results are very similar, however, and the simulation's overall accuracy has been endorsed by Downs and Vogel.

Back to Topics

Optimal steady states

Clicking HERE will download a zip archive called tables.zip.   Use Win Zip or equivalent to extract the following 3 files, which are in tab-delimited text (ASCII) format. It should be possible to read them into a spreadsheet program or text editor.  They contain both text (row labels and units) and numerical data, so you may need to re-format before reading into Matlab, etc.  The files give operating  conditions at 6 different steady state operating conditions described in the paper Optimal Steady state Operation of the Tennessee Eastman Challenge Process, N. L. Ricker, Computers & Chemical Engineering, Vol. 19, No. 9, pp.  949-959(1995).

table2.txt

  • First 38 state variables.  To generate the complete state vector (50) variables, you must append the 12 manipulated variables (from table4.txt) to the end of the states in table2.txt.

table3.txt

  • Output variables (41)

table4.txt

  • Manipulated variables (12)

To verify a steady state I suggest that you call subroutine tefunc (part of teprob.f) to calculate the derivatives for the given states and manipulated variables.  All derivatives should  be less than 1.e-3.  You can also use the te_mex function in Matlab.

Nonlinear MPC

The following zip archives contain Matlab and Fortran code described in the papers Nonlinear Model Predictive Control of the Tennessee Eastman Challenge Process, Computers & Chemical Engineering, Vol. 19, No. 9, pp. 961-981(1995), and Nonlinear Modeling and State Estimation for the Tennessee Eastman Challenge Process, ibid, pp. 983-1005(1995).  Click on the link to download the archive.  Use Win Zip or equivalent to expand it.

NOTE:  Since 2002, some researchers have had trouble getting this software to work.  This appears to be caused by changes in MATLAB.  I have been unable to correct this problem.  If you encounter problems with instability, etc., and are able to correct them, please let me know.

teest3.zip

  • Directory containing code related to the above papers.  After expanding the archive, see the file teest3.doc for more information on these files.

 
teest6.zip

  • Directory containing code related to a newer version of the teest3 model.  After expanding the archive, see the file example.doc for more information.

 
nl_mpc.zip

  • Code to implement nonlinear MPC.  After expanding the archive, see contents.txt for more information.

Back to Topics

 Simplified TE process

The following directory contains Matlab and Fortran code described in the paper MPC of a continuous, nonlinear, two-phase reactor, N. L. Ricker, J. Process Control, vol. 3, 109-123(1993).  Click on the link to download the archive.  Use Win Zip or equivalent to expand it.

te4.zip

  • Expand the archive, then see the file contents.txt for more information on the contents of the TE4 directory.

Back to Topics

Decentralized control

Control strategy simulation

See Simulink code available in the temex archive.

Zip archives

The following zip archives document the performance of the decentralized control strategy described in Decentralized control of the Tennessee Eastman Challenge Process, N. L. Ricker, J. Proc. Cont., Vol. 6, pp. 205-221(1996).  Each archive is a recording of the response to a particular disturbance (numbered 1-15 in the original problem description of Downs and Vogel).  Click on an archive to download it.  Then use Win Zip or equivalent to extract the data.

idv1.zip    Step in A/C feed ratio in stream 4
idv2.zip    Step in B composition in stream 4
idv3.zip    Step in D feed temperature (stream 3)
idv4.zip    Step in reactor cooling water inlet temperature
idv5.zip    Step in condenser cooling water inlet temperature
idv6.zip    Sudden loss of A feed (stream 1).  This is a tough one!
idv7.zip    Stream 4 header pressure loss.  (step change)
idv8.zip    Random variations in A,B,C compositions in stream 4 (another tough one).
idv9.zip    Random variations in D feed temperature
idv10.zip    Random variations in C feed temperature
idv11.zip    Random variations in reactor cooling water inlet temperature
idv12.zip    Random variations in condenser cooling water inlet temperature
idv13.zip    Slow drift in reaction kinetics (also difficult)
idv14.zip    Reactor cooling water valve sticking
idv15.zip    Condenser cooling water valve sticking

Data format

Click HERE to see a text file describing the data format of the above archives.  You can save this for future reference.

Using the TE code in Matlab

Click HERE to see a text file containing additional suggestions on using the Basic (original) TE code with MATLAB.
NOTE:  The information has not been updated since the release of MATLAB 5.  You may need to change
some aspects to make it work in the latest version of MATLAB.