DE10041439A1 - Automatic tuning facility for a control loop based on non-linear estimators of the tuning rules - Google Patents

Abstract

A system for tuning a process control loop includes a tuning module for receiving an error signal representative of the difference between a setpoint and a process variable, the module generating a first process control signal to control the process. The system further includes a controller module for receiving the error signal and a parameter signal from a non-linear module to generate a second process control signal for controlling the process, the non-linear module using a non-linear procedure to generate the parameter signal. The system further includes a switching device that is coupled to the tuning module and the controller module in order to select the appropriate process control signal for controlling the process. The system provided uses non-linear approaches in the non-linear module to approximate the desired tuning parameters of the controller. The non-linear approaches include neural network tuning, fuzzy logic tuning, and non-linear functions, including sigmoid tuning. According to one embodiment of the present invention, the non-linear module includes a module for determining a process module and a tuning module for the controller, which provides controller parameters and model determination parameters using neural networks, fuzzy logic and non-linear functions, including sigmoid tuning.

Description

BACKGROUND OF THE INVENTION Scope of the invention

The present invention relates to a system and a method for tuning a process controller based on not linear estimates of voting rules including new ronal networks and fuzzy logic.

Description of the prior art

A proportional, integral, differential (PID) controller is a Regulator that is common for industrial processes, including computer-controlled industrial processes. Such PID controllers and their variations and combinations, for example P, PI, PD, enjoy wide application in the regulation of industrial processes. Typical industrial Processes are controlled by one or more control loops with PID Controllers regulated.

A fuzzy logic controller (FLC) is also a well-known process controller that is used to control Process parameters serves by adding process variables within with desired setpoints related parameters  being held. FLC's are non-linear controllers and are used in industrial environments are increasingly used.

A type of known method for tuning parameters of a PID controller is the Ziegler-Nichols method. The automatic Tuning based on that excited by a relay Vibration is also a well known and other known technique of automatic tuning. The automatic Relay vibration tuning determines the limit gain (Ultimate Gain) and the limit period of a process the Ultimate Period. The settings of the PID controllers can use these parameters using the rules of Ziegler-Nichols and modifications thereof can be determined. A Continuation of the relay oscillation vote, which over the Determination of the limit gain and limit period "System and Method for Automatically Tuning a Process Controller ", U.S. Patent No. 5,453,925, issued in September 1995 to Wilhelm K. Wojsznis and Terrance L. Blevins (hereinafter Wojsznis) provided, hereby in his Whole is included.

Significant progress has been made in the past few years Field of model-based coordination, especially with the Internal Model Control - (IMC) and Lambda tuning achieved. Both approaches result in a closed answer First order control loop according to the changes in the Setpoints. One related to the speed of response standing tuning parameter is used to the com to set promiss between performance and robustness. With both methods, the integral time parameter of the PID Controller (or the integral time parameter and the differential time parameters) adapted to the pole point (s) of the process eliminate and adjust the controller gain so that the desired response of the closed loop is achieved. IMC and lambda tuning have prevailed since  Vibrations and oversteer are avoided and the rule behave over the time constant of the closed control loop can be specified in an intuitive manner.

One of the limits of model-based tuning is the need determination of a process model. An equivalent First-order process model plus dead time with the parameters static gain, apparent dead time and apparent time constant is usually used for self-regulating processes Right. For processes with integrating behavior Model parameters of the integral gain of the process and the Dead time determined. The model is typically determined wise by means of a jump test of the open control loop. Compared to the method of that excited by a relay Vibration methods cannot be used for open control loops easily automate. For procedures with an open control loop are due to non-linearities of the process, valve hysteresis and load disturbances are often interventions by one People required to ensure the accuracy of a model to deliver. For self-regulating and processes with integrating Behavior requires a different technique.

A system and procedure for voting is required in a relay vibration environment that has the required PID tuning parameters across all areas of the model parameters provide and determine model parameters of a process.

SUMMARY OF THE INVENTION

As a result, a system is used to coordinate a process rule circle provided. The system contains a voting module for receiving an error signal that is for the difference between a setpoint and a process variable is, wherein the module a first process control signal to Rege process. The system still contains one  Controller module for receiving the error signal and a para metersignals from a non-linear module to a second Process control signal to regulate the process where to generate a non-linear procedure for the non-linear module Generation of the parameter signal applies. The system contains the further a switching device that with the voting module and the controller module is coupled to the appropriate Process control signal to control the process.

The system provided uses non-linear approaches in the non-linear module to approximate the desired tuning control parameters. The non-linear approaches include voting using neural networks, voting using fuzzy logic and non-linear functions Lich sigmoid tuning.

A system also provides that the non-linear module non-linear approaches used to get the parameters you want to approximate the process model. According to one embodiment of the The present invention is the determination of a process model using neural networks, fuzzy logic and non-linear Features including sigmoid tuning achieved with what better model parameters can be achieved in an advantageous manner can as with the state of the art analy formulas for the determination based on a Relay excited vibrations.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention and its numerous objects Features and advantages for the expert on the basis of the Be writing of the accompanying drawings even more clearly.  

Fig. 1A is a schematic block diagram of a relay system based tuning system according to the vorlie invention.

Fig. 1B is a graphic representation of the tuning according to an embodiment of the present invention.

Fig. 1C is a graphical representation of the non-linear co-efficient functions for calculating the integral time and the gain of the controller.

Fig. 2 is a graphical representation of the vote and step responses for the process with L / T ≈ 0.2 and scanning of the loop = 0.1 s.

Fig. 3 is a graphical representation of the vote and step responses for the process with L / T ≈ 0.5 and scanning of the loop = 0.1 s.

Fig. 4 is a graphical representation of the tuning and step responses for the process with L / T ≈ 0.7 and sampling the control loop = 0.1 s.

Fig. 5 is a graphical representation of the vote and step responses for the process with L / T ≈ 0.7, sampling of the control loop = 0.1 s with a fast response.

Figure 6 is a diagram of a tuner supported by a neural network.

Fig. 7 is a flowchart of steps required for the development of the neural network according to an embodiment of the present invention.

Fig. 8 is a graphical representation of the neural network predicted results factory integral time - the actual / required integration time using the method according to an embodiment of the present invention.

Fig. 9 is a graphical representation of the results from the neural network model predicted integral time - did neuter integral time / required integration time using the method according to an embodiment of the front lying invention.

Fig. 10 is a graphical representation of the neural network model for the sub-region of the T _i values 15 to 33 predicted results according to an embodiment of the present invention.

Fig. 11 is a graphical representation of the functions of belonging of the input variables in an embodiment of the present invention.

By using identical reference numbers in different Drawings are made of similar or identical elements indicates.

DESCRIPTION OF THE PREFERRED EMBODIMENT (S)

Reference is now made to FIG. 1A, which is a schematic block diagram of a tuning system based on vibrations 100 excited by a relay that can be controlled in accordance with the new non-linear approaches. A process 190 can be any controllable process. An output signal 130 represented as a process variable (PV) is provided by the process 190 and added in a summer 120 , in which the PV is compared with a target value (SP) 110 . The difference between PV and SP, which is used for the determination of the tuning system based on relay vibrations 100 , provides the limit gain K _u and the limit period T _u for a process. As shown, based on relay vibrations 100 , the voting system includes a summer 120 , a relay 150, and voting rules 160 . The setpoint (SP) 110 is supplied to the summer 120 together with the process variable (PV) 130 . Summer 120 subtracts PV from SP and provides the result to relay 150 , tuning rules 160, and controller 170 . The process is coordinated either automatically or under the control of a user. Optionally, the process 190 tuning may include an automatically controlled tuning process based on the natural vibration, such as the procedure described in Wojsznis. The natural vibration procedure includes controlling a switch 180 so that either the output of controller 170 is coupled to process 190 or the output of relay 150 is coupled to process 190 . In one embodiment, the relay 150 executes the natural vibration procedure with the tuning rules 160 by controlling the switch 180 . Switch 180 couples either controller 140 or voting rules 160 and relay 150 to process 190 . The procedure is an oscillation procedure which determines the time delay, the limit gain and the limit period.

The apparent dead time Td is determined when the tuning is initialized. The apparent dead time Td is determined by using a tangent as the slope of the process output PV 130 during the initialization of the tuning. The tangent is extrapolated to intersect the process output setpoint (SP) or mean line before tuning. The time between the first relay jump and this cut is the apparent dead time. Other methods of determining Td are also within the scope of the present invention. Such other procedures include those provided by Wojsznis.

Figure 1B shows a graphical representation of the process input and output signals during parts of the oscillation procedure. The time duration between the time t1 ( 10 ), switching of the relay 150 , and the time t2 ( 20 ) at which the process output reaches the maximum approximately corresponds to the dead time. Otherwise the dead time is calculated as the difference between the time of the | SP-PV | increment and the time of the | SP-PV | decrement. The duration of the evaluation with this approach corresponds to the duration of one or more limit periods (Tu). Apparent dead time can be calculated as an average of two or three results.

Dead time, limit amplification and limit period are sufficient to calculate a first-order process model plus dead time. Equations (1) and (2) for calculating the first order plus dead time are:

in which:
T _c = time constant of the process
T _u = limit period
T _d = apparent dead time of the process
K _s = static reinforcement of the process
K _u = limit gain.

The time constant of the process in equation (1) is expressed by a tangent function that provides a good approximation for the arguments smaller than π / 3 when the dead time is relatively large compared to the time constant. For processes with insignificant dead time, there is a large error in the calculation of the time constant, even with a small error in the specification of the dead time (the tangent argument is approximately equal to π / 2, and a small error in the argument leads to a large error in the tangent value ). A certain improvement results from using a linear function as shown in equation (3) for arguments larger than π / 3:

Ziegler-Nichols (ZN) rules for tuning a PID Controller

The tuning based on vibrations excited by a relay naturally corresponds to the ZN rules and provides the limit gain K _u and the limit period T _u . The original ZN formulas for PI controllers are as follows:

K = 0.4 K _u ; and T _i = 0.8 T _u .

These formulas result in a phase reserve that is between about 20 ° and fluctuates strongly by 90 °, depending on the ratio τ of the dead time L. of the process at the time constant T of the process. Hence the behavior also varies considerably, namely from an extreme vibrating response in a process with a ratio close to 0.1 to an extremely sluggish response in a process a ratio close to 1.0.  

A PID controller that complies with the original ZN rules

K = 0.6 K _u ; T _i = 0.5 T _u and T _d = 0.125 T _u

matched behavior shows similar behavior. In order to solve this problem, various modifications of the original formula have been proposed. There is a tendency to decrease the gain and shorten the integral time, as can be seen from the following formulas:

K = 0.4 K _u ; T _i = 1/3 T _u and T _d = 1/12 T _u .

The above modification improves the behavior of the rule circle with τ close to 0.5; Control loops with small values of τ however, are swinging even more strongly.

Other, more flexible formulas (4), (5) and (6) provide the following for the design of the phase / gain reserve:

T _d = αT _i (5)

K = K _u cosϕ / G _m (6)

in which:
α Design selection of the ratio T _d : T _i with the standard value 0.15.
G _m desired gain reserve with the standard value 2.0.
ϕ phase reserve.
K, T _d and T _i are the parameters of the controller.

With a given phase and gain reserve, the formula (7) provides constant coefficients to calculate T _i , T _d and K from T _u and K _u . A typical design with ϕ = 45 ° gives the following coefficients:

K = 0.38 K _u ; T _i = 1.2 T _u and T _d = 0.18 T _u (7)

This design is suitable for small τ, but results in an extreme sluggish response behavior for τ greater than 0.2.

If it is assumed that the phase reserve ϕ = 33 ° and the amplification reserve = 3.0, the following coefficients result in formula (8):

K = 0.27 K _u ; T _i = 0.87 T _u and T _d = 0.13 T _u ; (8th)

These coefficients are suitable for the design of a narrow one τ range close to 0.25.

Another known modification sees the definition of controller parameters as functions of the normalized dead time

or the normalized gain k = 1 / K _p K _u , where K _{p is} the static gain of the process. However, the above approach uses both the dead time and the time constant or the static gain of the process. It cannot therefore be used directly for tuning based on relay vibrations.

Non-linear estimators of the voting rules

When developing non-linear estimators of voting re certain assumptions and considerations must be taken into account First, all input parameters during the Relay vibration tests obtained (i.e. the limit gain, the limit period and the dead time). Second, there is  biggest shortcoming of the Ziegler-Nichols rule in an unadjusted one Integral time of the controller in processes with low dead time and an excessive integral time in processes with significant Dead time. Third, the voting rules should vote provide parameters and responses from the controller that correspond to those come close to model-based tuning (IMC or Lambda).

The relay vibration test for the tuning shown in FIGS . 1A and 1B and described in the corresponding discussion provides the limit gain K _u and the limit period T _u . In order to overcome the shortcomings of the CN rules, non-linear estimators have proven to be an improvement in the definition of voting parameters.

A sigmoid expression provides a smooth transition between two different values and is used to develop non-linear estimators. The following formulas meet the above requirements:

T _i = f ₁ (T _u , L) T _u ; K = f ₂ (T _u , L) K _u ; T _d = d ₁ T _i ; (9)

f ₂ = (T _u , L) = a2 + (b2 - f ₁ (T _u , L) / c2; (11)

where a1, a2, b1, b2, c1, c2 are heuristic coefficients and
a1 is between 0.3 and 0.4;
b1 is between 0.25 and 0.4;
b1 ≈ 0.6; b2 ≈ 1.0; c1 ≈ 7.0; c2 ≈ 4.0; d1 ≈ 0.125; d2 ≈ 1.0.

Formula (10) gives the value of the coefficient used to calculate T _i , which fluctuates between a minimum value of a1 and a maximum value a1 + b1, as shown in FIG. 1C. Formula (11) provides the adjustment of the coefficient for the calculation of K in the range:

[a2 + (b2 - a1 - b1), a2 + (b2 - a1)].

This approach results in significantly improved tuning responses that are close to IMC or lambda tuning (other than the decay of ZN quarter amplitudes). Some typical step responses for different L / T are shown in FIGS. 2 to 6. An example of a loop tuning and a step response for a second order process with gain = 1; T1 = 10 s; T2 = 3 s; L = 2 s is shown in Fig. 2.

The tuning system based on relay vibrations 100 produced the following controller setting: K = 1.65; T _i = 12.36 and T _d = 1.97 compared to the IMC calculations: K = 1.0; T _i = 12.5 and T _d = 1.97. FIG. 3 shows a graph with L increased up to 5 s and FIG. 4 shows a graph with L increased up to or 8 s. In both FIG. 3 and FIG. 4, the step responses are similar to that of model-based voting, although the integrating behavior is somewhat weaker than that required for the IMC response (the voting device gave T _i = 15.7 and 18.37 in comparison 14.0 and 15.5 according to IMC calculations).

A design of the tuner in accordance with the new non-linear approaches of the present invention allows a user to adjust the voting behavior by providing the options Slow (normal) and Fast (fast). Fig. 5 shows e.g. B. the step response of the control loop for the process L / T ≈ 0.7 with the choice 'Fast'. The use of speed selection increases the overall flexibility of the design by better adapting the voting response to specific conditions and requirements. It should be obvious to one of ordinary skill in the art that a DeltaV ^™ system manufactured by Fisher-Rosemont Systems, Inc. implemented in a Windows NT application program or other suitable program can implement the new customizable choice of responses.

Reference is now made to FIG. 6, in which a tuning device based on relay vibrations is shown in a schematic block diagram, which uses the new non-linear approaches according to an embodiment of the invention opti onally. The control loop system shown can include both non-linear tuning techniques and non-linear process modeling techniques. Optionally, the system can include linear tuning techniques and non-linear process modeling techniques or non-linear tuning and linear process modeling techniques.

As the embodiment according to FIG. 6 shows, a process 630 is coupled via a switch 670 either to a tuning device based on relay vibrations 600 or to a controller 610 . The controller 610 is coupled via a second switch 690 in such a way that it receives either a first parameter signal 651 from the controller design 650 or a second parameter signal 641 , which is output by the analytical controller design 620 and the process model 640 . Controller 610 provides an output when it is switchably coupled to process 630 . The tuner based on relay vibrations 600 , process model 640 , controller design 650 , controller 610 , process 630, and analytical controller design 620 can all be implemented as software modules, hardware, or a combination of both.

As will be described later, the switch 670 is set to the corresponding process control signal to regulate the process 630 . The tuner based on relay vibrations 600 is also coupled to provide signals to a non-linear module 660 that includes process model 640 and controller design 650 . Both controller design 650 and process model 640 use neural networks, fuzzy logic, non-linear functions, or other non-linear or linear techniques, as described below. The tuner based on relay vibrations 600 and controller 610 receive an error signal 680 representative of the difference between a setpoint and a process variable and generate a first process control signal to control process 630 . The tuner based on relay vibrations 600 also provides signals to the non-linear module 660 .

The first parameter signal and the second parameter signal are coupled to the controller 610 via a switch 690 . The output of the process model 640 is shown coupled to the analytical controller model 620 . A parameter signal 641 contains process model parameters that are contained in a parameter signal 642 that is calculated using non-linear techniques in the process model module 640 . In a preferred embodiment, the parameter signal 641 further contains controller parameters; that are calculated in the analytical controller design 620 using the process model parameters determined in the process model module 640 and output in the signal 642 . For the average person skilled in the art, it goes without saying that controller parameters can be generated from process model parameters. Accordingly, the analytical controller design 620 optionally uses linear techniques to determine the controller parameters.

The parameter signal 651 output by the controller design 650 contains controller parameters that are calculated using non-linear techniques. The switch 690 coupled to the controller 610 provides the option to (1) select the controller parameters that are calculated in the analytical controller design 620 along with the process model parameters generated using non-linear techniques in the process model 640 ; or (2) choose the controller parameters that are calculated using non-linear techniques in controller design 650 .

Alternatively, the analytical controller design 620 is removed from the system implementation in one embodiment. Without the analytical controller design 620 , switch 690 can be coupled to both parameter signal 651 and parameter signal 642 . Accordingly, the switch 690 is not required in this embodiment. The controller 610 thus receives both the controller parameters in the parameter signal 651 and the process model parameters in the parameter signal 642 . Both parameter signals 651 and 642 are generated in the non-linear module 660 using non-linear techniques such as neural networks, fuzzy logic, non-linear function approaches or other non-linear or linear techniques, which are described below.

Tuning based on neural networks

Referring to Fig. 6, the controller design 650, the controller parameters are determined either by means of an approach of neural network modeling. Such an approach with the neural network modeling improves the adaptation of the three PID controller parameters over a wide range of changes in the model parameters. Control concepts using neural networks are generally divided into two broad categories. One approach involves replacing a controller with a neural network. The neural network is trained by mimicking a controller or a human expert. With this approach, each control loop must be trained separately so that it is not a good candidate for a voting model that requires voting results in a simple and standardized voting procedure. An alternative approach uses neural networks as support for modeling, the implementation of the rule laws or for monitoring. A specific design of a tuning device is shown in FIG. 6. The tuner optionally uses neural networks or other non-linear techniques described below to calculate the process model and parameters of the PID controller.

Similar to the natural vibration procedure described in connection with FIG. 1, the tuner of FIG. 6 also uses a natural vibration procedure. The natural vibration procedure provides control of the switch 670 in such a way that either the output of the controller 610 is coupled to the process 630 or the output of the tuning device is coupled to the process 630 based on relay vibrations 600 . In one embodiment, the controller 610 receives the controller parameters from the controller model 650 or the analytical controller model 620 via the switch 690 . In this embodiment, the analytical controller design 620 receives parameters for determining the process model from the process model 640 . In another embodiment, analytical controller design 620 is removed from the process control loop. Accordingly, controller 610 receives parameters for determining the process model from process model 640 and controller parameters from analytical controller design 620 . The switch 670 decouples either the controller 610 or the tuner based on relay vibrations 600 with the process 630 . The controller design 650 uses non-linear techniques to determine the controller parameters. One technique described below uses neural networks.

The procedure for developing neural network techniques includes several basic steps, which are shown in FIG. 7. Figure 7 is a flow diagram of the steps in creating neural networks. Step 700 shows the first step in which simulated control loop configurations are developed including future applications of the tuner. Usually a process module of either second or third order plus dead time is used. Step 710 specifies an assumed range for the changes in the process parameters. Step 720 dictates that the developer calculate the PID controller settings for any set of model parameters using IMC rules, lambda tuning, or another preferred controller concept. Step 730 provides that the developer runs the automatic voting facility for the same set of model parameters and records the voting results, T _u , K _u, and L. Step 740 provides that the developer checks the behavior of the control loop with the calculated parameters and adjusts the parameters as necessary. Step 750 provides that the developer trains the neural network using the simulated voting results as inputs to the neural network and the model and controller parameters as outputs of the neural network. Step 760 provides that the developer implements the trained neural network in the tuner.

The neural network described above with reference to Figure 6 can be of any type (sigmoid or radial basis functions). In one embodiment of the present invention, the neural network is used with the neuron transfer function, which is given by:

as a neural input. In is a weighted sum of external inputs.

In = Σw _i In _i .

The neural network is selectively used with multiple outputs (outputs K and T _i for the neural network of the controller and K _p and T for the neural network of the model) or as multiple neural networks with a single output. The advantage of the single-output neural network is faster training. For this reason, one embodiment implements a single output neural network. The following inputs and outputs of the neural network are defined in an exemplary concept for a voting device:
Inputs: T _u , K _u , L, noise level, relay hysteresis and scanning speed.
T _u , K _u , L and the noise level are defined during the tuning test.
The relay hysteresis and the scanning speed are parameters of the tuning device.
Outputs: K, T _i , T _d , K _p and T.
K, T _i and T _d are parameters of the PID controller.
K _p and T together with L are parameters of the process model of the first order plus dead time.

Implementation issues and test results

An example of the new non-linear methods will be neural network models implemented to automatic Voting device (autotuner) based on a Relay excited vibrations in a scalable controller improve system for industrial purposes. The car tuner has two blocks: the tuner function block, the one in the controller is implemented, and the application of the voting facility tung. The application of the voting facility is in one Windows NT console or another suitable console in the implemented. The neural network models are used added to the voting facility. With an execution form are the neural network models for the user of the Voting device transparent, the autotuner none options belonging to neural networks or Settings.

According to the method described here, the neural Network for the second-order process model plus dead time trained. The following specification defines the ver available input and output data for training neural Network models:

Process model based on a neural network

Inputs: limit gain K _u , limit period T _u and dead time L, defined during the relay-based tuning experiment.

Outputs: process gain: 0.5; 1.0; 1.5; Time constant 1 of the process in seconds: 1.0; 2.0; 5.0; 10.0; 20.0; 50.0; 100.0; 200.0; Time constant 2 of the process in seconds: 1.0; 2.0; 5.0; 10.0; Dead time in seconds: 1.0; 2.0; 5.0; 10.0; 20.0; 50.0; 100.0; 200.0.  

Controller model based on a neural network

Inputs: as for the process model based on a neural Network.

Outputs: Gain K of the PID controller, integral time T _i and differential time T _d , calculated from the process model parameters given above for the outputs, ie process gain 0.5; 1.0; 1.5; Process time constant 1 in seconds: 1.0; 2.0; 5.0; 10.0; 20.0; 50.0; 100.0; 200.0; Time constant 2 of the process in seconds: 1.0; 2.0; 5.0; 10.0; Dead time in seconds: 1.0; 2.0; 5.0; 10.0; 20.0; 50.0; 100.0; 200.0.

Experimental testing

An experimental test of the voting simulation was carried out 142 times, which is approximately the minimum number of scans that are suitable for training a simple neural network. The samples gave a good correlation coefficient and generally values of the predicted output that came close to that of the required output. The graphs of the results for T _i are shown in FIGS. 8 and 9.

At some intervals of the predicted parameter range, the prediction error was not acceptable. There are two specific cases in this regard. In one case, these are small values of the predicted controller parameters (relative to the prediction error). In this case, even a predicted parameter of only 1% of the maximum value gives unacceptable parameters. As can be seen in particular from Fig. 8, T _i is zero or even negative if the actual value is close to the prediction error can.

Another error situation occurs if the sub-area contains too few samples. For example, if a user wants to expand the area of neural network modeling, for an extended sub-area, e.g. B. T _i <50.0 in Fig. 8, only a few simulations are performed, an error could occur.

Voting rules based on a neural network

To avoid anomalies, the voting models apply Basis of neural networks according to the exemplary, non-linear Techniques following rules or steps. First are like that capture as many samples as possible. At least that's it Capture minimum number of samples for a simple Model are required. Second, the area for the choose predicted parameters so that there is an inside of the range of the parameters only a few times (approx. 5) different. The predicted values change frequently (approx. 15 times), is the range of parameter changes to several Subdivide and there are several neural networks to develop models. Third, are small predicted values to be treated with special attention. Are the previous ones said values compared to the prediction error is small for the small values to a separate neural network model put.

For example, a model with a narrowed sub-range from 15 to 33 (instead of 1 to 320 of the full range) was developed. Although the number of samples was below the minimum number, the prediction was much better than the original model for the whole area. A graph of the prediction of the neural network model for the subrange of T _i is shown in FIG. 10. The prediction error relative to the IMC parameters is less than 5%, while the error of the non-linear function estimator for the test cases L / T <0.5 exceeds 12.0%.

Linear and non-linear correction functions for determination of the process model

Reference is now made again to FIG. 6, according to which the use of linear and non-linear correction functions for specifying the process model is described together with formulas (1) and (3) (repeated below). As described above, the process model 640 provides parameters of the process model using either linear functions or linear correction functions. The techniques described below use linear, non-linear functions, neural networks, and fuzzy logic as correction functions to determine the process model that are executed in process model 640 .

The time constant (T _c ) generated from formulas (1) and (3) is considerably smaller for smaller ratios of apparent dead time of the process for limit amplification

and an excessively high time constant (T _c ) in relation to the apparent dead time of the process for limit amplification

greater than 0.25 as from Table 1 can be seen.

A correction function f (T _u / T _d ) applied to the specified time constant of the process

T _c (corr.) = T _c (given) f (T _u / T _d ),

should have f (T _u / T _d ) values greater than 1 for a small T _d / T _u and less than 1 for T _d / T _u greater than 0.25.

A simple linear function with coefficients developed from simulated tests corrects the time constant of the process according to equation (12):

For a non-linear function, a sigmoid expression ensures a smooth transition between two different values. A sigmoid correction function is well defined if the minimum and maximum values are known. The following formula (13) for estimating the time constant of the process was developed based on a small amount of simulated data:

Formula 12 gives the value of the coefficient used for the calculation of T _c , which varies between a minimum value of 0.8 and a maximum value of 2.2. After correcting the time constant, the static gain of the process is calculated again using formula (2), which is repeated below:

The coordination and calculation results for the process with the static gain 1.0 are in Table 1 and with the sta table gain 2.0 compiled in Table 2. In at In the cases, the process is a second order model with T1 = 10 s; T2 = 2 s plus dead time modeled according to column Td. The Identifier approached the model by a first order plus model Dead time on (Td determined). The model parameters shown are given first (Td, Kp and Tc) and by means of linear and nonlinear estimation formulas (Kp and Tc) corrected.

Table 1

Results of specifying the process model with Kp = 1.0

Table 2

Results of specifying the process model with Kp = 0.5

The application of both correction functions improved the loading process model considerably.

Fuzzy logic-based determination

A typical fuzzy logic controller is a non-linear controller type. Unique features of a fuzzy logic controller are the functioning and development of the controller. A fuzzy Control algorithm is defined by linguistic rules. The Controller input parameters are represented by fuzzy sets animals. The functioning of the controller is determined by the rule surface, especially for controllers, the input parameters use, well presented.  

The procedure for developing fuzzy logic (FL) functions is similar to the procedure for developing a fuzzy logic Controller. The following procedure is used to determine a FL correction function.

First, it is assumed that the correction function for the time constant of the process model has the following form:

Tc (corr.) = Tc [1 + ΔF]; (14)

ΔF is a fuzzy logic correction function. The value the function corresponds to the incremental output of the fuzzy Logic controller. The potential benefit of using it a fuzzy logic function compared to a normal one linear function is that the fuzzy logic Function has a more flexible structure and in a simple way can take two or more arguments.

Second, the function has two arguments to use where the ratio Tu / Td is the first and the limit gain the second argument is.

It is possible to use different numbers and different types of membership functions. For the simple implementation, simple triangle membership functions for the input parameters and singleton functions for the output signals are to be used. Fig. 11 and Table 3 show details of the structure of a fuzzy logic correction function. Regarding Table 3, P means positive and N means negative.

Negative: N = -0.4; negative small: NS = -0.2; positive: P = 0.5; positive large: PL = 1.2

Table 3 Inference rules for the correction function

Defuzzification of the output signal is for singleton Membership roles apply.

Table 4

Comparison between the specified time constant and process gain from various correction functions

Experimental results of fuzzy logic simulation

The fuzzy logic correction function resulted in time constants and Gain estimates that are comparable to the estimates using linear and non-linear correction functions.

According to Table 4, one based on relay vibrations Identifier with added correction functions a model first order plus dead time, for model-based tuning calculation of a PID controller can be used.

Both non-linear functions and neural network models improve the voting rules based on by a relay excited vibrations considerably. In compliance with the predetermined principles in the development of the vote ba Modeling on neuronal networks provides neuro networks a number of advantages. A user can any voting rules (not necessarily lambda or IMC) implement and develop on neural Network-based voting models use the same methodology apply. A neural developed in the simulation Network model may have certain features of the voting draft take up. Sampling speed and noise level that the The design of the controller can be influenced in a simple manner Add way as an input parameter to the prediction.

Other embodiments

Although the systems and methods described herein are one Use the voting facility, the limit period and the limit gain of a process calculated to informa  Any other type of voting can develop facility that measures process parameters, including open loop voting facilities and other closed rule voting facilities circle. Furthermore, by means of the disclosed systems and method certain factors and control parameters either be entered by a user or automatically.

Furthermore, you can in all schematic block diagrams elements referred to herein implemented in hardware or in a suitably programmed digital computer or processor, that with software - either in the form of separate programs or as modules of a common program - is programmed, be implemented.

Although the present invention is by reference to certain Examples that are illustrative but not as the invention intended to be restrictive, has been described it is obvious to the professional that changes, additions gene and / or deletions of the disclosed embodiments can be made without devoid of spirit and scope to deviate from the invention.

Claims (18)

1. System for coordinating a process control loop, the system comprising the following:
a tuning module ( 600 ) for receiving an error signal ( 680 ) representative of the difference between a setpoint and a process variable to generate at least one process control signal for controlling the process ( 630 );
a non-linear module ( 660 ) for applying a non-linear procedure to generate at least one parameter signal;
a controller module ( 610 ) for receiving the error signal and the at least one parameter signal from the non-linear module ( 660 ), the controller module ( 610 ) generating a second process control signal for controlling the process; and
a switch ( 670 ) coupled to the process to couple either the tuning module or the controller module to the process to select the appropriate control signal to control the process. 2. The system of claim 1, wherein the non-linear module ( 660 ) includes:
a non-linear module ( 640 ) for determining the process for applying a non-linear procedure for generating parameters for determining the process model. 3. System according to one of the preceding claims, in particular according to claim 1, wherein the non-linear module ( 660 ) contains the following:
a non-linear controller design module ( 650 ) for providing a non-linear estimate of a variety of parameters for controller tuning. 4. System according to one of the preceding claims, in particular according to claim 3, wherein the controller of the further coupled so that it can estimate the Tuning parameters for the controller from the non-linear Controller design module receives. 5. System according to one of the preceding claims, especially according to claim 3, wherein a non-linear Controller design module Adjustments to the controller module allowed, the adjustments the Change controller response speed. 6. System according to one of the preceding claims, in particular according to claim 1, further comprising:
an analytical controller design module ( 620 ) coupled to the non-linear module, the analytical controller design module ( 620 ) being able to provide the controller with a variety of controller parameters based on a variety of parameters received from the non-linear module to determine a model . 7. System according to one of the preceding claims, in particular according to claim 1, wherein the controller is a pro proportional, integral and differential controller.   8. System according to one of the preceding claims, in particular according to claim 1, wherein the linear module ( 660 ) contains at least one neural network module. The system of any preceding claim, particularly claim 1, wherein the non-linear module ( 660 ) includes a non-linear module ( 640 ) for determining a process and a non-linear controller design module ( 650 ), an output of the non-linear Module ( 640 ) for determining the process and an output from the non-linear controller design module ( 650 ) are coupled to the controller module. 10. System according to one of the preceding claims, in particular according to claim 8, wherein the neural Network either a sigmoid function or a radial one Basic function used. 11. System according to one of the preceding claims, in particular according to claim 8, in which the neural network module uses a sigmoid function in which a neuron transmission function is given to:
where In is a weighted sum of external inputs, which has the form In = Σw _i In _i . 12. System according to one of the preceding claims, in particular according to claim 1, wherein the non-linear Module contains a fuzzy logic module. 13. System according to one of the preceding claims, in particular according to claim 1, wherein the parameter signal contains a variety of control parameters that are non-linear  Estimator for a variety of tuning parameters for use the coordination of the process control loop. 14. System according to one of the preceding claims, especially according to claim 13, wherein the non-linear Module using the variety of control parameters non-linear estimator with non-linear functions calculated to nonlinear estimators for the multitude of Generate voting parameters. 15. System according to one of the preceding claims, especially according to claim 13, wherein the non-linear Module using the variety of control parameters nonlinear estimator calculated with neural networks, around the vibrations excited by a relay based tuning parameters. 16. System according to one of the preceding claims, especially according to claim 13, wherein the non-linear Module using the variety of control parameters non-linear estimator with fuzzy logic calculated to the on the vibrations excited by a relay based tuning parameters. 17. System according to one of the preceding claims, in particular according to claim 14, wherein the non-linear Function is a sigmoid function. 18. System according to one of the preceding claims, in particular according to claim 17, in which the heuristic Coefficients can be used with the sigmoid function around parameters, including an integral time constant, a gain factor and one To provide differential time constants.