(67a) A Systematic Approach to Alarm Design with Application to the Tennessee Eastman Problem

 Abstract
for 2017 AIChE Spring Conference

A systematic
approach to alarm design with application to The Tennessee Eastman Problem

JOSHIBA ARIAMUTHU VENKIDASALAPATHY*, COSTAS KRAVARIS and M. SAM MANNAN

Artie McFerrin Department of Chemical Engineering

Texas A&M University

College Station, Texas 77843-3122, USA

*+1(979)402-9823, joshvenkat03@tamu.edu

When
a process plant faces an abnormal situation, the alarm management system that
is meant to assist process operators, may hinder effective remedial measures by
simultaneously displaying several alarms, making it nearly impossible for the
operator to identify the critical variable that needs immediate attention. This
condition is known as 'alarm flooding'. This study focuses on efficient alarm
identification which is the first critical step in designing the alarm system
of a process plant.   

We
formulate a systematic approach to alarm identification wherein a subset of
measured variables is chosen for configuration in the alarm system. The approach
may be outlined as (i) Identification of potential hazardous scenarios, (ii)
Simulation of these scenarios in order to quantify process measurement response
time, (iii) Formulation of a mixed integer linear programming problem (MILP) that
is solved using MATLAB. The solution to this problem gives a set of process
variables that minimizes the time taken for abnormality detection while taking
into account all potential hazards. Additionally, this optimal set ensures that
the number of alarms activated for any fault is not more than a threshold
number.

The
formulation is as follows:

For this
particular study, the MILP routine ‘intlinprog’ has been used which is based on
the branch and bound algorithm.

The
proposed approach is then applied to a benchmark industrial plant control
problem, the Tennessee Eastman Process Control Problem [1], a well-defined
simulator of a chemical process plant which is extensively used in process
control research. It consists of a reactor, separator, condenser, stripper and
a recycle compressor. The exothermic reactions involved and its operation under
pressurized conditions make it a safety-critical plant, and hence a potential application
for the proposed algorithm.

A number of major hazards were identified for this
plant including control valve 'stiction’ and sensor faults. The abnormal
scenarios were simulated using the closed loop simulator [2], [3] in a MATLAB
environment. The simulations were carried out, one scenario at a time. An
important observation based on the results obtained is that the number of
alarms for any scenario should be not less than 4, in order to detect any
abnormal situation soon enough to provide the operator with adequate response
time. The optimal set of process variables identified to be configured in the
alarm system are ‘Reactor pressure’, ‘Separator liquid Level’, ‘Stripper Liquid
Level’ and ‘Purge gas composition- G component’.

References

[1]   Downs, James J., and
Ernest F. Vogel. "A plant-wide industrial process control
problem." Computers & chemical engineering 17, no. 3 (1993):
245-255.

[2]   Ricker, N. Lawrence.
"Decentralized control of the Tennessee Eastman challenge
process." Journal of Process Control 6, no. 4 (1996): 205-221.

[3]   Ricker, N. Lawrence. "Tennessee
Eastman challenge archive." [Simulink Model], Retrieved from http://depts.washington.edu/control/LARRY/TE/download.html
(accessed November 21, 2016).