APC III - First-Principles Advantage

Recently, I wrote a posting about how one of one of the super-geniuses at PAS was busy inventing the next best thing in Advanced Process Control – something we code-named Galaxy.

Galaxy simplified the whole business of implementing APC schemes – which is a good thing because – until now – putting APC schemes in place required far too much expertise (and cost).

But Galaxy does more than save money – it also makes more money than a traditional APC scheme because it works better.  Here’s why…

For the most part, traditional APC applications are based on linear approximations of the process plant, which are then solved across a near-term horizon to generate a series of control moves that optimize a pre-defined objective function, subject to constraints.

This linear approximation approach works pretty well - as long as plant operations are kept in a small, pre-defined region.  As it turns out, a linear approximation works well for anything if you stay close enough to some starting point.

But the linear approach breaks down real fast once things start to change. 

Even the linear gurus recognize this and so they have a bunch of tricks to fix that –  step-wise linear modeling, gain scheduling, quadratic optimization and more.  All to get a more accurate model across a wider operating range so the controller can still generate (somewhat) accurate results. 

The better approach – the holy grail to the optimization of any plant – is to use first-principle chemical engineering models right in the control problem. 

Although lots of smart engineers have worked to do this, the mathematics and computing technology have not been there to support it.  After all, solving a complex multivariable non-linear problem in real-time needs a bit more than everyday freshman mathematics.

Galaxy changes all that.

Galaxy includes the mathematics and technology to allow a complex multivariable non-linear first-principles model to be solved fast enough on a pretty standard PC to be useful for real-time control. 

With Galaxy, the control engineer can use rigorous first-principles models and simpler linear approximations as desired.  The Galaxy solver and optimization engine then works out the whole problem in real time to give more accurate control predictions across a much broader range of plant operations.

So what’s the big deal?

The big deal is that Galaxy makes more money at the plant – because it does a better job of more closely controlling and optimizing, but also because it stays in control in those regions where linear modeling just doesn’t work.

This means that Galaxy can optimize and control a plant right through a grade change, feedstock switch, turndown and just about any other abnormal situation thrown at it.

And this isn’t just a nice story or wishful thinking.  Galaxy is already doing this in plants – going where no APC scheme has gone before…

Want to know more? Email me.

Comments

[Tony]

What about controlling processes where first principles are virtually non-existant or totally emphirical, derived with lots of guess work and hand waving?

[Roland Heersink]

Tony - yes, you are right.

For those processes where the underlying physical phenomena are not known well enough to develop a first principles based model one will have to use a different approach, an empirical approach. Our Galaxy technology is designed to handle this reality as well, as was its predecessor STAR. In this mode, all the user needs to determine are the steady state gains for the relevant MV/CV pairs.

These gains are easily determined from historical data or even discussions with the operators on the unit. As a last resort, a small bump test - one up and one down - can get the job done as well.

Remember, we are after approximate gains only, so we don't need to do the full model identification as you might have to do with other technologies.

Hope this helps!

Roland.