MXPA00003670A - Optimal auto-tuner for use in a process control network - Google Patents

Abstract

A device and method that automatically tune a valve controller coupled to a process control loop generate a plurality of sets of tuning parameters for use by the controller, deliver a test signal, such as a blocked sinusoidal signal, to the controller to force the process control loop through a test cycle while each of the plurality of sets of tuning parameters are being used by the controller and measure a response of the process control loop during each of the test cycles. The device and method then calculate a performance index for each of the plurality of sets of tuning parameters based on the measured responses and select one of the sets of tuning parameters based on the calculated performance indices. The selected set of tuning parameters is then loaded into the controller for use during normal operation of the process control loop.

Description

OPTIMAL SELF-TUNER FOR USE IN A PROCESS CONTROL NETWORK Technical field _ The present invention relates in general to auto tuners for use in process control networks and, more particularly, to process control auto tuners that determine an optimal set of tuning parameters for use to control a process or for its use for control a valve setter and a valve device in a process environment. Description of the Related Technique In the mid-1950s, process controllers with self-tuning (self-tuning) have been used in certain industries, such as the aerospace and process control industries, to automatically determine a set of tuning parameters, such as a set of gains, for use to control a process or a process control device such as a valve. In general, these adaptive or self-tuning controllers implement a method of identifying the system that determines one or more characteristics of a process or a device and a control design procedure that determines a suitable set of tuning parameters based on the determined characteristics of the system. process or device. System identification processes typically induce controlled oscillation within a process or device, measure the values of one or more process variables during controlled oscillation and then determine certain characteristics of the process or device, such as the final gain, the period end, and the time delay of the process or device based on the measured variables. These system identification procedures then use the characteristics of the process or device to identify the type of process or device that is being controlled based on standard mathematical procedures. Alternatively, some system identification procedures perform selection techniques by comparing models or tuning analysis to determine which of a set of stored mathematical models (or process syntheses) more closely matches or adjusts the data associated with the variables of the process measures. After the characteristics of a process or device are derived, or a model for the process or device is determined, the process or device is identified as one among several different types of, for example, linear processes so that it can be generated a set of definition equations for them. The control design procedure then calculates or otherwise determines a suitable set of tuning parameters (such as gains) based on the results of the system identification procedure and loads these tuning parameters in a process controller or a controller. device for use to control the process or the device. Because self-tuning or adaptive controllers have been known for a significant period of time, many system identification strategies, such as recursive least-squares approximations, Poisson's functional moment approximations, and description function approximation, have been developed for characterize a process or a device. Similarly, many control design strategies, such as pole positioning methods, Zeigler-Nichols methods, and modified Zeigler-Nichols methods, have been developed to determine a set of tuning parameters for use to control a process or device. after this process or device has been characterized. One of the most versatile control design techniques uses the Gaussian linear quadratic approximation (LQG) to select a set of tuning parameters for a controller. Nevertheless, most of these known methods, including the approximation by LQG, are only optimal when they are used with linear processes or linear devices or when they are used in processes or devices for which a set of linear equations can be identified. Consequently, most of these approaches, including the LQG approach, are suboptimistic when used to determine a set of tuning parameters to control processes or process control devices, such as control valves, which are non-linear in nature or They include peculiar and non-linear regions that are difficult to quantify. SUMMARY OF THE INVENTION The autosetter of the present invention uses a systematic experimental approach to determine a set of tuning parameters, such as a set of gains, for use to control a process or process control device and, as a result, it does not explicitly need to characterize a process or a device which, in turn, allows the auto-tuner of the present invention to develop an optimal set of tuning parameters for process control devices, both linear and non-linear. In general, the auto-tuner of the present invention forces a process or a process control device through a test cycle using each of a plurality of different sets of tuning parameters, measures the response of the process or of the control device of processes during each of the test cycles and determines which of the sets of tuning parameters minimizes a previously defined performance index. The auto tuner of the present invention then loads the determined set of tuning parameters in a controller that controls the process or process control device during normal operation. According to one aspect of the present invention a self-tuner develops a set of operational tuning parameters, such as a set of gains, for use by a process or a device driver that is connected within a process to receive a signal from reference. The autosetter includes a tuning parameter generator that generates a plurality of sets of tuning parameters for use by the controller during a tuning procedure, a test signal un-generator that supplies a test signal to the controller as the signal of reference during the tuning procedure and a data collector adapted to receive measurements of an input variable and an output variable associated with the process during the tuning procedure. A performance index generator determines a performance index associated with each of a plurality of sets of tuning parameters from the measurements of the input variable and the output variable and, after that, a parameter selection unit of tuning selects one of the plurality of sets of tuning parameters as the set of operational tuning parameters based on the performance indices. Preferably, the controller is coupled within a process control system for driving a valve setter that is connected to a valve actuator / valve device via a fluid pressure line. In this configuration, the data collector can be coupled to the valve setter and the valve to collect data related to, for example, the valve placement, the driving signal supplied or developed by the valve setter and one or more variables of intermediate states such as the pressure of the activator inside the valve activator. Also, preferably, the test signal generator develops a blocked sine signal or some other test signal that includes a multiplicity of discrete changes therein and the data collector collects a series of measurements for the input variable and for the variable output after each of the discrete changes in the test signal for each of the plurality of sets of tuning parameters. Even more, the performance index generator can calculate the performance index associated with each of the plurality of sets of tuning parameters as an expected value of performance indices associated with each of the discrete changes in the test signal or as a combination of a norm of difference between the reference signal and the output variable and a norm of the input variable. According to another aspect of the present invention, a self-tuner that develops a set of operational tuning parameters for a valve setter includes a data collector coupled with a valve for collecting data related to an input variable and an output variable and includes a signal generator that supplies a predetermined test reference signal to the valve setter during a tuning procedure. The auto-tuner also includes a computer program incorporated into a computer-readable medium that performs the steps of generating a plurality of sets of tuning parameters for use by the valve setter during the tuning procedure, calculating a performance index associated with each of the plurality of sets of tuning parameters from the collected data related to the input variable and the output variable and selecting one of the plurality of tuning parameter sets as the set of operating tuning parameters based in the performance indices.

According to still another aspect of the present invention, a method of automatically tuning a controller coupled with a process control circuit includes the steps of generating a plurality of sets of tuning parameters for the controller, forcing the control circuit of processes through a test cycle while the controller uses each of the plurality of sets of tuning parameters, measuring a response of the process control circuit during the test cycle associated with each of the plurality of parameter sets of tuning and calculates a performance index for each of the plurality of sets of tuning parameters based on the measured responses. After that, the method selects one of the set of tuning parameters as a set of operational tuning parameters for the controller based on the calculated performance indices and loads these operational tuning parameters in a controller. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a block diagram of a process control circuit including a prior art autosetter and controller. Figure 2 is a block diagram of a process control circuit including a setter, a control valve and a self-tuner according to the present invention. Figure 3 is a diagram illustrating a control waveform used by the auto tuner of the present invention; and Figure 4 is a graph of the performance index (J) associated with five different sets of gain vectors (k) calculated according to the tuning method of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring now to Figure 1, a standard process control system or circuit 10 includes a controller 12 that produces a control or impulse signal (u) which controls the operation of a process or a plant 14. A process variable (y) developed by the plant 14 is fed back to an adder 16, where it is subtracted from the reference signal or fixed point (r) to produce an error signal (e) which, at its once, it is provided to the controller 12. The controller 12 implements a standard control algorithm that changes the impulse signal (u) so as to force the process variable (y) to match the fixed point (r) whenever the signal of error (e) is different from zero. The process control system 10 also includes a self-tuner 17 having an identification block - of the system 18 and a control design block 20. The auto-tuner 17 implements a tuning procedure to develop a set of tuning parameters, such as a set of gains, for use by the controller 12. During this tuning procedure, the system identification block 18 provides a forcing function (such as a wave quadrature signal) to the adder 16 as the reference signal (r) which, in turn, causes the plant 14 to go into a state of controlled induced oscillation. At this time, the system identification block 18 measures the control or impulse signal (u) and the process variable (y), using these measured signals to characterize the plant 14 or to characterize a performance index associated with the plant 14 according to any identification technique or standard process model. Although auto tuner 17 is illustrated in Figure 1 by implementing an identification system in a closed loop manner, auto tuner 17 can implement system identification in an open circuit manner as well. After the system identification block 18 has characterized the plant 14, the defining characteristics are provided to the control design block 20 which develops a set of tuning parameters for use in the controller 12 based on the identified characteristics of the plant 14. In some systems of the prior art, the control design block 20 uses a Gaussian linear quadratic (LQG) approach to select an appropriate set of gain values. The LQG approach can generally be implemented by solving equation (1) below which is constrained by equations (2), (3) and (4). The coefficients in equations (2) and (3) can be estimated by, for example, the identification block of system 18. Equation (4) describes the feedback structure and does not have to be estimated. N-l minj. J = E { x ^ t-N] Sx [N] +% i li] Rx [i] + uT [i] Qu [i])} x [i + l] = Ax [i] + Bu [i] + w li] (2) and [i] = Cx [i] + Du [i] + w2 [i] (3) u [i] = k [i] x [i] (4) where: k [i] a gain vector in the i-th time period; J a performance index; N number of measured samples used to determine the performance index; [i] the plant state vector in the i-th time period; u [i] the impulse signal vector in the i-th time period; and [i] the output variable vector in the i-th time period; Wj_ [Í] = random-gaussian plant perturbation vector in the i-th time period; 2 [i] Gaussian random measurement perturbation vector in the i-th time period; A, B, C, D = matrices that describe the dynamics of a plant 14; and S, R, Q weighting matrix matrices that identify a preferred optimum operation of the plant 14. As a clarification, all vector variables are indicated herein by bold letters, lowercase, all matrices are indicated by bold uppercase letters , and all the scalars are indicated with standard letters. The expression E { f (x) } identifies the expected value of the function f (x) while the expression min ^ - J minimizes the scalar J with respect to the vector k. As will be understood from the above equations, when the LQG approach is used, the control design block 20 solves equation (1) to identify the gain vector k associated with the minimum performance index J for an identified set of control dynamics. the plant (defined by matrices A, B, C and D) and for a given optimal performance index (defined by the weighted coefficient matrices S, R and Q) • The right side of equation (1) can generally be thought of as the expected value of the sum of three penalty components. The first penalty component J ^ INJ Sx [N] places a penalty on the final value of the state vector x. Thus, when the final components of the state vector x are different from zero, the performance index J increases by an amount proportional to the sum of the squares of the components multiplied by the matrix of weighting coefficients S. The second penalty component N-i t £ XxX iij Rxíi] t = 0 places a penalty on the movement (eg, slow or oscillatory movement) of the state vector x in response to a change in the control vector or pulse u . The movement of the state vector x in a way that causes the non-zero evaluated components of the state vector x to increase the performance index J as a function of the sum of the square error of the state vector x multiplied by the weighting matrix R. The third component Nl £? uT [i] Qu [i]? = 0 places a penalty on large values of the input vector or impulse u. The large values of the impulse vector u cause the performance index J to increase by a function of the sum of the square value of the impulse vector u multiplied by the weighting matrix Q. If the plant 14 is linear and its parameters are well known or can be identified, a closed-form solution for the gain vector k can be found which leads to optimal control of the plant 14. For the most well-known adaptive controls or auto-tuners using the LQG approach, matrices A, B, C and D are estimated online and equation (1) is solved to identify the gain vector k associated with the optimal performance of the controller 12. However, as indicated above, the self-tuning algorithms, such as those using the LQG approach, are suboptimal when used in situations in which the dynamics of the process or device are very non-linear or when they have strong stochastic components because, b Under these conditions, it is difficult or almost impossible to identify a set of equations that precisely define the operation of the process or the device. In addition, many self-tuning algorithms are sensitive to noise, such as process noise or measurement noise, and are unable to operate optimally when a high level of noise is present within the signal that is being used to characterize the plant. In many cases, known auto-tuning algorithms operate satisfactorily only when the noise within the signal being measured is less than about one percent of that signal. Referring now to Figure 2, a process control circuit 30 having a self-tuner 32 that overcomes these problems is illustrated in detail. As will be apparent, the process control circuit 30 includes a setter 34 having a current-to-pressure transducer (I / P) 36 coupled with a pressure relay 38. The pressure relay 38 can be modeled by a first non-linear gain function using the output of the I / P transducer 36 to produce a relay path (v) and a second non-linear gain function which uses the -relay v path to produce a controlled amount of an air flow (w) in a fluid line 39 connected to the relay output 38. An activator 40 is connected through the fluid line 39 with the relay 38 and uses the air flow in line 39 to produce the pressure (p) over an activation area to thereby produce a force (f) which causes the movement of a valve element within a valve 42. A controller 44 (which may be part of the valve setter 34) is connected within the process control circuit 30 for receiving a fixed point or reference signal (r) and one or more process or device parameters which may be, for example, the valve position (z), the pressure (p) inside the activator 40, the air flow (w) within the fluid line 39 and the relay path (v) in the relay 38. Of course, these signals can be measured by any convenient measuring device and other signals, such as other device or process parameters, including process variables, may be used by the controller 44 if desired. As illustrated in Figure 2, the controller 44 includes an adder 46 that subtracts signals developed by five feedback paths from the reference signal (r) to develop an error signal in a forward path. The forward path includes an amplifier 48 which multiplies the error signal by a gain K ^ which supplies the output of the amplifier 48 to the I / P transducer 36"as the control signal or pulse (u). controller 44 includes a transfer function block 50 responsive to the relay path (v) and an amplifier 51 which multiplies the output of block 50 by a gain K2.The second feedback path includes a transfer function block 52 responsive to flow of air (w) and an amplifier 53 which multiplies the output of block 52 by a gain K. Similarly, the third feedback path includes a transfer function block 54 responsive to the trigger (p) pressure and an amplifier 55 which multiplies the output of block 54 by a gain K4 while the fourth feedback path includes a transfer function block 56 responsive to the position of the valve (z) and an amplifier 57 that multiplies the output of block 56 by a gain K5. The fifth feedback path simply provides the position of the valve (z) to the adder 46. As will be understood, the transfer function blocks 50, 52, 54 and 56 can be, for example, filters, and can implement any type or desired types of transfer functions. Of course, the specificity of the controller 44 of Figure 2 is merely exemplary, it being understood that other forward and forward paths could alternatively or additionally be used and that the controller 44 could implement any other type of control scheme including, for example, a Proportional integral control scheme (Pl), a proportional integral-derivative control scheme (PID), an internal model control scheme (IMC), and so on. The auto tuner 32 includes a system quantization unit 60 and a gain selection unit 62 that operates to implement a tuning method that selects or determines an optimal set of tuning parameters, such as a set of gains Kj, K2,. .., K5, for use by the controller 44. In general, during the tuning procedure, the gain selection unit 62 sends a plurality of different sets of gain values to the controller 44 and the system quantization unit. sends a test reference signal or a force function to the input of the reference signal of the controller 44 for each of the sets of gain values. During this time, the quantification unit of the system 62 measures data indicating the operation of the circuit 30 and determines, from the measured data, a performance index J associated with the control of the setter 34 and / or the valve 42 for each of the sets of Gain values. The gain selection unit 62 then selects a set of gains associated with the minimum performance index J and stores these gains in the controller 44 for use to control the setter 34 and the valve 42 during normal operation of the control circuit. process 30. As illustrated in Figure 2, the system quantization unit 60 includes a data collection unit 63 that collects and stores measurements of one or more variables that may be, for example, an input variable such as a impulse signal (u), one or more intermediate state variables such as the relay path (v), the air flow (w), the activating pressure (p), etc., and one or more output variables such as the position of the valve (z) or any convenient process variable. Of course, if desired, the data collection unit 63 can respond to other desired input variables, intermediate variables and / or output variables including, for example, the output of the adder 46, the output of the I / P transducer. 36, an indication of movement of the activator 40, or any other signal indicative of the control or operation of the setter 34 or of the valve 42. Likewise, the data collection unit 63 may collect other signals in addition to the pulse signal (u) as the input variable including, for example, the relay path (v) or one of the other intermediate variables. Also, the data collection unit 63 can collect other signals such as the output variable including, for example, the force developed by the trigger 40. During a tuning procedure, a gain generator 64 with the gain selection unit 62 provides a first set of gain values (e.g., K1 # K2, ..., K5) by forming a gain vector k to the controller 44. A test signal generator 66 within the quantization unit of the system 60 then provides a force function or test signal from the fixed point or reference input of the controller 4-4. If desired, the force function can be similar to that illustrated in Figure 3 which includes an initialization phase (which causes the valve 42 to move through a dead band to a particular location) followed by a phase of test that has a repetitive or periodic sine wave form blocked associated with it. After each change in the force function during each period of the test phase, for example, beginning at each of the times T T2, T3 and T4 in period 1 of Figure 3, the data collection unit 63 measures and records N values or time samples of one or more input variables, for example, the pulse signal (u), zero or more intermediate state variables, such as the air flow (w) or the pressure signal of activator (p), and one or more output variables, such as the position of the valve (z). Using this recorded data, a performance index calculator 68 calculates or determines a performance index J according to, for example, the following equation (which assumes that the impulse signal (u) is the only measured input variable and that the position of the valve (z) is the only measured output variable): my . J = E & (¿* (| (R [i] -r [0]) - (z [i] - z [0]) \ +? \ U [i] - u [0] |).}.? = 0 in where k gain vector comprising, for example, the gains Kj, K2, ..., K5; N the number of time samples for which the data is collected for the input variable (u) and the output variable (z) after each change in the force function (r); r [i] = the value of the reference signal or force function in the i-th time sample; z [i] = the value of the position of the valve (or other output variable) in the i-th time sample; u [i] = the value of the impulse signal (or other input variable) in the i-th time sample; Y ? = a scalar weighting coefficient (experimentally chosen to be approximately 0.3). As will be understood, the function within the keys. { } from equation (5) is calculated for N points of time samples after each change in the reference signal (r) resulting in four different calculations of the performance index J for each period of the blocked sine wave form of the Figure 3. These different calculations of the performance index J are desirable to fully quantify the controlled operation of the valve 42 during each different type of movement thereof. The graph in Figure 4 illustrates the values of the performance index J defined by the function within the keys. { } from equation (5) for five different sets of gain values (ie, for five different gain vectors k). In each set of gain values, four separate points J are illustrated and a different one of these four values J is associated with a different one of the changes in the force function of Figure 3. For example, in the first set of The gain values of Figure 4, a first value J 71 is associated with the change in force function (r) at time Tj, a second value J 72 is associated with the change in force function (r) at the time T2, a third value J 73 is associated with the change in the force function (r) at time T3, and a fourth value J 74 is associated with the change in force function (r) at time T . According to equation (5) the expected value (for example, the average) of the J values associated with each of the changes in the force function (r) of Figure 3 for a particular set of gains is determined by the performance index calculator 68. After the performance index J is calculated for a first gain vector k (or the data needed for that calculation is collected), a new set of gain values is chosen (Kj, K2, ..., K5) or gain vector k by the gain generator 64 and is sent to the controller 44 by the gain selection unit 62. At this time, the second period of the blocked sine-wave function of Figure 3 is supplied to the input of the reference signal of the controller 44 and the response of the process control circuit 30, the setter 34 and / or the valve 42 is measured by the data collection unit 63. If desired, the second period of the blocked sinusoidal signal may include the phase initializac to ensure that the valve 42 is in the same position as it was when it was tested with the first set of gains. After that, the performance index calculator 68 determines the performance index J associated with each of the movements of the force function (r) for this second set of gain values using equation (5) above. This procedure is repeated for any number of sets of gain values or gain vectors k. Five sets of four J values for five different sets of gain vectors k are illustrated in the graph of Figure 4, where the third set of gain values is associated with the minimum expected value of the performance index J over the four changes in each period of the blocked sine wave form of Figure 3. Of course, the gain generator 64 may choose or develop the different sets of gain values k in any desired manner. For example, one of the gains. Just as K ^ can vary and the other gains K2, K3, K4 and K5 can be calculated as a function of Kj. Alternatively, other minimization routines may be employed including, for example, a Downhill Simplex Nelder-Mead method, a simulated annealing method or a multidimensional conjugate gradient method, to name a few. Still further, predetermined gain sets can be stored in a memory and retrieved by the profit generator 64. However, preferably different gain vectors k are chosen to cover the stable-operation range of the process control circuit 30 or the setter 34-and the valve 42 in combination so that an optimum local or global set of gains can be determined at the end of the tuning procedure. After all the different sets of gain values or gain vectors k are stored in the controller 44 and used during a test cycle to operate the process control circuit 30, a gain selection unit 70 determines which set of gain (ie, which gain vector k) produced the minimum expected value of the performance index J and then directs the gain generator 64 to load the gain values in the controller 44. After that, the control circuit process 30 operates using the chosen set of gains until another tuning procedure is implemented. Of course, the tuning procedure described herein may be repeated at any time, including when the process circuit or the setter / valve combination is either on line or off line. It is important to note that equation (5) does not specify or require another plant model or a set of equations that defines any operation of a process control circuit or a process control device. As a result, this equation can be used to determine an optimal set of gains (or other tuning parameters) to be used to control any type of system or device, including any system or device that has non-linear characteristics and any system or device that uses non-Gaussian random processes, such as the processes associated with control valves and control valve placers. In addition, the tuning procedure and the tuner described herein may use other performance index calculations that calculate a norm in a vector space. These other calculations of performance indices can generally be expressed by the following equation: mink j = E. { a \ \ r- z \ | + / 31 | u | | + IT IP | I} <6 > where: a, ß, y = constants; r = a reference signal; z = an output variable, such as a valve position; u = an input variable, for example, a pulse signal, - and p = an intermediate variable, for example, a pressure signal. || * || denotes a norm, which can be, for example, a sum or an absolute value, a sum of a square value, a supreme, a power signal (such as an RMS signal), or any other norm of vector space. Choosing the absolute value rule, as in equation (5), tends to specify the minimum. Moreover, more or less input variables and intermediate variables could be used in equation (6) although the differences between a reference signal and other output variables could also be added to equation (6), as desired. Still further, any other force function, including a force function with only a single change, two changes, etc. can be used to test each of the different sets of gain values according to the present invention. However, it is preferable to use a force function that causes the movement of a valve (or that operates any other device or circuit under test) in each of the operating regions of that device or circuit. Of course, the components of the system quantization unit 60 and the profit selection unit 62 can be implemented in software in any suitably programmed processor, such as a microprocessor, or alternatively, they can be implemented in hardware, firmware, or any combination of software and hardware. Thus, for example, the data collection unit 63 can be any analog or digital storage unit coupled with a measuring device that measures the appropriate signals. Similarly, the test signal generator 66 may store one or more test signals in the memory and send an analog or digital signal to a reference signal input of the controller 44 using any standard or known method. Alternatively, the test signal generator 66 can be any known analog signal generator. Likewise, the profit generator 64 can generate profits in any convenient manner and, if desired, can generate profits in different ways selectable by a user. Preferably, the gain generator 64, the gain selector 70, and the performance index calculator 68 are implemented in software in a microprocessor but, instead, can be implemented in hardware or firmware.

The auto-tuner 32 can be located in the controller 44, in a separate unit, in the setter 34, in the valve 42 or in any other device capable of accepting inputs from the process control circuit 30 and of calculating gains and minimum values of according to, for example, equation (5) or equation (6). Likewise, the tuning method of the present invention can determine any other desired tuning parameter in addition to gains including, for example, time constants or filters, and so on. Although the tuning method of the present invention is very useful for controlling valves and valve setters (which are very non-linear), the tuning method of the present invention can be used to develop tuning parameters for any process control circuit or system that includes process control systems that include devices other than control valves and control valve setters. However, for the purposes of this invention, a combination of valve setter and valve device is considered to be a process control circuit while a valve setter is considered to be a device driver as well as a process controller. . When the auto tuner described herein is used to tune an entire process control circuit or system, the reference signal will typically be an operator input (instead of an output from a process controller), the input variable may be an output of the process controller and the output variable can be a process variable. It will be understood that the self-tuning technique described herein is not parametric, as a result, it does not require a parametric identification algorithm that identifies a process or a device. On the contrary, this technique can be applied to any algorithm or any process control system that includes linear and non-linear components. In particular, this technique can be applied to very non-linear and highly variable systems such as control valves and does not require the usual assumptions of random Gaussian process perturbations. Furthermore, this technique does not require that variations in the performance index J be stationary with respect to changes in profit. Moreover, by incorporating measures of the impulse signal (u) or, alternatively, the intermediate signals such as the relay path or the activating pressure (p) in the performance index J, the technique of the present invention allows one " see inside "of setter 34 and activator 40 to detect marginally stable behavior before it becomes a problem in the travel of the valve. Furthermore, instead of tuning a setter to determine stability, the auto tuner of the present invention optimally tunes, which minimizes the dead time and response time of a control valve while maintaining a reasonably stable pulse signal. . Also, unlike many description function auto-tuners, the autosetter of the present invention does not require that driver feedback gains be located in the forward control path but, instead, allows the gains to be applied from any way desired, including in reverse control paths. Although the present invention has been described with reference to specific examples that are intended to be illustrative only and not limiting of the invention, it will be apparent to those of ordinary skill in the art that changes, additions or deletions to the described embodiments can be made without departing from of the spirit and scope of the invention.

Claims (30)

1. A self-tuner for developing a set of operational tuning parameters for use by a controller receiving a reference signal and developing a pulse signal to effect changes in an output variable, the self-tuner comprising: a tuning parameter generator that generates a plurality of sets of tuning parameters for use by the controller during a tuning procedure; a test signal generator that supplies a test signal to the controller as a reference signal during the tuning procedure, - ^ a data collector adapted to receive measurements of an input variable and the output variable during the tuning procedure; a performance index generator that determines a performance index associated with each of the plurality of sets of tuning parameters from the measurements of the input variable and the output variable; and a tuning parameter selection unit that selects one of a plurality of sets of tuning parameters as the set of operating tuning parameters based on the performance indices. The auto-tuner of claim 1, wherein the controller is a device controller and is coupled within a process control system to drive a valve and wherein the data collector is coupled to the valve to collect data related to The valve. 3. The auto-tuner of claim 2, wherein the output variable is indicative of a valve position. The auto-tuner of claim 2, wherein the input variable comprises the pulse signal developed by the device driver. The autosetter of claim 2, wherein the collector is further adapted to receive a measurement of an intermediate variable during the tuning process and wherein the performance index generator determines the performance indices as a function of the intermediate variable . The auto-tuner of claim 5, wherein the intermediate variable is a valve actuator pressure. The auto-tuner of claim 1, wherein the test signal generator develops a blocked sine signal during the tuning procedure. The auto-tuner of claim 1, wherein the test signal generator develops a test signal that includes a multiplicity of discrete changes therein and wherein the data collector collects a series of measurements for the input variable and for the output variable after each of the discrete changes in the test signal for each of the plurality of sets of tuning parameters. 9. The autosetter of claim 8, wherein the performance index generator calculates the performance index associated with each of the plurality of sets of tuning parameters as an expected value for additional performance indices for each of the discrete changes in the test signal. The auto tuner of claim 1, wherein the tuning parameters comprise gains. The auto-tuner of claim 1, wherein the performance index generator calculates the performance index as a function of a standard of a difference between the reference signal and the output variable and as a function of a standard of the input variable. The auto-tuner of claim 1, wherein the performance index generator calculates the performance index generally according to the equation: min? J = E { i_ (| (r [í] -r [0]) - (z [i] - z [0]) | +? | u [i] - u [0] |)} where: J the performance index; k = a set of tuning parameters; r [i] = the value of the reference signal in the i-th time sample; z [i] = the value of the output variable in the i-th time sample; u [i] = the value of the input variable in the i-th time sample; N = a number of time samples for which the data is collected for the input variable (u) and the output variable (z), - and? = a weighting coefficient 13. The auto-tuner of claim 1, wherein the tuning parameter selection unit selects the set of tuning parameters associated with the minimum performance index as a set of operational tuning parameters. The auto-tuner of claim 1, wherein the controller is a process controller and wherein the output variable is a process variable. 15. A self-tuner for developing a set of operational tuning parameters for a valve setter that controls an output variable associated with a valve in response to a reference signal, the self-tuner comprising: a data collector coupled with the valve setter and the valve for collecting data related to an input variable and an output variable during a tuning procedure; a signal generator that supplies a test reference signal previously determined to the valve setter during the tuning procedure, and a computer program incorporated in a computer readable medium for the implementation of a computer that performs the steps of (a) ) generating a plurality of sets of tuning parameters for use by the valve setter during the tuning procedure; (b) calculating a performance index associated with each of the plurality of sets of tuning parameters of the collected data related to the input variable and the output variable generated in response to the supply of test reference signals previously determined to the valve setter, - and (c) selecting one of the plurality of sets of tuning parameters as the set of operational tuning parameters based on the performance indices. 16. The auto-tuner of claim 15, wherein the output variable is indicative of a valve position. The auto-tuner of claim 15, wherein the input variable comprises a pulse current signal developed by a control algorithm of the valve setter. The auto-tuner of claim 15, wherein the data collector collects additional data related to an intermediate variable during the tuning procedure and the computer program calculates the performance indices from the other data collected related to the intermediate variable . 19. The auto-tuner of claim 18, wherein the intermediate variable is a valve activating pressure. The auto-tuner of claim 15, wherein the signal generator develops a blocked sine signal during the tuning procedure. The auto-tuner of claim 15, wherein the computer program calculates the performance-index associated with one of the plurality of sets of tuning parameters as an expected value of other performance indices for each of a set of changes. discrete in a previously determined test reference signal. 2

2. The autosetter of claim 15, wherein the tuning parameters comprise gains. The auto-tuner of claim 15, wherein the computer program selects the set of tuning parameters associated with a minimum performance index as the set of operational tuning parameters. The auto tuner of claim 1-5, wherein the computer program calculates the performance index associated with each of the plurality of sets of tuning parameters as a function of a standard of a difference between the reference signal and the output variable as a function of a norm of the input variable. 25. A method for automatically tuning a coupled controller within a process control circuit, comprising the steps of: generating a plurality of sets of tuning parameters for the controller; forcing the process control circuit through a test cycle wherein the controller uses each of the plurality of sets of tuning parameters; measuring a response of the process control circuit during the test cycle associated with each of the plurality of sets of tuning parameters; calculating a performance index for each of the plurality of sets of tuning parameters based on the measurement response; and selecting one of the sets of tuning parameters as a set of operating tuning parameters for the controller based on the calculated performance indices. 26. The method of the claim. 25, wherein the step of measuring includes the step of measuring an output variable and an input variable. The method of claim 26, wherein the process control circuit includes a valve and wherein the output variable is indicative of a valve position. The method of claim 26, wherein the controller is a valve setter that receives a pulse signal and that includes a relay which is fluidly coupled to a valve actuator of a valve and wherein the variable of input comprises one of the pulse signal and one relay path associated with the relay. The method of claim 25, wherein the step of forcing the process control circuit through the test cycle includes the step of supplying a test signal having a multiplicity of discrete changes therein to the controller and in where the step of measuring the response includes the step of collecting a series of measurements for an input variable and for an output variable after each of the discrete changes in the test signal. The method of claim 25, wherein the step of calculating a performance index includes the steps of determining a first standard associated with a difference between a reference signal and an output variable, determining a second standard associated with an input variable and combine the first and second standards.