Self-monitoring tuner for feedback controller

• US 5,159,547 A
• Filed: 10/16/1990
• Issued: 10/27/1992
• Est. Priority Date: 10/16/1990
• Status: Expired due to Fees

• Current Assignee: Rockwell International Corporation
• Original Assignee: Rockwell International Corporation
• Inventors: Chand, Sujeet
• Primary Examiner(s): Smith, Jerry
• Assistant Examiner(s): MUIR, PATRICK
• Application Number:
  US07/598,488
• Time in Patent Office: 742 Days
• Field of Search: 364/157, 364/159, 364/151, 364/150, 364/148, 364/149, 364/152, 364/159, 364/162
• US Class Current: 700/37
• CPC Class Codes: G05B 13/0275 using fuzzy logic only
• CPC Class Codes: Y10S 706/90 Fuzzy logic
• CPC Class Codes: Y10S 706/906 Process plant

Abstract

An automatic tuner is provided for continuous, on-line tuning of proportional, integral, and derivative (PID) feedback controllers. The tuner compares the system input signals with the system response to generate estimates of system damping, frequency, and steady-state error, and then applies a set of if-then rules derived from mathematical and/or empirical analysis of the system parameters. The tuner represents each parameter of the rule-set by a "fuzzy" membership function, and an inference mechanism uses fuzzy logic for computing system outputs from the input values. The tuner also incorporates a self-monitoring mechanism to adjust the tuner output based on past performance. At each running cycle, the tuner computes a Euclidean distance between present values and desired values in the parameter-space represented by damping, frequency, and steady-state error. The output of the tuner is then scaled by a "reward factor" that is a function of the tuner effectiveness and consistency. If tuner effectiveness degrades, the self-monitoring mechanism diminishes the output of the tuner. If the tuner is operating with consistency, the self-monitoring mechanism increases the output of the tuner. Thus, the self-monitoring mechanism improves performance and robustness of the tuner by scaling the output to produce faster slew rates when the tuner is performing well and by reducing tuner output when performance is poor.

Claims

1. A method of feedback control, comprising the steps of:

connecting a feedback controller to a controlled system for receiving an input signal, providing a gain signal to said controlled system, and receiving a response signal from said controlled system;

connecting a tuner to said controller for comparing said input signal with said system response signal and for applying incremental gain value adjustments to said controller;

converting selected parameters of said compared signals to fuzzy logic parameters;

applying if-then rules to said fuzzy logic parameters for computing said incremental gain value adjustments;

measuring effectiveness and consistency of previously applied gain value adjustments;

scaling said computed incremental gain value adjustments based on said measured effectiveness and consistency; and

applying said scaled incremental gain value adjustments to said controller.

2. The method of claim 1, wherein the step of converting selected parameters further comprises the steps of:

computing parameters of system damping, frequency, and steady-state error; and

transforming said parameters of system damping, frequency, and steady-state error to corresponding fuzzy membership functions.

3. The method of claim 2, wherein the step of measuring effectiveness and consistency further comprises the steps of:

defining a system state-space having coordinates of damping, frequency, and steady-state error;

mapping said computed parameters of system damping, frequency, and steady-state error into said state-space;

defining an equilibrium point in said state-space comprising desired values of system damping, frequency, and steady-state error;

measuring a distance between said mapped parameters and said equilibrium point in said state-space;

computing effectiveness as a difference between two successive measurements of said distance divided by a moving average of said differences; and

computing consistency as an inverse of variance of tuner effectiveness over three successive computations of effectiveness.

4. The method of claim 3, wherein the step of scaling said computed incremental gain value adjustments further comprises the steps of:

defining a reward factor as a function of effectiveness and consistency;

defining an effectiveness multiplier as a function of effectiveness; and

multiplying said computed incremental gain value adjustments by said reward factor and said effectiveness multiplier.

5. A method of adjusting gain of signals provided to a controlled system by a feedback controller, comprising the steps of:

comparing a system input signal with a response of the controlled system;

transforming system parameters of damping, frequency, and steady-state error to fuzzy logic membership functions;

computing gain value adjustments by applying if-then rules to said fuzzy logic membership functions;

measuring effectiveness and consistency of previously applied gain value adjustments;

scaling said computed gain value adjustments based on said measurements of effectiveness and consistency; and

applying said scaled gain value adjustments to the feedback controller.

6. The method of claim 5, wherein the step of measuring effectiveness and consistency further comprises the steps of:

defining a system state-space having coordinates of damping, frequency, and steady-state error;

mapping said system parameters of damping, frequency, and steady-state error into said state-space;

defining an equilibrium point in said state-space comprising desired values of system damping, frequency, and steady-state error;

measuring a Euclidian distance between said mapped parameters and said equilibrium point in said state-space;

computing effectiveness as a difference between two successive measurements of said Euclidian distance divided by a moving average of said differences; and

computing consistency as an inverse of variance of effectiveness over three successive computations of effectiveness.

7. The method of claim 6, wherein the step of scaling said gain value adjustments further comprises the steps of:

computing a reward factor as a function of effectiveness and consistency;

computing an effectiveness multiplier as a function of effectiveness; and

multiplying said computed gain value adjustments by said reward factor and said effectiveness multiplier.

8. A tuner for applying incremental gain value adjustments to a proportional, integral, and derivative controller, comprising:

means for comparing an input signal with a response of a system controlled by the controller;

means for transforming system parameters of damping, frequency, and steady-state error to fuzzy membership functions;

means for computing the incremental gain value adjustments by applying if-then rules to said fuzzy membership functions;

means for measuring effectiveness and consistency of previously applied incremental gain value adjustments;

means for scaling said computed incremental gain value adjustments based on said measurements of effectiveness and consistency; and

means for applying said scaled incremental gain value adjustments to said controller.

9. The tuner of claim 8, wherein the means for measuring effectiveness and consistency further comprises:

a system state-space defined with coordinates of damping, frequency, and steady-state error;

system parameters of damping, frequency, and steady-state error mapped into said state-space;

an equilibrium point comprising desired values of system damping, frequency, and steady-state error defined in said state-space;

a Euclidian distance measured between said mapped parameters and said equilibrium point in said state-space;

means for computing effectiveness as a difference between two successive measurements of said Euclidian distance divided by a moving average of said differences; and

means for computing consistency as an inverse of variance of effectiveness over three successive computations of effectiveness.

10. The tuner of claim 9, wherein the means for scaling said gain value adjustments further comprises:

means for computing a reward factor as a function of effectiveness and consistency;

means for computing an effectiveness multiplier as a function of effectiveness; and

means for multiplying said computed gain value adjustments by said reward factor and said effectiveness multiplier.