JPH07271421A - Learning type numerical controller - Google Patents

Abstract

PURPOSE:To provide high-accuracy learning effects by executing the learning of deviation between a commanded value and a present value while limiting it to a prescribed frequency band. CONSTITUTION:Concerning position deviation data SPepsilon composed of the difference between shape command data SPref and machine position data SPc, a shape deviation processing part 11 executes the filter processing of the position deviation data and transmits the result to a position commanded value generation part 2. Concerning the position deviation including a desired frequency component because of the filter processing, the position commanded value generation part 2 corrects the commanded value. This processing is continued until the position deviation data SPepsilon satisfy fixed conditions.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a learning-type numerical control device that learns a tracking error that occurs when controlling a machine tool or the like and performs numerical control based on the learning result, and more particularly to a learning mode of this device.

[0002]

2. Description of the Related Art In order to improve the machining accuracy of a numerically controlled machine tool, there is a method of increasing the position loop gain and the velocity loop gain to improve the sensitivity of the control system to reduce the tracking error. However, if the position loop gain and the velocity loop gain are excessively increased, the control becomes unstable, so that there is a limit in reducing the tracking error of the control system. In addition to the control system, the motor, ball screw,
There is an inherent error in the servo system including the load, etc., and it is difficult for the control system to compensate for it. Therefore, there is a limit in reducing the shape error in the entire servo system including the control system. Therefore, learning control has been performed as a method for further reducing the shape error.

FIG. 4 is a block diagram showing an example of processing in a conventional learning control device. SW1 and SW2 are switches for switching between the normal processing mode and the learning mode, and the switching is performed online by a command from the end determination unit 10 to end the learning mode. The a side is the normal mode, the b side is the learning mode, and SW1 and SW2 are interlocked. In the normal mode, the position command value generator 2 generates a position command value at regular intervals, and servo system control and mechanical operation are performed by known techniques. Depending on the learning mode,
First, the position command value generation unit 2 receives the shape command value data from the shape command value generation unit 1 and generates a position command value for every constant period for the control system including the servo control unit 4. Position control is performed in the servo control unit 4 by a known technique, and a current command is sent to a mechanical system drive unit 6 including a motor and a feed shaft to operate the mechanical system drive unit 6. The machine position detection unit 7 detects the current position of the mechanical system drive unit 6 at regular intervals and sends the machine position data SP to the time delay correction unit 8.
There is a time lag between the transmission of the shape command value data SPref from the shape command value generation unit 1 to the position command value generation unit 2 and the detection of the machine position data SP by the machine position detection unit 7. Therefore, the machine position data SP detected by the machine position detecting unit 7 is subjected to the time delay correcting process on the shape command data SPref by the time delay correcting unit 8,
At the addition point 9, the position deviation SPε from SPref is calculated. The command values handled by the shape command value generator 1 and the position command value generator 2 are point group data, and point group data is also created for the position deviation SPε. This is added to the position command value point group data in the position command value generation unit 2 to correct the position command value. The learning control is completed once by the series of operations described above. The learning control is continued until the position deviation SPε is satisfied by the process in the end determination unit 10, and the corrected position command value is used in the position command value generation unit 2 in the second and subsequent learning controls. When the learning control ends, SW1 and SW2 are switched to the normal mode, and in the normal mode, the position command value at the end of the learning control is used.

[0004]

FIG. 5 shows an example of the positional deviation with respect to time. In the conventional learning control device, the position deviation is directly added to the position command value to correct the position command value, so that the corrected position command value includes the high frequency component of the position deviation. It was Here, what differentiated the position command value with respect to time is called a velocity component of the position command value, and what differentiated the speed component of the position command value with respect to time is called an acceleration component of the position command value. FIG. 6 shows an example of the acceleration component of the position command value before the position command value correction by learning control, and FIG. 7 shows an example of the acceleration component of the position command value after the position command value correction. Even if the position command value before the learning control is the acceleration component that does not include the high frequency component as shown in FIG. 6, if the position deviation including the high frequency component as shown in FIG. 5 is learned, the acceleration including the high frequency component as shown in FIG. Become an ingredient. In the normal control system, a torque command proportional to the acceleration component of the position command value is generated. Therefore,
When performing control with the position command value after learning control, a torque command proportional to the acceleration component in FIG. 7 is generated. Since it is difficult for the control system to follow a signal having a frequency higher than the frequency band determined by the transfer function, it is difficult to follow the torque command including a high frequency component. It is difficult to follow the control system, that is, the servo system including the mechanical system is also difficult to follow. As a result, the reduction in position deviation is smaller than before the position command value was corrected. Become. As described above, in the conventional learning control device, when the learning control is performed for the control system that generates the position deviation including the high frequency component, the high frequency component of the position deviation is also learned, so that the learning accuracy is improved by the learning control. There was a problem that it was difficult.

As described above, the cutoff frequency is changed every learning control, and the learning control is performed in order from the low frequency component of the position deviation, and finally all the frequency components of the low frequency from the frequency band of the control system are learned. Then, the learning control can be completed within the range that the control system can follow.

The problem to be solved by the present invention is that when learning control is performed on a servo system which generates a position deviation including a high frequency component, the learning deviation includes the high frequency component of the position deviation so that the position deviation is reduced by the learning control. For the problem that it does not reduce, filtering for position deviation,
It is to reduce the position deviation by performing learning control of only a desired frequency component and achieve improvement of machining accuracy.

[0007]

Means for Solving the Problems In a learning type numerical control device according to the present invention for solving the above-mentioned problems,
A first filter processing unit that cuts a frequency component having a frequency higher than the highest frequency that can be tracked by the servo system controlled by the learning type control device, and a filter process by a predetermined cutoff frequency in a frequency band that the servo system can track. And a second filter processing means for performing. Further, frequency analyzing means for performing frequency analysis of the deviation, minimum frequency selecting means for extracting a frequency component having a predetermined gain or more of the deviation based on the analysis of the frequency analyzing means, and selecting the lowest frequency among them, The cutoff frequency determining means determines the cutoff frequency of the second filter processing means based on the frequency selected by the lowest frequency selecting means.

[0008]

According to the present invention, among the frequency components of the deviation, a high frequency component is cut from the frequency band which the servo system can follow, and the frequency component which can be followed by the servo system is narrowed down to a predetermined frequency component. You can do different learning. Further, it is possible to perform learning by focusing on frequency components having a deviation level of a certain value or more in a frequency band in which the servo system can follow.

[0009]

EXAMPLE FIG. 8 shows an example of frequency analysis of the positional deviation of FIG. As shown in FIG. 8, the position deviation in FIG. 5 is composed of many frequency components. In the conventional learning control method, learning is performed for all these frequency components. If the frequency band of the servo system is set to fbHz or less, it is difficult for the servo system to follow the position deviation of higher frequency than fbHz as described above. However, in other words, f
The servo system can follow the position deviation of bHz or less. If tracking of the servo system is possible, the position deviation can be reduced by learning control. Therefore, only the position deviation having a frequency lower than fbHz should be learned. FIG. 9 shows an example of the positional deviation obtained by applying a low-pass filter having a cutoff frequency fbHz to the positional deviation shown in FIG. The position deviation in FIG. 9 does not include a higher frequency component than fbHz, but fbHz
It contains many frequency components below. Since each of these frequency components can be tracked by the servo system, learning is possible, but it is difficult to learn all the frequency components at once. Therefore, it suffices that the position deviation is again low-pass filtered and only desired frequencies are extracted and learned. For example, it is assumed that the (fb / 4) Hz low-pass filter is further applied to the acceleration component obtained by cutting the high-frequency component from the frequency band of the servo system by applying the fb Hz low-pass filter. FIG. 10 shows an example of the position deviation obtained by applying a low-pass filter of (fb / 4) Hz. Since the low-pass filter cuts off the acceleration component having a frequency higher than (fb / 4) Hz, when the position command value is corrected by the position deviation due to this acceleration component, (fb /
4) Learn the frequency component of the position deviation that is lower than Hz,
The position deviation caused by the frequency component is reduced. After the learning control is completed once by the above procedure, in the second time, similarly to the first time, the low-pass filter of fbHz is applied to the position deviation to cut the high frequency component from the frequency band, and then the cutoff frequency is set to (fb / 3) Set to Hz and apply a low pass filter. Then, I learned the last time (f
It is possible to perform learning on frequency components lower than (fb / 3) Hz, including a portion of b / 4) Hz or less.
By repeating the above operation to increase the cutoff frequency for each learning control and repeating the learning control until the cutoff frequency becomes fbHz, the position caused by the frequency component within the frequency band of the servo system is caused. All deviations can be reduced. Preferred embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a process of the learning control device in the present invention, and FIG. 2 is a diagram showing a process of the shape deviation processing unit 11 in FIG. Since the functions and processes of the components indicated by the same numbers as those in FIG. 4 showing the prior art are the same, the description thereof will be omitted, and only the shape deviation processing unit 11 of the present invention will be described below. The processing performed on the position deviation data by the shape deviation processing unit 11 is different from the other portion that performs the processing in real time on the data transmitted at regular intervals, and the entire position deviation data is processed as a data string. Put it on the table and process the entire data table. Position deviation data SPε created from the difference between the shape command value data SPref in the shape command value generation unit 1 and the machine position data SPc after time delay correction.
In the shape deviation command unit 11, first, the first filter processing unit 1
At 11, the first filtering process is performed. The cutoff frequency in the first filter processing is the first cutoff frequency determination unit 112.
Is determined by. The first cutoff frequency fc1 is determined by the frequency band of the servo system.

Generally, in the Bode diagram of the frequency response of the servo system, the servo system can theoretically follow up in a frequency range lower than the frequency of the point where the gain decreases by 3 dB. Since the frequency response of the servo system is determined from the transfer function, if the transfer function of the servo system can be determined, the frequency band of the servo system can also be obtained. The transfer function of the control system excluding the mechanical system can be determined by a known method based on the position loop gain, the velocity loop gain, etc., and is invariable unless each parameter of the control system changes. Therefore, the control system transfer function is set as a constant off-line and a constant value is used.

Now, the control system transfer function H (s) is calculated by a known method.

## EQU1 ## H (s) = (1 + b × s) / (1 + a × s) is obtained. However, a and b are constants determined by each parameter of the control system. In this case, the band frequency fb0 of the control system is

## EQU00002 ## fb0 = k / a. However, k is a constant. The band frequency fb0 in this case is a band frequency obtained from the transfer function of the control system, and is not the transfer function of the servo system including the mechanical system. In the mechanical system, there is a tracking delay of the machine. Therefore, when the transfer function of the servo system is considered, the time constant is considered to be smaller than that of the transfer function of the control system. Therefore, the band frequency fb of the servo system is smaller than the band frequency fb0 of the control system. Theoretically mechanical system drive unit 6 and machine position detection unit 7
Since it is considered that the operation of is constant every time, the reduction rate α of the band frequency fb of the servo system with respect to the band frequency fb0 of the control system is constant. Therefore, the band frequency f of the servo system
b is

## EQU3 ## This is obtained from fb = α × fb0, and this fb is determined as the first cutoff frequency fc1. The first cutoff frequency determination unit 112 receives the data of α from the parameter table unit 118 and determines the first cutoff frequency fc1 by the method described above.

Further, generally, when filtering is performed, a phase delay occurs. The phase delay can also be obtained from the phase angle at the band frequency in the Bode diagram of the frequency response of the servo system. In the first phase delay processing unit 113, the first cutoff frequency f from the first cutoff frequency determination unit 112.
c1 and servo system transfer function data are received, the phase delay is calculated by a known method, and the first filter processing unit 11
The data shift is performed on the position deviation data sent from 1. By the above method, the position deviation data SP
ε can cut frequency components higher than the frequency band of the servo system.

The position deviation data generated by the first phase delay processing unit 113 is sent to the second filter processing unit 114 and at the same time sent to the frequency analysis unit 116. In the frequency analysis unit 116, a Fourier transform is performed on the data string by a known technique to analyze each frequency component. On the other hand, in the parameter table unit 118, the threshold value TL for the frequency component analyzed by the frequency analysis unit 116, the threshold value TL2 serving as the positional deviation reduction determination reference, and the cutoff frequency shift amount fd are set as parameters. 2 to the cutoff frequency determining unit 117. FIG. 3 is a flowchart showing a series of processes in the second cutoff frequency determination unit 117.
Shown in. The second cutoff frequency determination unit 117 first reads from the second cutoff frequency storage unit 119 the second cutoff frequency fc2 ′ determined in the previous learning control. Then, the gain data at the second cutoff frequency fc2 ′ of the previous time is detected from the data subjected to the frequency analysis this time. When the gain data is TL2 or more, the previous second cutoff frequency fc
The second cutoff frequency fc2 of this time is determined assuming that 2 ′ is used as it is also this time. TL2 is another threshold value for determining the reduction of the position deviation due to the desired frequency component by the previous learning control, and is set to a value of about 30 to 50% with respect to TL. If it is confirmed by the determination that the gain is reduced in the desired frequency component, the second
In the cutoff frequency determining unit 117, for the frequency component data transmitted from the frequency analyzing unit 116, for the lowest frequency component among the frequency components having the gain exceeding the threshold TL transmitted from the parameter table unit 118, By adding the cutoff frequency shift amount fd to the second cutoff frequency f
Determine as c2. The cutoff frequency shift amount fd is an amount added to the frequency in order to surely cut a desired frequency component, and is an amount of about 0.5 to 5 Hz.
Further, the TL, TL2, and fd can be set as parameters. The second cutoff frequency fc2 determined by the second cutoff frequency determination unit 117 is sent to the second filter processing unit 114, the second phase delay processing unit, and the second cutoff frequency storage unit 119. The second cutoff frequency storage unit 119 stores only the cutoff frequency of the previous time, and the data is updated every learning control. The second filter processing unit 114 performs the second filter processing, and sends the filtered positional deviation data to the second phase delay processing unit 115.
The second phase delay processing unit 115 calculates a phase delay based on the second cutoff frequency fc2 determined by the second cutoff frequency determination unit 117 by the same method as described above, shifts the data, and performs the shape deviation processing. It is sent to the position command value generation unit 2 as subsequent data SPε '. The above is the description of the series of processes in the shape deviation processing unit 11. In the present invention, frequency components that the servo system cannot respond are cut in each learning control to prevent an increase in position deviation due to difficulty in tracking the servo system. Further, for frequency components that can respond to the servo system, frequency components above a certain level. Out of
Learning control is performed in order from the low frequency component. Therefore, while the first cutoff frequency is constant, the second cutoff frequency changes every learning control. Also, theoretically, each time learning control reduces the position deviation due to the frequency in order from the low frequency component, but if the reduction is actually insufficient, perform filtering again with the same cutoff frequency. Since the learning control is performed again for the desired frequency component, the frequency component can be surely learned.

In this embodiment, the first cutoff frequency and the second cutoff frequency are automatically determined, but the method of determining each cutoff frequency is not limited to this. In addition, although a method of performing frequency analysis on the data after the first filter processing is used when determining the second cutoff frequency, the method is not limited to this, and many variations can be considered. Furthermore, although the filters are low-pass filters, the calculation method for the cutoff frequency and the phase delay is as described above, and the frequency analysis method is the Fourier transform, the method is not limited to this and many other methods are possible. Many modification examples of the method of setting the servo system transfer function and the parameter table are possible.

[0015]

As described above, according to the learning control device of the present invention, since it is possible to perform learning on a desired frequency component in the frequency band of the servo system, a servo that generates a deviation including a high frequency component. Even in the system, there is an effect that the deviation can be reduced by the learning control and the machining accuracy can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing an example of learning control processing according to the present invention.

FIG. 2 is a block configuration diagram showing a detailed example of a shape deviation processing unit 11 which is a main part of FIG.

FIG. 3 is a flowchart illustrating an operation of a second cutoff frequency determination unit 117.

FIG. 4 is a block diagram showing an example of conventional learning control processing.

FIG. 5 is a diagram showing an example of position deviation.

FIG. 6 is a diagram showing an example of an acceleration component of a position command value before position command value correction by learning control.

FIG. 7 is a diagram showing an example of an acceleration component of a position command value after correction of the position command value by learning control.

FIG. 8 is a diagram showing an example of a result of frequency analysis of the positional deviation shown in FIG.

9 is a diagram showing an example of position deviation obtained by applying an fbHZ low-pass filter to the position deviation shown in FIG.

FIG. 10 is (fb / 4) HZ for the position deviation of FIG.
FIG. 6 is a diagram showing an example of position deviation that has been subjected to the low-pass filter of FIG.

[Explanation of symbols]

 1 Shape command value generation unit 2 Position command value generation unit 3 Addition point 4 Servo control unit 5 Amplifier 6 Mechanical system drive unit 7 Machine position detection unit 8 Time delay correction unit 9 Addition point 10 End determination unit 11 Shape deviation processing unit 111 1st filter processing part 112 1st cutoff frequency determination part 113 1st phase delay processing part 114 2nd filter processing part 115 2nd phase delay processing part 116 Frequency analysis part 117 2nd cutoff frequency determination part 118 Parameter table part 119th 2 Cutoff frequency storage

Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI Technical display location G05D 3/12 305 V

Claims (2)

[Claims] 1. A learning-type numerical controller for learning a deviation between a control command value and an actual present value and correcting the control command value based on the learning result. The servo system controlled by the learning-type controller. Has a first filter processing means for cutting frequency components higher than the highest frequency that can be followed, and a second filter processing means for performing a filtering process with a predetermined cutoff frequency in a frequency band which the servo system can follow. A learning-type numerical control device, wherein the deviation is processed by the first and second filter processing means, and learning is performed based on the processed deviation. 2. The learning type numerical control device according to claim 1, further comprising: a frequency analysis means for performing a frequency analysis of the deviation.
Based on the analysis of the frequency analyzing means, a frequency component having a predetermined gain or more of the deviation is extracted, and the lowest frequency selecting means for selecting the lowest frequency among them, and the second frequency selecting means based on the frequency selected by the lowest frequency selecting means. A learning-type numerical control device comprising cutoff frequency determining means for determining a cutoff frequency of a filter processing means.