JPH03290702A - Learning control system - Google Patents

Abstract

PURPOSE:To obtain the step answer of a control subject with no direct measurement by sampling the ramp answer of the control subject and using the value obtained by dividing the differential value of the ramp answer by the slope of a ramp command as the step answer. CONSTITUTION:A step answer Hk is obtained as Hk = (Rk - Rk-1)/D based on the sample value Rk of the ramp answer obtained by sampling the step answer of a control subject. In this equation D shows the slope of a sampling interval set in a steady state of a ramp command or the ramp answer. Based on this step answer, a future control deviation is estimated and the control input is decided so that the estimated control deviation is minimized. Thus a follow-up control system having the high accuracy is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a control system for machine tools, robots, etc. that perform repetitive operations.

[Conventional technology]

As a design method for a learning control system for repeated target values,
There is a method proposed in Japanese Unexamined Patent Publication No. 1-237701, which was previously filed by the present applicant! This method repeats the operation for the same target value, predicts the future deviation based on the past deviation, and corrects the control input so that the value becomes the minimum. Since the value and the output match (1+TS), highly accurate tracking operation is achieved.

[Problem to be solved by the invention]

In the above method, the step response of the controlled object is required to predict the deviation, but measuring the step response places a burden on the mechanical system, for example in the case of an NC machine tool.
I don't like it very much.

Therefore, an object of the present invention is to obtain the step response of a controlled object without directly measuring it.

[Means to solve the problem]

In order to solve the above problem, the first invention of the present application repeats trials to match the output of the controlled object to a target value that matches the same pattern, and uses the step response of the controlled object in each trial. In a control system that predicts the future control deviation based on the predicted value and determines the control input so that the predicted value is minimized, the ramp response of the controlled object is sampled, and the difference value divided by the slope of the ramp command is calculated as a step value. It is characterized by being used as a response.

The second invention of the present application repeats trials to match the output of the controlled object to a target value that repeats the same pattern, and predicts future control deviation using the step response of the controlled object during each trial. In a control system that determines the control input so that the predicted value is minimized, the time from the manual power time of the lamp command until the output of the controlled object starts to rise is regarded as the dead time td, and the steady-state deviation value E
The time constant T is calculated from the slope D of the ramp command by T=E/D, and the step response is determined by simulation using the above transfer function.

[For production]

In the present invention, a ramp response that does not place much burden on the mechanical system is measured, and a step response is calculated from the result.

〔Example〕

Hereinafter, a specific example will be described in which the method of the first invention of JP-A-1-237701 is adopted as the learning control law and the present invention is used as the step response calculation method.
This will be explained with reference to the diagram.

The 11th concrete embodiment of the first invention of the present application! Shown in l.

1 in the figure is a command generator, and the current time 11. : Generate the target value r (i) at . 2 is a subtractor, and the deviation e
It is used to find and memorize (i). 3 is constant Q+
, ql..., q,, Q, g++...
+g*-+ memory, 4 is the deviation e(i) for the current time 1 and the past one cycle (j=i, i-1, ......,
i'+1. i') memory. However, i'
=i-1.

Further, 5 is the incremental correction amount a(j) from the time one sampling before to the time N-1 times before N=i-1°1-2.・・・
..., i-N+1), and 6 is the memory for the control human power u(j)(
j=i-1, i-2,..., i'+1゜i')
memory.

In addition, 7 is an arithmetic unit, a(0=Σqie(i'10K) 0Q (e (i)
-e (i')) -ΣgI a(i-n) ・・
The current incremental correction amount a (i) is calculated by the calculation (1).

8 is an integrator that calculates the current correction amount Σa('').

J=1゜9 is an adder, and the control human power u at the time one cycle before
(io> and the current correction amount Σa (j) are added: 1
0, the current control human power u(1) is output. At the start of control, u (i-)=O, a (i,)=0.

10 and 11 are samplers that close at the sampling period T, and 12 is a hold circuit.

13 is a controlled object, the input is u (t), and the controlled quantity which is the output is x (t).

In the control system, 2 to 12 are parts usually called controllers, which can be easily realized using general-purpose digital circuits or microcomputers. Moreover, it does not matter if some kind of control system (compensator, etc.) is already included in the 13 objects to be controlled.

Here, the constants Q L Q, g- in equation (1) are given by the following equation. Next, in the second method, the transfer function of the controlled object 13 is calculated.

The time from the 1+TS time until the output X of the controlled object starts to rise is regarded as the dead time td, and the time constant T is calculated from the steady-state deviation value E and the slope D of the ramp command as T=E/D.

Hk in the above equation is the sample value of the step response of the controlled object 13 (Fig. JJ2), but in reality, the ramp response of the slope D is sampled. I) We will show you two ways to convert it into a step response.

First, in the first method, the step response Hk is calculated from the sample value Rk of the sampled lamp response using the following formula. Here, D is
It is the slope of the steady state sampling interval of the lamp command or lamp response.

We will simulate the step response and find Hk.

〔Effect of the invention〕

As described above, according to the present invention, the ramp response of the controlled object is measured, information such as the dynamic characteristics of the controlled object is obtained without placing a burden on the mechanical system, and control input is performed using learning control based on the information. Therefore, it is possible to realize a highly accurate tracking control system.

[Brief explanation of the drawing]

FIG. 1 is an embodiment of the present invention, and FIGS. 2 and 3 are explanatory diagrams of the operation of the present invention. l...Command generator 2...Subtractor 3...Constant memory 4...Deviation memory for the past one cycle 5...Past incremental correction amount memory 6...Past control Input memory 7...Arithmetic unit 8...Integrator 9...Adder 10, 11...Sampler 12...Hold circuit 13...Controlled object Figure 2 Patent applicant Yasuyo Co., Ltd. Kiku, Representative of Denki Building Works
Ike Isao d

Claims (2)

[Claims] (1) Repeat trials to match the output of the controlled object to the target value that repeats the same pattern, and use the step response of the controlled object in each trial to predict future control deviation,
In a control system that determines a control input so that its predicted value is minimized, the learning is characterized in that the ramp response of the controlled object is sampled, and a value obtained by dividing the difference value by the slope of the ramp command is used as the step response. control method. (2) Repeat trials to match the output of the controlled object to the target value that repeats the same pattern, and use the step response of the controlled object in each trial to predict future control deviation;
In a control system that determines the control input so that the predicted value is minimized, the ramp response of the controlled object is sampled, and the transfer function of the controlled object is assumed to be (e^-^t^d^■)/(1+TS). The time from the input time of the ramp command until the output of the controlled object starts to rise is regarded as the dead time td, and the time constant T is calculated from the steady-state deviation value E and the slope D of the ramp command as T=E/
A learning control method characterized in that the step response is calculated by D and is obtained by simulation using the transfer function.