Home

From the innovation track:Model Intelligence for Process Control-- Tai-Ji

May 8, 2007

Yucai Zhu, president of Tai Ji Control, www.taijicontrol.com Tai-Ji (Tai Chi) is the Chinese system of balanced philosophy. "I thought, wow! This is control! So I named my company Tai-Ji." A few words about MPC: MPC is the standardtool for advanced process control with tremendous benefits in the refining/petrochem industry, with maybe 10,000 applications according to some people. Other industreies are starting MPC trials. MPC success factors are good understanding of the proc...

waltboyes

functional design and benefit study (10%)
pretest (10%)
Plant step test and model identification (40%)
Controller simulation and tuning (15%)
Controller commissioning and operator training (25%)
Controller Maintenance (mainly re-identification)

These two italicized steps take most of the time. Review of MPC identification:

the tradtional approach is a manual, single variable test, very long.
Open loop test, more disturbing.
Nonparametric model (FIR) based, less accurate.
Model validation/selection is trial and error and process knowledge.

Here's a new approach:

Automatic multivarialbe test; much shorter
Closed loop test if possible, less disturbing
parametric (compact) model, more accurate
model validation using grading (A, B. C and D)

This "new" approach isn't new. It was first applied on an FCC unit at Statoil in 1993. Tai-Ji Cotnrol was founded in 1996; Tai-Ji ID released in 1997. Earlier users of Tai-Ji ID were Dow, Statoil, and a BP refinery. Mobil and Exxon both run pilot projects in 1999. ExonMobil decided to use Tai-Ji ID worldwide in 2000. In 2002, Tai-Ji Control and Matrikon started collaboration and Process Doctor-Tai-Ji (Tai-Ji Online) was developed. In 2004, AspenTech performed automatic multivariable test. In 2007, Honeywell announced a new tool for multivariable closed loop test and identification. New approach. Hah. ProcessDoctor-Tai-Ji (Tai-Ji Online)

combines the strength of Tai-Ji Control and Matrikon
testing and identification in one piece

This creates the models automatically and can be exported in DMC format, or other formats. After the first project, Hovensa decided to do the rest of them themselves. They didn't want to pay us for the testing. That's all in the past, for me. What can we do now to extend MPC?

Challenges in MPC and process control:

cost of MPC is too high, even for the refining/petrochem industry
due to high cost, it is difficult to apply MPC in other process industries
over 50% of PID loops perform poorly
experienced control engineers are leaving/retiring

Think about it: PID loops have been around for more than 50 years, and still half of them don't work properly. That's a real challenge. Model intelligence: Model intelligence is a class of computer programs, that can, for a given class of processes, automatically develop, maintain and use dynamic process models for control, prediction, and monitoring/diagnosis. Model intelligence consists of three modules:

connectivity module (OPC)
modeling module: online identification
application module: MPC tuning/control, PID tuning/control. monitor and diagnosis program
Key differences: automatic model development and maintenance, online and in realtime
online identification plays the key role
adaption is included, though not sample-wise.

If you can develop models automatically, it is an automatic system, yes? Tai-Ji PID-- A multi-loop PID auto-tuner (coming in the next release)

obtains models using closed loop identification
closed loop tests do not disturb unit operation
highly efficient, can test and tune 10 loops simultaneously
tuning cascade loops using one test
using multivariable models to tune interactive loops and to provide decoupling parameters

New Generation MPC-- Tai-Ji MPC "Get the design right, the rest is automatic." This is the official announcement of a new product. An adaptive MPC with an MPC Controller, MPC Monitor, and Online Identification built in, performs steps 2 to 5 (go back to the top of this post if you don't remember what they are) in parallel, and reduces MPC cost b a factor of 5! Doing things in parallel gives us great efficiency! Now all the non-refining applications are doable at a reasonable cost. We can do more! Nonlinear MPC

many process units change opeartion range frequently, like polymer units, lubricate oil units, power plants, air separation plants
good control at working points and fast/smooth transition control are desired
Linear MPC will nto perform well
use multi-models or LPV

Model is the key to MPC technology Considerable process has been mae in MPC identification: more efficient, saving 70% time; more accurate models for control; user friendly" all procedures are automatic. Model intelligence has a bright future: adaptive MPC can reduce costs by a factor of 5; Nonlinear MPC can be used for difficult processes; all PID loops will have diagnose and auto-tuning; all processes can benefit from MPC technology.