DETAILED DESCRIPTION OF THE INVENTION

[0020] Illustrated in FIG. 1 is a manufacturing plant for carrying out a non-linear process such as a chemical or polymer process. The processing plant is formed of a series of vessels (e.g., holding tanks, feed tanks, catalyst feeds), reactors (including mixing tanks, etc.) and pumps (or compressors) connected by various conduits and pipes. Sensors 17 senses temperature, volume, pressure, compositions and other physical properties at various points in the process and provide sensor data to a software or computer-implemented controller 19 . The controller 19 utilizes the sensor data to maintain setting for controlled variables such as temperature, pressure, compositions and product properties by adjusting manipulated variables such as feed rates, flowrate, temperature of the vessels, reactor 13 and pumps/compressors 15 . Controller 19 physically effects adjustment of the manipulated variables through actuators 21 coupled to respective valves 23 of the vessels 11 , reactors 13 and pumps/compressors 15 .

[0021] In particular, the controller 19 employs a linear dynamic model of the manufacturing process. These linear models relate the dynamic responses of controlled variables to manipulated variables in terms of process gains, response time, and dead time. The sensor data define values of control variables in the model equations. The model predicts how controlled variables will change with respect to step changes in manipulated variables.

[0022] An optimizer 25 uses the sensor data and an internal model of the manufacturing process to provide target values for the controlled variables. In the preferred embodiment, commonly known first principles equations for thermodynamics, kinetics, and heat and mass balances define the model. As a result, the model is capable of predicting the nonlinear relationship (or gains) between the controlled variables and the manipulated variables based on first principles.

[0023] In the case of a highly non-linear process being controlled, the linear model of the controller 19 is problematic. A common practice in the prior art is to update the process gains of the linear model in a pre-programmed manner based on process experiences. Applicants have discovered an improved non-linear optimizer 25 for such a controller 19 . Specifically, the optimizer 25 of the present invention employs a non-linear first principles model which is suitable for frequently calculating target controlled variable values and non-linear gains between the controlled variables and manipulated variables. Frequency of optimizer 25 calculations and operation is about every 2-3 minutes or about the same rate as the controller 19 determinations are made. The non-linear gains computed by the optimizer 25 are used continuously to update the process gains of the linear model of the controller 19 . In contrast, typical optimizers used with process control of the prior art perform calculations only for target control variables at about once every four hours or so and hence not at a rate helpful to the controller 19 of a highly non-linear process.

[0024] Referring now to FIG. 2 is an illustration of the preferred embodiment in a computer system 31 . Generally, the computer system 31 includes a digital processor 33 which hosts and executes the invention controller 19 and optimizer 25 in working memory. Input 35 to digital processor 33 is from sensors 17 , another software program, another computer, input devices (e.g., keyboard, mouse, etc.) and the like. Output 37 (i.e., the controlled variable settings determined by controller 19 and gain adapted by optimizer 25 ) is provided to actuators 21 , another computer, storage memory, another software program and/or output devices (e.g., display monitor, etc.) and the like. It is understood that computer system 31 may be linked by appropriate communication links to a local area network, external network (i.e., the Internet) or similar such networks for sharing or distributing input and output data.

[0025] In FIG. 2 , the controller 19 is preferably DMC Plus by Aspen Technology, Inc. of Cambridge, Mass. Optimizer 25 is preferably DMO/SQP®) also of Aspen Technology, Inc. Other similar non-linear optimizers are suitable. In the preferred embodiment, Polymers Plus (module 27 in FIG. 2 ) of Aspen Technology Inc. supplies the internal rigorous, non-linear model 29 to optimizer 25 . That is, optimizer 25 uses a steady state process model 29 based on first principles, rigorous, mechanistic Polymers Plus models.

[0026] By way of the below example, the preferred embodiment is illustrated around a fluidized bed, gas phase polyethylene process. The non-linear optimizer 25 solves the non-linear steady state model 29 from Polymer Plus module 27 . To accomplish this, the optimizer 25 employs a:

[0027] Sparsity file,

[0028] Nonlinear steady state model 29 , and

[0029] Objective function

[0030] The non-linear model 29 is, preferably, formulated in open-equation form. In this case, optimizer 25 supplies values to the model 29 for all the variables of interest, including manipulated variables and controlled variables. The model 29 returns to optimizer 25 the values of the constraint equation residuals, and as many of the Jacobian elements (partial derivatives) as it can calculate.

[0031] If an open-equation model is not available, a closed-form model may also be used. In that case, optimizer 25 supplies values to the model 29 for its ‘input’ variables, and the model 29 returns values of its ‘output’ variables.

[0032] The nonlinear model 29 user interface package allows users to:

[0033] Set a scenario—initial values of controlled variables and manipulated variables

[0034] Set constraints—targets or upper and lower limits on controlled variables and manipulated variables

[0035] Set the objective function—costs on controlled variables and manipulated variables

[0036] Run a simulation of the scenario

[0037] View the calculated trajectory of the controlled variable targets and manipulated variable targets over the nonlinear simulation interval.

[0038] As such, Optimizer 25 (via the nonlinear model 29 ) supplies the following to the DMCplus controller 19 :

[0039] Controlled variable targets or upper and lower limits

[0040] Manipulated variable targets

[0041] Model gains (i.e. derivatives of controlled variables with respect to manipulated variables, derivatives of controlled variables with respect to Feed Forward variables) for the current operating point

[0042] Other model variables (such as residence time)

[0043] Example Reactor Model

[0044] The nonlinear model 29 incorporates all of the equipment of interest required to represent the polyethylene process—the reactor, heat exchanger, compressor, mixers, splitters, component separator, valves, etc. Details of the reactor model 49 are given below for illustrative purposes.

[0045] As shown in FIG. 3 , the reactor 49 is modeled as a steady-state CSTR (continuous stir tank reactor), with a vapor phase 45 and a liquid phase 47 in equilibrium. Two feeds—a gas feed 41 and a catalyst feed 43 —are required. The two products are (1) a vapor stream 51 , and (2) a liquid stream 53 containing the polymer.

[0046] The reactor model 49 is composed of the following nonlinear equations derived from first principles of the Polymers Plus module 27 :

[0047] Component material balances

[0048] Total material balance

[0049] Pressure balances

[0050] Energy balance

[0051] Vapor enthalpy calculation

[0052] Liquid enthalpy calculation

[0053] Vapor-liquid equilibrium

[0054] Reactor volumes

[0055] Component reaction rates

[0056] Polymer attribute calculations

[0057] Catalyst attribute calculations

[0058] The equations require calculations on the stream enthalpy, stream density, reaction kinetics, vapor-liquid equilibrium K-values, and the polymerization reaction kinetics. Such calculations are performed by subroutines in Polymers Plus module 27 .

[0059] Component Slate

[0060] The component slate for the reactor model 49 consists of 1

[0061] Variables

[0062] Streams into and out of the reactor model 49 have a standard format and units of measurement to enable automatic connection to other equipment models. Conversion to internal model units of measurement takes place inside the model 49 . The stream variables are: 2

[0063] Analysis

[0064] Fixed variables are those having values specified to define the problem. The number of equations must equal the number of calculated variables.

Number of Variables

[0065] 3

Number of Variables

[0066] 4

[0067] In the preferred embodiment, the reactor model 49 is formulated to fit into pre-defined stream conventions:

[0068] Mole fractions must sum to 1.0

[0069] A stream contains mole fractions, flow, temperature, pressure, and enthalpy

[0070] In Polymers Plus (module 27 ) formulation, the following variables are included in the streams:

[0071] Catalyst feed stream 43 5

[0072] The manipulated variables of the reactor 49 include the catalyst feed 43 variables and gas feed 44 variables as solved by subroutine ZNMECH. Thus, optimizer 25 forwards, in pertinent part, the output of subroutine ZNMECH to DMCPlus controller 19 . In response, DMCPlus controller 19 uses the subroutine outputs to update the process gains in its dynamic linear model of reactor 49 .

[0073] The corresponding dynamic linear models of reactor 49 employed by the DMC controller 19 are of the form

βΔu ( k )= e ( k +1)

[0074] where β is the reactor's dynamic matrix having m columns of the reactor's step function appropriately shifted down in order. Δu(k) is an m-dimensional vector of control moves. e(k+1) is the projected error vector. See B. A. Ogunnaike and W. H. Ray, “Process dynamics, modeling, and control,” Chapter 27, pp. 1000-100.7, Oxford University Press 1994.

[0075] According to the foregoing example and description, the present invention utilizes first principles kinetics, thermodynamics and heat and mass balances to develop the non-linear relationships between the manipulated variables and controlled variables. Based on these non-linear relationships (non-linear model 29 ), optimizer 25 effectively and timely gain adapts the dynamic linear model for use in controller 19 for the subject manufacturing process. As such, the preferred embodiment provides a novel combination of (i) Polymers Plus rigorous, non-linear models 19 , (ii) a non-linear optimizer 25 and (iii) DMCPlus (or similar dynamic linear model based) controller 19 as heretofore unapplied to process control.

[0076] While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention encompassed by the appended claims.