petrocontrol/GDS

GDS		(GENERALIZED DISTILLAION SHORTCUT) INFERENTIAL CONTROL OF NARROW CUT DISTILLATION COLUMNS

	To overcome the challenge of distillation column control, Dr. Y. Zak Friedman of Petrocontrol pioneered the development of GDS (Generalized Distillation Shortcut), a first principles model inferring product purities from column operating conditions, indicating and controlling the product qualities minute-by-minute. No longer dependent on frequent lab support, the advanced control applications (or operators) are free to move the unit up against constraints while responding to disturbances and keeping product purities at targets. GDS is a versatile tool and has been implemented on a wide range of column configurations, from stabilizers to gas plants to superfractionators. About fifty applications have been implemented to date.



The distillation inferential modeling technology has proven to be:
	Reliable enough to replace distillation analyzers; Inferential precision is for example 0.2% on stabilizer C5 in LPG, where the target of C5 in LPG is 1%, and with periodic biasing the error is reduced to 0.1%
	Can work with analyzers, in the way of resetting a bias in the calculation via a dead time compensator.
	Calculates liquid and vapor traffic, and infers flooding if necessary. This is a useful inference because the usual flooding indicators, such as pressure difference and mass balance disturbances are "post mortem" measurements.
	Responsive to feed composition changes. The inference model quickly recognizes feed composition disturbances, and changes the column operation to keep product purities constant. There is no need to input any data whatsoever by the operator. Detection of feed changes is completely automatic.
	Integrates smoothly with multi variable control packages available on the market.
	Easily understood. The model follows standard thermodynamic procedures for equilibrium constants and other column calculations. All equations and assumptions are well understood by chemical engineers.
	Packaged implementation: Model, constraints and dynamics operate as one package, integrating inferential cut points with other control variables, constraints and where exist, analyzer measurements.
	The model applies a simulation shortcut technique to obtain the best fit between bottom (or top) product composition and column measurements.
	Requires only a simple calibration procedure, which makes use of steady state data. GDS is robust enough to work without calibration, but given inaccuracies of flow, pressure and temperature measurements, precision improves after steady state calibration.

Reference literature.
	First-principles inference model improves deisobutanizer column control, Hydrocarbon Processing Journal, March 2003. 2003_DIB.pdf
	First principles distillation inference model for a toluene – xylene separation column. ERTC Computer Conference, June 2002. 2002_Toluene_Inference_ERTC.pdf
	First-principles distillation inference models for a superfractionator product quality prediction. Hydrocarbon Processing Journal, February 2002. 2002_superfractionator.pdf
	Experience with GDS, a first principles inference model for distillation columns. Presented ERTC Computer Conference, June 2001 and NPRA computer conference, October 2001, published in Petroleum Technology Quarterly, Autumn 2001. 2001_GDS.pdf
	Simulation based inferential controls. Paper presented at the AICHE spring conference 1995, Houston TX. 1995_Inferentials_AIChE.pdf


back to inferential control

GDS

The distillation inferential modeling technology has proven to be:

Reliable enough to replace distillation analyzers; Inferential precision is for example 0.2% on stabilizer C5 in LPG, where the target of C5 in LPG is 1%, and with periodic biasing the error is reduced to 0.1%

Can work with analyzers, in the way of resetting a bias in the calculation via a dead time compensator.

Calculates liquid and vapor traffic, and infers flooding if necessary. This is a useful inference because the usual flooding indicators, such as pressure difference and mass balance disturbances are "post mortem" measurements.

Responsive to feed composition changes. The inference model quickly recognizes feed composition disturbances, and changes the column operation to keep product purities constant. There is no need to input any data whatsoever by the operator. Detection of feed changes is completely automatic.

Integrates smoothly with multi variable control packages available on the market.

Easily understood. The model follows standard thermodynamic procedures for equilibrium constants and other column calculations. All equations and assumptions are well understood by chemical engineers.

Packaged implementation: Model, constraints and dynamics operate as one package, integrating inferential cut points with other control variables, constraints and where exist, analyzer measurements.

The model applies a simulation shortcut technique to obtain the best fit between bottom (or top) product composition and column measurements.

Requires only a simple calibration procedure, which makes use of steady state data. GDS is robust enough to work without calibration, but given inaccuracies of flow, pressure and temperature measurements, precision improves after steady state calibration.

Reference literature.

First-principles inference model improves deisobutanizer column control, Hydrocarbon Processing Journal, March 2003. 2003_DIB.pdf

First principles distillation inference model for a toluene – xylene separation column. ERTC Computer Conference, June 2002. 2002_Toluene_Inference_ERTC.pdf

First-principles distillation inference models for a superfractionator product quality prediction. Hydrocarbon Processing Journal, February 2002. 2002_superfractionator.pdf

Experience with GDS, a first principles inference model for distillation columns. Presented ERTC Computer Conference, June 2001 and NPRA computer conference, October 2001, published in Petroleum Technology Quarterly, Autumn 2001. 2001_GDS.pdf

Simulation based inferential controls. Paper presented at the AICHE spring conference 1995, Houston TX. 1995_Inferentials_AIChE.pdf

back to inferential control