WO2013087973A1

WO2013087973A1 - Method of tuning a process controller

Info

Publication number: WO2013087973A1
Application number: PCT/FI2011/051118
Authority: WO
Inventors: Timo Harju; Risto Kuusisto
Original assignee: Metso Automation Oy
Priority date: 2011-12-16
Filing date: 2011-12-16
Publication date: 2013-06-20

Abstract

An model predictive control (MPC) type process controller(31) already implemented in the process control layer of the automation system is utilized in automatic determining initial values of MPC tuning parameters. The MPC controller (31) can be switched between an online process control mode (A) and an offline MPC tuning mode(B). In the online process control mode (A) the MPC controller (31) performs an online process control task in a real-time operating system environment on a process control layer of a process automation system. In the offline MPC tuning mode (B), the MPC controller is operated in the real-time operating system environment on the process control layer of the process automation system under control of an optimizer unit (33) to determine MPC tuning parameters meeting a predetermined quality criterion.

Description

METHOD OF TUNING A PROCESS CONTROLLER

FIELD OF THE INVENTION

The present invention relates generally to control of an industrial process, and particularly to tuning of a process controller. BACKGROUND OF THE INVENTION

A process control or automation system is used to automatically control an industrial process such as chemical processes, oil refineries, power plants, timber industry, mineral processing (e.g. rock-crushing, screening, grinding, etc.), and paper and pulp factories. The process automation sys-tem often uses a network to interconnect sensors, controllers, operator terminals and actuators. Process automation involves using computer technology and software engineering to help power plants and factories operate more efficiently and safely. An automatic controller must be able to facilitate the plant operation over a wide range of operating conditions. Also many machines, such as crushers, screeners, wood harvesters, etc., are very complicated systems with a plurality of sensors, controllers, and actuators.

Proportional-integral (PI) or proportional-integral-derivative (PID) controllers are commonly used in many industrial control systems. These controllers are tuned with different tuning techniques to deliver satisfactory plant performance. However, specific control problems associated with the plant operations severely limit the performance of PI or PID controllers. The increasing complexity of plant operations together with tougher environmental regulations, rigorous safety codes and rapidly changing economic situations demand the need for more sophisticated process controllers. Most processes require the controlling of more than one variable. Controller-loop interaction exists such that the action of one controller affects other loops in a multi-loop system.

Model predictive control (MPC) refers to a wide class of control algorithms that use an explicit process model to predict the behavior of aprocess. The most significant feature that distinguishes MPC from other controllers is its long range prediction concept. The idea of MPC is to calculate a control function for the future time in order to force the controlled system response to reach the reference value, i.e. set point (SP), or multiple set points SPs. Therefore, the future reference values are to be known and the system behavior must be predictable by an appropriate model. An example block diagram illustrating a model predictive controller scheme is shown in Figure 1 . MPC 2 uses a dynamic model of the process 3 in order to predict the controlled variable or multiple variables CV. The predicted controlled variables are used in an on-line optimization procedure, which minimizes an appropriate cost function to determine the respective multiple manipulated variables MV. The controller output is implemented in real time such that only the first control change is used from the predicted sequence of the control changes, and then the procedure is repeated every sampling time with actual process data. The difference between the plant measurement (feedback CVs) and the model output (predicted CVs) may be used by the controller 2 to eliminate a steady state offset. Thus, using an internal optimizing algorithm, the MPC controller 2 determines a manipulated variable (MV) profile that optimizes some open-loop performance objective over a finite horizon extending from the current time into the future, i.e. attempts to control the CVs towards the set points (SP). The internal optimizing algorithm may identify control actions which minimize a "cost" function which considers CV errors, MV changes, and how fast the process approaches the optimum steady state operating point. For example, a cost function may depend on the quadratic error between the future reference variable SPs and the future controlled variable CVs within a limited time horizon. This manipulated variable profile is implemented until new plant measurements become available. Feedback is incorporated by using the measurement to update the optimization problem for the next time step. MPC models predict the change in the controlled variables (CV) of the process 3 that will be caused by changes in the manipulated (MV) variables. In an industrial process, CVs that can be adjusted by the controller often include either the set points of regulatory PID controllers (pressure, flow, temperature, etc.) which control the final control elements (valves, dampers, etc.). Variables that cannot be adjusted by the controller may be considered as disturbance varia- bles (DV).

The tuning of the MPC involves searching, usually by trial and error, finding the optimal parameters in the internal cost function of the MPC, i.e. parameters that minimize a characteristic representing a control error. Basically, the MPC may have four main tuning parameters, namely; the control horizon, the prediction horizon, the controlled variable (CV) weights and the manipulated variable (MV) weights in the cost function. The controlled variable (CV) weights are used to define mutual significances of the variables. The larger the value of the weigh is, the more accurate control is targeted, but with a risk of reducing the stability of the control. The manipulated variable (MV) weights are used to define the trade-off between the amount of movement allowed in the manipulated variables MV and the rate at which the output deviation from set point is reduced over the prediction horizon. In other words, larger MV weights will slower the response of the process. Naturally, the higher is the number of MVs and CVs associated with the MPC (a multivariable MPC), the higher is also the number of the tuning parameters. The prediction horizon represents the time span (e.g. the number of samples) into the future over which MPC computes the predicted process variable profile and attempts to minimize the prediction error between the predicted measurements and the set points. The control horizon represents the time span (e.g. the number of MV moves) into the future over which the MPC computes at each sampling time to eliminate the current prediction error between the predicted measurements and the set points.

Tuning of the MPC control is, in fact, based on mutual relationships of the weights rather than on their absolute values. Moreover, the control and measurement ranges are not necessarily scaled separately, and the absolute values of the weights may freely vary case-by-case. Even with the same MPC controller and in the same process, the same control performance can be achieved with a myriad of parameter combinations. For example, all weights can be multiply with a constant, such as 10 or 100, and the control performance still remains the same. This is problematic to a person attempting to tune a MPC controller, since he is not able to conclude anything about rationality of the tuning based on the absolute values of the weights and parameters. This differs significantly from tuning of a PID controller, for example, in which both the control and measurement ranges are scaled. Assuming that field devices are selected and configured appropriately, gains of different PID control- lers are very similar to each other and therefore easier to evaluate.

There are offline simulation and tuning tools which allow searching default values (thumb values) for MPC tuning parameters. The simulation and tuning tools have been implemented in software run on a PC. A MPC algorithm to be used and the characteristics and special parameters thereof are pro- grammed in the tuning software tool with a suitable programming language, such as Matlab. A problem in the prior art tuning tools is that if the MPC con- trailer is developed further (i.e. the MPC algorithm is modified), the same features, additional parameters, etc., must separately be programmed into the tuning tool software with the Matlab programming language or the like. In other words, changes in the MPC controller employed may require double work when the changes are programmed both into the MPC controller and into the tuning tool. There are also problems regarding the version management, because versions of the MPC controllers in the tuning tool and the automation system must be compatible.

US2007/0225835 discloses an adaptive MPC controller provided with auto-tuning of controller variables.

SUMMARY OF THE INVENTION

An object of the present invention is to provide an improved tuning scheme for a process controller. This object of the invention is achieved by the subject matter of the attached independent claims. The preferred embodiments of the invention are disclosed in the dependent claims.

An aspect of the invention is a method of tuning a model -based process controller, particularly a model predictive control (MPC) type process controller, comprising

operating a model-based controller in an online process control mode in a real-time operating system environment on a process control layer of a process or a machine,

switching said controller from said online process control mode to an offline tuning mode for a controller tuning,

operating said controller in said offline tuning mode in the real-time operating system environment on the process control layer of the process or machine under control of an optimizer unit to determine controller tuning parameters meeting a predetermined quality criterion, and

switching said controller from said offline tuning mode to said online process control mode upon determining said controller tuning parameters.

In an embodiment of the invention, said switching to the offline tuning mode comprises disconnecting said controller from actual process signals and connecting predicted signals to the optimizer.

In an embodiment of the invention, said switching to the offline tuning mode comprises

disconnecting a feedback of controlled variables from a process or a machine to the controller,

disconnecting a supply of manipulated variables from the controller to the process or machine,

connecting predicted controlled variables and predicted manipulated variables from the controller to the optimizer, and

providing tuning parameters or tuning controls from the optimizer to the MPC controller or to a user interface.

Another aspect of the invention is a method of tuning a model - based process controller, particularly a model predictive control (MPC) type process controller, comprising

running online a process control task in a model-based controller in a real-time operating system environment on a process control layer of a process or a machine,

providing a duplicate of said controller with parameters thereof to enable a controller tuning, and

operating one of said controllers in the real-time operating system environment on the process control layer of the process automation system under control of an optimizer unit to determine controller tuning parameters meeting a predetermined quality criterion.

In an embodiment of the invention, the method comprises operating said optimizer in an online process control mode in a realtime operating system environment on a process control layer of a process or a machine,

switching said optimizer from said online process optimization mode to an offline tuning mode for the tuning,

operating said optimizer in said offline tuning mode in the real-time operating system environment on the process control layer of the process or machine to control said controller of the duplicate of said controller to determine the controller tuning parameters meeting the predetermined quality crite- rion, and

switching said optimizer from said offline tuning mode to said online process optimization mode upon determining said controller tuning parameters.

In an embodiment of the invention, said switching of said optimizer to the offline tuning mode comprises disconnecting said optimizer from actual process signals, connecting to the optimizer the predicted signals from said controller or from said duplicate of said controller, and changing to use said predetermined quality criterion in place of a process cost function in optimization.

In an embodiment of the invention, said switching of said optimizer to the offline tuning mode comprises

disconnecting a feedback of controlled variables from a process to the optimizer,

disconnecting a supply of manipulated variables from the controller, connecting predicted controlled variables and predicted manipulated variables from the controller or from the duplicate of the controller to the optimizer, and

providing tuning parameters or tuning controls from the optimizer to the controller or to the duplicate of the controller.

In an embodiment of the invention, said optimizer is dedicated for the controller tuning and operated in the real-time operating system environment on the process control layer of the process automation system.

In an embodiment of the invention, the method comprises providing said optimizer by duplicating an optimizer operating in an online process control in a real-time operating system environment on a pro- cess control layer of a process automation system, and

using said predetermined quality criterion in place of a process cost function in optimization.

In an embodiment of the invention, said optimizer is at least partly integrated into the controller or into the duplicate of the controller.

In an embodiment of the invention, the method comprises triggering said tuning from a user interface on plant management layer of the process automation system.

In an embodiment of the invention, the method comprises accepting said determined controller tuning parameters for use from a user interface on plant management layer of the process automation system.

In an embodiment of the invention, the method comprises, upon determining said controller tuning parameters, continuing tuning of the controller manually from a user interface on plant management layer of the process automation system, while said controller is operating in an online process control task in the real-time operating system environment on the process control layer of the process automation system. Another aspect of the invention is a system for tuning a model - based process controller, particularly a model predictive control (MPC) type process controller, comprising means for implementing for implementing any of embodiments above.

Another aspect of the invention is an executable program product comprising program code means stored on a processor readable medium for performing a method according to any one of embodiments above when said program product is run on one or more processors.

BRIEF DESCRIPTION OF THE DRAWINGS

In the following the invention will be described in greater detail by means of exemplary embodiments with reference to the attached drawings, in which

Figure 1 is a block diagram illustrating an exemplary model predictive controller (MPC),

Figure 2 is a simplified block diagram illustrating an example of a distributed process automation system;

Figure 3 is a simplified block diagram illustrating a tuning arrangement according to an exemplary embodiment of the invention;

Figure 4 is a flow diagram illustrating an example operation of a tun- ing arrangement according to an exemplary embodiment of the invention;

Figure 5 is a block diagram illustrating a tuning arrangement according to another exemplary embodiment of the invention;

Figure 6 is a block diagram illustrating a tuning arrangement according to a still another exemplary embodiment of the invention;

Figure 7 illustrates an exemplary view at the user interface for controlling an MPC controller; and

Figure 8 show examples of step response models in a paper machine application.

EXAMPLE EMBODIMENTS OF THE INVENTION

The present invention can be applied in connection with any computer-aided control or operating system, such as automation system (process control system), and any technical process, such as industrial process or the like. The technical processes may include, but are not limited to, processes in a processing industry, such as pulp and paper, oil refining, petrochemical and chemical industries, or processes in power plants, processes in timber indus- try, mineral processing (e.g. rock-crushing, screening, grinding, etc.), operation of processing machines, etc. In the following, the invention is illustrated with exemplary embodiments without restricting the invention to any of the control systems and processes described in these exemplary embodiments.

There are various architectures for an automation system. For example, a process automation system may be a Direct Digital Control (DDC) system or a Distributed Control System (DCS), both well known in the art. One example of such a decentralized automation system is Metso DNA (DNA, Dynamic Network of Applications) delivered by Metso Inc.

Figure 2 schematically illustrates an exemplary distributed process automation system, to which the example embodiments of the invention may be applied. The user interfaces and workstations of the automation system for managing, operating and maintenance of an entire plant, such as a paper mill, are typically referred to as a control room, which may be comprise one or more control room computers/programs 1 1 , maintenance/monitoring/engineering computers/programs 19, databases 23, etc., for example, with multipurpose operating systems, such as Windows XP, Windows 7, Linux, etcThe process automation system may comprise a control room bus/network 12 which may interconnect the user interface components and control room computers on a plant management layer of the process automation system. The control room bus/network 12 may be a local area network, for example, based on the standard Ethernet technology. The plant management layer, as used herein, refers generally to the higher-layer plant operating and management equipment available for the operator of the plant, such as control room computers, maintenance/monitoring tools, simulator tools, etc.

The process automation system may further comprise a process bus/network 13 which may, in turn, interconnect the process control components, such as process controller units 17 and 18 as well as an optimizer unit 22 on a process control layer with each other as well as with the equipment with the plant operating and management layer. The process bus/network 13 may be based on a deterministic token passing protocol, for instance. The process control components 17, 18, and 22 may also be connected to the control room network 12, allowing the communication between the process controllers and the user interfaces. It must be appreciated, however, that Figure 2 only illustrates one example of an automation system and it is not the intention to limit the application area of the invention to any specific implementation of an automation system.

There are various alternative ways to arrange the interconnection between the process controller units 17, 18 on the process control layer and field devices 16-1 ,..., 16-N, such as actuators, valves, pumps and measuring devices (e.g. sensors). Traditionally, field devices have been connected to the control system by two-wire twisted pair loops, each device being connected to the control system by a single twisted pair providing a 4 to 20 mA analog input signal. In the example architecture of Figure 2, the process controller units 17 and 18 may be connected with one or more interface units or I/O (input/output) units, such as a fieldbus interface unit 14 and a HART interface unit 20. The HART interface unit 20 supports a Highway Addressable Remote Transducer (HART) protocol, that allow the transmission of digital data together with the conventional 4 to 20 mA analog signal in a twisted pair loop 21 to and from field devices 16. The HART protocol is described in greater detail for example in the publication HART Field Communication Protocol: An Introduction for Users and Manufacturers, HART Communication Foundation, 1995. The HART protocol has also been developed into an industrial standard. The fieldbus interface unit 20 may interface any field buses 15, such as industry standards Foundation Fieldbus, PROFIBUS, AS-i bus, CAN bus and Modbus. However, it is to be understood that the type or implementation of the interconnection between the process controllers and the field devices may be based on any one of the alternatives described above, or on any combination of the same, or on any other implementation.

Each of the process controller units 17 and 18 may comprise one or more process controllers, such as PI, PID, and MPC controllers. Examples of commercial controller units suitable for the process controller units 17, 18 may include ACN RT, ACN CS, ACN SR1 and Metso DNA VME controllers from Metso Inc, which provide multifunction controllers designed for embedded or centralized applications. The process controller units 17 and 18 may be microprocessor-based units and they may employ a real-time operating system (such as MetsoRTS) rather than general-purpose operating systems (such as Windows) employed in control room workstations. Real-time operating systems are typically designed to run one application very reliably and with precise tim- ing. The process control layer of an automation system may also include one or more optimizer units 22 that, based on a cost function of the overall process or a subprocess (preferably larger than the part of process controlled by an individual MPC), attempts to determine the most optimal operating point within the process or the subprocess.

An exemplary embodiment of the invention will be illustrated with reference to Figure 3. A model predictive controller (MPC) 31 may be any controller that uses an explicit process model to predict the behavior of a plant for the future time in order to force a controlled system response to reach a reference value, i.e. set point (SP). The MPC 31 may be implemented, for example, with ACN RT, ACN CS, ACN SR1 and Metso DNA VME controllers. In the example shown in Figure 3, the MPC 31 is a multivariable controller that uses a dynamic model of the process 32.

According to an aspect of the present invention, the MPC 31 can be used in two different modes: an online process control mode A; and an offline tuning mode B. In the online process control mode A (switches S1 , S2, S3, and S1 in position A), the MPC controller 31 performs an online process control task in a real-time operating system environment on a process control layer (such as in the process controller unit 17 or 18) of a process automation sys- tern. In other words, the MPC controller 31 operates as a conventional MPC controller. The MPC controller uses an internal optimization algorithm that determines manipulated variable MV profile that optimizes some open-loop performance objective over a finite horizon extending from the current time into the future, i.e. attempts to control the feedback CVs towards the inputted set point/set points (SP). This manipulated variable MV profile may be implemented until a plant measurement becomes available. Feedback may be incorporated by using the measurered CVs to update the optimization problem for the next time step. The internal optimization algorithm may identify control actions which may minimize a "cost" function which considers CV errors, MV changes, and how fast the process 32 approaches the optimum steady state operating point SP. For example, a cost function may depend on the quadratic error between the future reference variable SP and the future controlled variable CVpred within a limited time horizon. MPC models predict the change in the controlled variables (CV) of the process 32 that will be caused by changes in the manipulated (MV) variables. In the example embodiment shown in Figure 3, in the online process control mode A (switches S1 , S2, S3, and S4 in position A) of the MPC 31 , the controlled variables CV feedback from the process 32 as well the manipulated variables MV fed from the MPC to the process 32 may also be fed to an opti- mizer 33. In embodiments of the invention, having assumed the online process control mode A, the optimizer 33 may perform an online process optimization in a real-time operating system environment on a process control layer (such as in the optimizer unit 22) of a process automation system. The optimizer unit 22 may be a microprocessor-based unit with a real-time operating system. The optimizer 33 may continuously determine and implement optimal set points or optimal levels for MVs that reflect a selected control strategy based on a cost function of the overall process or the sub process (preferably larger than the part of process controlled by an individual MPC). The position A of the switch S4 represents that the cost function of the process is utilized in the mode A. The cost function of the process may be stored in the optimizer 33 and/or loaded from plant management layer. The operator may change parameters of the cost function via a user interface at the control room computer 1 1 and/or the maintenance computer 19. The optimizer 33 may maintain a required product quality and target a technical and economical objective. This may re- suit in an increase in capacity, product quality and value and/or a decrease in production costs, for example. Single optimizer 33 may be assigned for optimizing a plurality of process controllers. The optimizer 33 (implemented in the optimizer unit 22, for example) and the MPC 31 (implemented in the controller unit 17 or 18, for example) may communicate over the process bus 13, for ex- ample. The actual MPC controller 22 may eliminate system delays and cross- correlations. Moreover, below the MPC controller may be provided a PID control layer that may provide control of field devices, a safety logic, etc. Figure 8 show examples of step response models which illustrate an impact of various manipulated variables (MV) on various controlled variables (CV) over a time in a paper machine application.

As noted above, the prior art MPC tuning tools have been implemented in simulation software run on a PC in general-purpose operating system environment, such in the maintenance/engineering computer 19 shown in Figure 2. A MPC algorithm to be used and the characteristics and special pa- rameters thereof are programmed in the tuning software tool with a suitable programming language, such as Matlab, whereas the actual MPC controller may be implemented in the micro-processor-based process controller unit 17 or 18 in the real-time operating system environment. As a result, when the MPC controller is developed, double programming work may be needed, i.e. both in the tuning tool and in the actual MPC controller, and there are also problems regarding the version management, because versions of the MPC controllers in the tuning tool and the automation system must be compatible.

According to an aspect of the present invention, an MPC controller already implemented in the process control layer of the automation system can be utilized in automatic determining initial values (called thump values herein) of MPC tuning parameters. With embodiments of the invention, thumb values for MPC parameters can be determined without need for programming the MPC algorithm and the characteristics and special parameters thereof in a separate tuning software tool with a suitable programming language, such as Matlab, Moreover, there is no version management or updating requirement for a tuning tool, as the tuning tool may utilize a real MPC controller already available in a microprocessor and real-time operating system environment.

According to another aspect of the invention, in addition to the MPC controller, also an optimizer already implemented in the process control layer of the automation system for another process optimizing task can be utilized for the automatic determining the thump values of MPC tuning parameters. As a result, the MPC tuning parameters can be obtained without separate special- purpose tuning tools. The MPC controller and the optimizer utilized as tuning tools are available in the process automation system at any time.

Referring again to Figure 3, in the offline tuning mode B, the MPC controller 31 is still operating in a real-time operating system environment on a process control layer (such as the process controller unit 17 or 18) of a process automation system, but the MPC controller is now operationally disconnected from the online process control task (i.e. from the process 32) and operationally connected to the optimizer 33. In the example embodiment shown in Figure 3, the disconnection means are generally represented by the switches S1 , S2, S3, and S4 having switch positions A and B. The switches S1 , S2, S3, and S4 may be implemented by hardware, such as with analog and/or digital switches or selectors, or by software, or by a combination of hardware and software. When the MPC controller is in the offline tuning mode B, switches S1 -S2 are in the position B, the controlled variables CV feedback from the process 32 are disconnected from the MPC 31 and the optimizer 33, the manipu- lated variables MV fed from the MPC 31 are disconnected from the process 32 and the optimizer 33, and the set point/target MV inputs SP/MP are not provided from the optimizer 33 to the MPC 31 . Instead, the predicted controlled variables CVpred and the predicted manipulated variables MVpred are now fed from the MPC 31 to the optimizer 33, and tuning parameters/controls are fed from the optimizer 34 to the MPC 31 . Moreover, the cost function of the process is disconnected from use, and the optimizer 33 now assumes a tuning optimization mode wherein tuning parameters which optimize a quality value or values representing a control error are searched. The disconnection of the MPC 31 from the control task may require a process condition wherein the process 32 can be safely run solely with lowerlayer controls, such as PID controls. Such process conditions may be found, for example, in connection with setups, shutdowns, product changes, etc.

Operation of the exemplary embodiment of Figure 3 is now illustrat- ed with reference to an example flow diagram shown in Figure 4. It is assumed that the MPC 31 and the optimizer 33 are first operating in the online process control mode A. Then, at step 42, the MPC 31 and the optimizer 33 are disconnected from the online control and assume the offline tuning mode B as described above. The transition to the offline tuning mode B may be triggered by the operator via a user interface (e.g. at the control room computer 1 1 ), or it may be a scheduled operation, or automatically triggered by suitable criteria. Preferably, the structure and parameters already available in the MPC 31 are utilized in tuning, but new structures or parameters may also be loaded in to the MPC 31 for the tuning operation (step 43). At step 43, a first set of values for tuning parameters are inputted to the MPC 31 from the optimizer 34, such as the control horizon, the prediction horizon, the controlled variable (CV) weights and the manipulated variable (MV) weights in the cost function. In the beginning of the tuning the first set of values may comprise the values presently used in the MPC 31 . As the MPC 31 has the process models used in the online process control, the MPC 31 is able to simulate its operation, i.e. produce the predicted controlled variables CVpred and the predicted manipulated variables MVpred that are to the optimizer 33 (step 44). The optimizer 33 may calculate a quality value of the tuning based on the first set of the predicted controlled variables CVpred and the predicted manipulated variables MVpred (step 45). The quality value may be determined with any appropriate performance criterium of control, such as SSE (Sum of Squared Error), STSE (Sum of Time weighted Squared Error), etc. (step 45). The determined quality value may be compared with a reference value (step 46). If the quality value is not acceptable, the optimizer 33 determines a new set of values for the tuning parameters such that the quality value will be improved, e.g. the error is reduced or minimized (step 47). The new set of tuning parameter value are again fed to the MPC 31 , the response is simulated by the MPC (step 44), a new quality value of the tuning is determined based on the resulting new set of the predicted controlled variables CVpred and the predicted manipulated variables MVpred (step 45), and the quality value is checked against the reference value (step 46). This procedure (steps 44-47) may repeated to search for optimal tuning parameter values until the resulting quality value is acceptable or a predetermined number of iterations has been performed. The new tuning parameters found may then be stored in the MPC 31 and taken into use, either automatically or if approved by the operator via the user interface (step 48). Finally, the MPC 31 and the optimizer 33 may be reconnected to control process, i.e. they may return to the online process control mode A (S1 -S4 in position A) at step 49.

According to a still another aspect of the invention, an MPC controller having an online process control task and already implemented in the pro- cess control layer of the automation system may be copied or duplicated so as to provide a duplicate MPC controller for tuning purposes. As a result the additional advantage is obtained that the actual MPC controller can always be maintained in the online process control task, while merely an optimizer already implemented in the process control layer of the automation system for another process optimizing task may be disconnected from the primary optimization task and be utilized for the automatic determining the thump values of MPC tuning parameters. A disadvantage may be an increase in the memory capacity required and extra module or modules (i.e. MPC) in the control application.

An exemplary embodiment according to this still further aspect of the invention will be will be illustrated with reference to Figure 5. The model predictive controller (MPC) 31 be implemented in a similar manner as the MPC 31 shown in Figure 3, expect that the MPC 31 may now operate in the online process mode all the time. Therefore, switches S1 and S2 in the CV input and in the MV output can be omitted. Also outputs CVpred and MVpred to the optimizer 33 can be omitted. The optimizer 33 can be used in two different modes: an online process optimization mode A; and an offline tuning mode B. In the example embodiment shown in Figure 5, in the online process optimization mode A (switches S1 , S2, S3, and S4 in position A) of the optimizer 33, the optimizer 33 may perform an online process optimization in a real-time operating system environment on a process control layer (such as in the optimizer unit 22) of a process automation system, based on the controlled variables CV feedback from the process 32 as well the manipulated variables MV fed from the MPC to the process 32. Thus, operation in mode A may be similar to that described with reference to Figure 3.

The actual MPC 31 may be copied or duplicated into another MPC module 34 also run in in a real-time operating system environment on a process control layer (such as in the process controller unit 17 or 18) of a process automation system. The actual MPC 31 is copied or duplicated such that the MPC duplicate 34 exactly corresponds to the actual MPC 31 regarding signals, parameters, etc. The duplication may be hot standby duplication. More preferably, the copying or duplication may be performed on demand, in other words when tuning is needed. Referring to Figure 5, in the offline tuning mode B, the MPC controller 31 is maintained in the online process control task operating in a real-time operating system environment on a process control layer (such as the process controller unit 17 or 18) of a process automation system, but the optimizer 33 is now operationally disconnected from the online process optimization task (i.e. from the process 32) and operationally connected to the MPC duplicate 34. In the example embodiment shown in Figure 5, the disconnection means are generally represented by the switches S1 , S2, S3, and S4 having switch positions A and B. The switches S1 , S2, S3, and S4 may be implemented by hardware, such as with analog and/or digital switches or selectors, or by software, or by a combination of hardware and software. When the optimizer 33 is in the offline tuning mode B, switches S1 -S2 are in the position B, the controlled variables CV feedback from the process 32 and the manipulated variables MV fed from the MPC 31 are disregarded by the optimizer 33, and the setpoint/target MV inputs SP/MP are not provided from the optimizer 33 to the MPC 31 . Instead, the predicted controlled variables CVpred and the predicted manipulated variables MVpred are now fed from the MPC duplicate 34 to the optimizer 33, and tuning parameters/controls are fed from the optimizer 33 to the MPC duplicate 34. Moreover, the cost function of the process is dis- connected from use, and the optimizer 33 now assumes a tuning optimization mode wherein tuning parameters which minimize a quality value representing a control error are searched. The tuning process may then proceed as described above with reference to Fig. 4, except that now the MPC duplicate 34 performs the simulation instead of the actual MPC 31 . The new tuning parameters finally found may then be stored in the MPC 31 and taken into use, either automatically or if approved by the operator via the user interface. Finally, the optimizer 33 may be reconnected to control process, i.e. it may return to the online process optimization mode A (S1 -S4 in position A). Also MPC duplicate 34 may be reset for a new tuning process of another MPC 31 .

According to still another aspect of the invention, in addition to the MPC controller, also an optimizer may be duplicated in the process control layer of the automation system for the process control task and the MPC tuning. As a result, the MPC tuning has no effect to the actual process control task. It is now also possible to integrate the optimizer into the MPC. Moreover, an accelerated processing can be enabled during the tuning process and thereby the time required for tuning can be reduced.

An exemplary embodiment according to this still further aspect of the invention will be will be illustrated with reference to Figure 6. The model predictive controller (MPC) 31 and the optimizer 33 can be implemented in a similar manner as in Figure 3, expect that the MPC 31 and the optimizer 33 may now operate in the online process mode all the time. Therefore, switches S1 -S4 can be omitted.

The actual MPC 31 and the optimizer 33 may be copied or duplicat- ed into another MPC module 34 and another optimizer module 35 may also run in in a real-time operating system environment on a process control layer (such as in the process controller unit 17 or 18) of a process automation system. The actual MPC 31 is copied or duplicated such that the MPC duplicate

34 exactly corresponds to the actual MPC 31 regarding signals, parameters, etc. The duplication may be hot standby duplication. More preferably, the copying or duplication may be performed on demand, in other words when tuning is needed. The latter approach allows using the same MPC duplicate module for a plurality of different online MPCs, for one at time. The duplicated optimizer

35 may alternatively be different from the actual optimizer 33 and configured for the MPC tuning optimizing task only. Referring to Figure 6, when the tuning process is needed, the predicted controlled variables CVpred and the predicted manipulated variables MVpred are now fed from the MPC duplicate 34 to the optimizer 35, and tuning parameters/controls are fed from the optimizer 35 to the MPC duplicate 34. The tuning process may then proceed as described above with reference to Fig. 4, except that now the MPC duplicate 34 performs the simulation instead of the actual MPC 31 , and the optimizer 35 performs the optimizing of the tuning parameters instead of the optimizer 33. The new tuning parameters finally found may then be stored in the MPC 31 and taken into use, either automati- cally or if approved by the operator via the user interface. Finally, the optimizer 35 and/or the MPC duplicate 34 may be reset for a new tuning process.

Tuning operation may be triggered by the operator via a user interface (e.g. at the control room computer 1 1 ). Figure 7 illustrates an exemplary view at the user interface for controlling an MPC controller that is performing an online process control task (e.g. operation in the mode A in some of the embodiments above). The view at the user interface may show the current state and parameters of the MPC. The user interface may also be provided with means, such as a soft key shown in Figure 7, for triggering an offline tuning according to the invention. The operator may also be to accept or reject the new MPC tuning parameters for use from the user interface. The operator may, upon taking into use the new MPC tuning parameters, continue fine tuning of the MPC manually from the user interface, e.g. by manually changing parameters in the view shown in Figure 7, while the MPC is operating in the online process control task.

The techniques described herein may be implemented by various means. For example, these techniques may be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or combinations thereof. For a firmware or software, implementation can be through modules (e.g., procedures, functions, and so on) that perform the functions described herein. The software codes may be stored in any suitable, processor/computer-readable data storage medium(s) or memory unit(s) and executed by one or more processors. The data storage medium or the memory unit may be implemented within the processor or external to the processor, in which case it can be communicatively coupled to the processor via various means as is known in the art. Additionally, components of systems described herein may be rearranged and/or complimented by additional components in order to facilitate achieving the various aspects, goals, advantages, etc., described with regard thereto, and are not linnited to the precise configurations set forth in a given figure, as will be appreciated by one skilled in the art.

The description and the related figures are only intended to illustrate the principles of the present invention by means of examples. Various alternative embodiments, variations and changes are obvious to a person skilled in the art on the basis of this description. The present invention is not intended to be limited to the examples described herein but the invention may vary within the scope and spirit of the appended claims.

Claims

1 . A method of tuning a model -based process controller, particularly a model predictive control (MPC) type process controller, comprising

switching said controller from said online process control mode to an offline tuning mode for a tuning of the controller,

operating said controller in said offline tuning mode in the real-time operating system environment on the process control layer of the process or the machine under control of an optimizer unit to determine controller tuning parameters meeting a predetermined quality criterion, and

2. The method as claimed in claim 1 , wherein said switching to the offline tuning mode comprises disconnecting said controller from actual process signals and connecting predicted signals to the optimizer.

3. The method as claimed in claim 1 or 2, wherein said switching to the offline tuning mode comprises

disconnecting a feedback of controlled variables from a process or machine to the MPC controller,

providing tuning parameters from the optimizer to the controller.

4. A method of tuning a model -based process controller, particularly a model predictive control (MPC) type process controller, comprising

providing a duplicate of said controller to enable a controller tuning, and

operating said controller or said duplicate of said controller in the real-time operating system environment on the process control layer of the process or machine under control of an optimizer unit to determine controller tuning parameters meeting a predetermined quality criterion.

5. The method as claimed in any one of claims 1 to 4, comprising operating said optimizer in an online process optimization mode in a real-time operating system environment on a process control layer of a process or a machine,

operating said optimizer in said offline tuning mode in the real-time operating system environment on the process control layer of the process or machine to control said controller or the duplicate of said controller to determine the controller tuning parameters meeting the predetermined quality criterion, and

6. The method as claimed in claim 5, wherein said switching of said optimizer to the offline tuning mode comprises disconnecting said optimizer from actual process signals, connecting to the optimizer the predicted signals from said controller or from said duplicate of said controller, and changing to use said predetermined quality criterion in place of a process cost function in optimization.

7. The method as claimed in claim 5 or 6, wherein said switching of said optimizer to the offline tuning mode comprises

disconnecting a feedback of controlled variables from a process or a machine to the optimizer,

disconnecting a supply of manipulated variables from the controller, connecting predicted controlled variables and predicted manipulated variables from the controller or from the duplicate of the controller to the opti- mizer, and

providing tuning parameters from the optimizer to the controller or to the duplicate of the controller.

8. A method as claimed in any one of claims 1 to 4, wherein said optimizer is dedicated for the tuning and operated in the real-time operating sys- tern environment on the process control layer of the process or machine.

9. A method as claimed in claim 8, comprising providing said optimizer by duplicating an optimizer operating in an online process optimization in a real-time operating system environment on a process control layer of a process or a machine, and

using said predetermined quality criterion in optimization.

10. A method as claimed in claim 8 or 9, wherein said optimizer is at least partly integrated into the controller or into the duplicate of the controller.

1 1 . A method as claimed in any one of claims 1 to 10, comprising triggering said tuning from a user interface on plant management layer of the process automation system.

12. A method as claimed in any one of claims 1 to 1 1 , comprising accepting said determined tuning parameters for use from a user interface on plant management layer of the process automation system.

13. A method as claimed in any one of claims 1 to 12, comprising, upon determining said tuning parameters, continuing tuning of controller tuning manually from a user interface on plant management layer of the process automation system, while said controller is operating in an online process control task in the real-time operating system environment on the process control layer of the process automation system.

14. A tuning system for a model -based process controller, particu- larly a model predictive control (MPC) type process controller, comprising means for implementing for implementing steps of any one of method claims 1 to 13.

15. An executable program product comprising program code means stored on a processor readable medium for performing a method ac- cording to any one of claims 1 to 13 when said program product is run on one or more processors.