TW201220010A - Parameter learning controller in a machine device and learning method thereof - Google Patents

Abstract

A parameter learning method of a CNC machine device is designed to modify parameters of the CNC machine device according to fast, accurate, and stable preferences. Weightings of these preferences can be set by user's requirement. This learning method will attend the objective including saving working time, improving workpieces accuracy, or enhancing processing stability.

Description

201220010 VI. Description of the Invention: [Technical Field] [0001] The present invention relates to a numerical controller in a CNC mechanical device, and more particularly to a controller of a CNC mechanical device having a parameter learning function and parameters thereof Learning control method; according to the user's preference for processing characteristics, through the parameter learning method, the parameters of the mechanical device can be adjusted to shorten the processing time, improve the processing precision, and improve the processing stability of the mechanical device. [Prior Art] 〇 [00〇2] First, please refer to Fig. 1, which is a numerical value of a conventional CNC mechanism. As shown in FIG. 1, the digital controller module 10 is composed of a road planning unit 12, a motion planning unit 13, and an interpolation unit 14, and the path planning unit 12 is a user-edited program. Interpreting and planning a processing route; then, the motion planning unit 13 further predicts the motion characteristics of the processing path, such as the speed acceleration, according to the parameter 18 provided by the digital control module 1; After the interpolation unit 7014 performs interpolation processing on the single-section data with motion characteristics of the motion planning, the command is sent to the driver T5; then, the driver 15 receives the interpolated command 彳, and then retransmits the signal rotation. The motor 16 is controlled and, in addition, the sensed motor 16 is positionally fed back to the actuator 15 by position sensing elements 17, such as motor encoders or optical scales.闺 During the operation of the above-mentioned CNC mechanical device, when setting the parameters 18 of the digital control module 10, 'often due to the difference between the rigidity of the machine, the characteristics of the motor and the various mechanisms, it is necessary to repeat the test by experience to set a set. More suitable 099137664 Form No. A0101 Page 3 / Total 28 Page 0992065649-0 201220010 [0004] [0005] [0007] [0007] 099137664 When the parameter is 18. This can cause a lot of inconveniences. For example, setting parameters requires experienced tuning personnel, and it takes a lot of time to repeat testing and verification. Due to the use of prior art in CNC machinery, a set of parameters is used to apply to all machining processes. Since the user does not know the parameters deeply, and the processing status often changes due to different situations, the user often cannot determine whether the current parameters are suitable for the processing file that is being prepared for processing, whether there is too long processing time and machine jitter. Such problems; in the face of these problems, users are usually unable to adjust the parameters themselves to solve the problem of 'the need for the assistance of the mechanical plant personnel to have a way' causing considerable trouble and inconvenience to the user. SUMMARY OF THE INVENTION In order to solve the problems and disadvantages of the prior art, the main object of the present invention is to provide a numerical controller having a parameter learning function in a CNC mechanical device, and the CNC mechanical mechanism is provided by a parameter learning function of the numerical controller. The device can be more convenient in tuning and parameter setting. Another main object of the present invention is to provide a numerical controller with a parameter learning function in a CNC mechanical device, so that the mechanical factory personnel can turn on the parameter learning function and execute the standard test program when adjusting the CNC mechanical device. The digital controller records the fast and quasi-stable performance indicators, and automatically updates the parameters of the mechanical device through the parameter learning algorithm, so that the CNC mechanical device can achieve the processing effect more accurately in the tuning and parameter setting. Still another main object of the present invention is to provide a numerical controller having a parameter learning function in a CNC mechanical device, so that the machine factory personnel can turn on the parameter learning function when the CNC mechanical device is adjusted and the form number A0101 is 4 Page / Total 28 pages 0992065649-0 201220010 [0008] 0009 [0009] [0010]

G

[0011] performing a standard test program, so that the digital controller records fast and quasi-stable performance indicators, and automatically updates the parameters of the mechanical device through the parameter learning algorithm, so that the user can adjust the preference according to the processing characteristics through the parameter learning method. The parameters of the mechanical device can reduce the processing time, improve the machining accuracy, and improve the processing stability of the mechanical device. · The method of parameter learning of the present invention sets the total performance index according to the weight of the user's fast, accurate, and stable performance indicators, and pursues the goal of pursuing the best overall performance index as a parameter. Therefore, the main advantages of the present invention include: 1. When adjusting the machine, using the parameter learning function, it can ensure that the adjusted parameters will improve the overall performance index of the CNC mechanical device to a fairly good level, and can avoid being adjusted due to lack of experience. Occurs when the appropriate parameters or tuning time is too long. 2. In the past, when the user encountered an unsatisfactory processing result, the parameters could only be changed and the processing was repeated, which was very difficult to solve. Through the parameter learning function, you can directly input the weight of the index preference such as fast quasi-stable, without having to change the parameters yourself, you can learn the parameters well, improve the convenience and solve the problem timeliness. According to the above objects and advantages, the present invention firstly provides a controller having a parameter learning function, comprising: a path planning unit for receiving a processing program and planning a processing path message according to a processing program; a motion planning unit, An input terminal is connected to the path planning unit and receives the processing path message, and the other input end is connected with a digital control parameter, and the processing path information is planned according to the processing path information and the digital control parameter. 099137664 Form No. A0101 Page 5 of 28 0992065649-0 201220010 The motion message; an interpolation unit is connected with the motion planning unit, and the motion information output by the motion planning unit is interpolated, and then a control command is output to the motor driver; The calculation unit has an input end connected to a position sensing component, and receives a feedback signal input by the position sensing component, and calculates a fast performance indicator, a quasi-performance indicator and a stable performance indicator according to the feedback signal, to output a fast Quasi-stable indicator message; a learning function unit with its first input terminal Set by the wearer fast metastable weight message, a second input terminal of fast metastable index calculation unit connected to and receiving fast quasi-stationary indicator message, the newly calculated digital control parameters, and outputs it to stroke the planning unit. [0012] The present invention further provides a parameter learning method of the controller, comprising: providing a path planning unit for receiving a processing program, and planning a processing path message according to the processing program; providing a motion planning unit, an input end thereof Connected with the path planning unit and received the processing path message, and the other input end is connected with a digital control parameter, and the processing path message is output according to the processing path information and the digital control parameter to output a motion message; and an interpolation unit and a motion planning unit are provided. Connecting, and performing an interpolation operation on the motion message output by the motion planning unit, and then outputting a control command to the motor driver; providing a fast quasi-stable index calculation unit, one input end of which is connected to a position sensing component, and Receiving a feedback signal input by the position sensing component, and calculating a fast performance indicator, a quasi-performance indicator and a stable performance indicator according to the feedback signal to output a fast quasi-stable indicator message; providing a learning function unit, the first input end thereof Receiving a fast quasi-weighted message set by the user, the second input thereof The quasi-stable index calculation unit connects and receives the fast quasi-stable index message, and recalculates the new digit control parameter 099137664 Form number AOiOl Page 6/28 page 0992065649-0 201220010 [0013] [0014] Ο ο [0015] and outputs To the motion planning unit. [Embodiment] Since the present invention discloses a digital control vibration having a parameter learning function and a control method thereof, the digital control device can introduce a parameter learning function control mechanism, and therefore, in the following description, the digital control will be described in detail. A method by which a device implements a parameter learning function. First, please refer to Figure 2, which is a schematic diagram of the CNC mechanical control file (NC Fi le) with learning function. As shown in Figure 2, G90 G01 Z-5. F2000 represents a general machining single section 'where G90 is the absolute command, G01 is the linear cutting command, Z is the single block end point, and F2000 is the set cutting it. G7. 7 PIQnRnKn represents the start parameter learning function, in which Pn sets the n-th set of high-speed and high-precision parameters, and QnRnKn represents the set fast (Q), quasi (E·), and stable (IL) weights. For example: P1Q1R1K1 represents the learning of the first group of high-speed and high-precision parameters and the setting is fast, accurate and stable. The weight is 1:1:1; and G01. Χ0· ΥΙΟ. /Χ30· Y20. represents the processing order with learning function. Section, where X, Y arguments set the single-node end point; G5. 5 represents the end (or close) parameter learning function; G01.xi2.Y5. represents the general processing block, where X ' Y argument list End point coordinates. In the invention, 'when the G 5. 7 code is recognized in the numerical control slot of the processing program edited by the user', the program for starting the learning function is turned on; when the G5. 5 code is recognized, the program for ending the learning function is indicated. . When the processing program executes the program that starts the learning function, it will record the performance indicators such as fast, accurate, and stable. After the parameter learning algorithm, a set of parameters satisfying the best total performance indicator "P" will be obtained. Obviously, the learning function of the present invention is 099137664 Form No. A0101 Page 7 of 28 0992065649-0 201220010 The learning function or the end learning function is activated at any time of the processing program; the present invention is not limited thereto. [0016] Next, please refer to FIG. 3, which is a functional block diagram of the numerical controller of the present invention. As shown in FIG. 3, the numerical controller of the present invention comprises a digital control module 10, a position sensing component 17, a fast quasi-stable index computing unit 20 and a learning function unit 30; wherein the digital control module 10 is composed of a path planning unit 12. The motion planning unit 13 and the interpolation unit 14 are composed. The position sensing component 17 is coupled to the motor 16 for reading the feedback signal and transmitting the feedback signal to the driver and the digital control device; wherein the position sensing component 17 can be a motor encoder 'optical scale or an acceleration gauge; This feedback signal can include the position, velocity and acceleration of the motor 16, and the like. Then, the fast quasi-stationary computing unit 20 is connected to the position sensing component 17, and can be based on the feedback signal sent from the position sensing component 17, and the feedback signal is calculated by the fast quasi-stationary computing unit 20, respectively. The fast indicator message 25, the quasi-indicator message 26, and the stable indicator message 27 are outputted to the learning function unit 30 for learning functions. [0017] The learning function unit 30 is connected to the fast quasi-stationary computing unit 20 for providing an algorithm, and can calculate the fast quasi-stable index signal according to the fast quasi-stationary weight information 31 set by the user and the fast quasi-stationary computing unit 20. 28. After the algorithm of the learning function unit 30 is calculated, the corrected digital control device parameter 18 is obtained for processing after the CNC mechanism. [0018] Next, please refer to FIG. 4, which is a block diagram of the fast quasi-stable index calculation unit of the present invention. As shown in FIG. 4, the fast quasi-stable index calculation unit 20 is connected to the feedback signal 21, which may be the feedback position of the motor or the optical ruler and the acceleration value transmitted by the acceleration gauge; 099137664 Form No. A0101 Page 8 of 28 0992065649-0 201220010 22 When the machining program is executed to start the parameter learning function, the time required for each block processing will be recorded, so the fast index of the present invention (Fast In_ dex) Can be set to, a is a fast indicator constant. The quasi-indicator calculation sub-unit 2 3 will

The aF servo command position is subtracted from the feedback position signal 2丨, and the relative trajectory error is calculated. Therefore, the quasi index (precisi〇n Index) of the present invention can be set to be fast = αρ

Heart 仏 * number. The stability index calculation sub-unit 24 is configured to feedback the acceleration signal 2 to measure the acceleration change value of the bed or the workpiece during processing, or may also pass the difference from the feedback bit signal 21, and (4) estimate the acceleration change value, especially The purpose of defining this stability index is because the change of acceleration will cause the machine to be irregularly red, so the invention is for the acceleration time signal 'using wavelet transform to convert UaVeiet Transformation) or fast Fourier transform ( Fast p . f〇urior Frequency Transformation) Converts the time signal into a frequency signal, so it can monitor the intensity of the acceleration signal at the frequency at which the machine is prone to resonance. 099137664 (intensity) Form No. A0101 Therefore, the stability index of the present invention (Stability Index) ) ^ 9 / only 'total 28 pages 0992065649-0 201220010 can be set to, where is the fast indicator constant, for monitoring the intensity of the as h frequency. The fast indicator calculation sub-unit 22, the quasi-indicator calculation sub-unit 23 and the steady-state calculation sub-unit 24 calculate the corresponding fast indicator message 25, the quasi-indicator message 26 and the stability indicator message 27, and the three are quickly stabilized. Indicator message 28 is provided for use by the learning function module. [0019] Obviously, the above-mentioned fast index calculation sub-unit, quasi-index calculation sub-unit and stable index calculation sub-unit are respectively calculated according to the feedback signal, and the fast performance index, the quasi-performance index and the stable performance index are respectively calculated by numbers. . In addition, it should be emphasized once again that the present invention is applicable to the calculation of only one or two performance indicators, and it is not limited to the calculation of the fast-steady three performance indicators in order to achieve the learning effect. In addition, since the stable performance index is related to the surface roughness or surface finish of the machined surface, the stable performance index can also be described by the surface roughness, and the present invention does not limit the manner in which the performance indicator is described. [0020] Please refer to FIG. 5 again, which is a block diagram of the learning function unit of the present invention. Referring to Fig. 5, the learning function unit 30 is formed by a series connection of the total performance index calculation sub-unit 32, the learning effect check sub-unit 33, the Jacobian matrix correction sub-unit 34, and the parameter correction sub-unit 35. 099137664 Form No. A0101 Page 10 / Total 28 Page 0992065649=0 201220010 The fast quasi-station weight message 31 in the figure is the fast quasi-stable performance index preference weight "wi" set by the user, which can be input when the control command is entered. And inputting to the learning function module by the control command; wherein the fast quasi-stable index message 28 is the fast quasi-stable performance index calculated by the fast quasi-stable index calculation unit 2前述. When the fast quasi-stable indicator message 28 is defined as the time, the total performance indicator sub-unit 32 including the fast quasi-stable indicator message "1" and the fast quasi-stable performance indicator preference weight "'" is obtained by "()). "P":

[0021] ❹

Eq(1) [0022] Substituting the current total performance indicator "P" into the learning effect test sub-unit 33, judging whether the total performance indicator "P" has improved, and if the good performance means that the previous learning is effective, the effective number of learning is increased. 1. The flag "L" is set to 丨, and the parameter correction result of the previous learning is stored; if it is not good, the previous learning is invalid, the learning invalid number is increased by 1, the flag "L" is set to 0, and the previous learning is abandoned. Parameter 099137664 Form No. A0101 Page 11 / Total 28 Page 0992065649-0 201220010 Corrected the result. The way to determine whether the total performance indicator "p" has improved is to compare the total performance indicator "P" calculated this time with the previous performance indicator "p-1"; for example When the total performance index P" is smaller than the total performance 4a "P 1" value (or convergence), the total performance indicator "P" is judged to be better; otherwise, the total performance indicator "P" is judged to be bad. . [0024] After verifying the learning effect, enter the jac〇bian matrix correction sub-unit 34. If the flag "L" is 1, the Jacobian matrix "κ" is modified; if the flag "L·" If it is 0, it means that the previous learning is invalid, then the Jacobian matrix "K" is not corrected. The correction method of the Jacobian matrix "κ" is as follows: Since each performance index "q" is a function of each parameter "χ, 1 j The amount of change is: [0025]

Eq( 2 ) [0026] Calculate the estimated performance indicator change using Eq(2) "099137664 Form No. A0101 Page 12 of 28 0992065649-0 201220010 [0027]

❹ [0028] [0029] Ο

Eq( 3 ) defines the Jacobian matrix "K

Eq( 4 ) [0030] Therefore, "K" can be used to calculate the estimated change in performance indicator "q/ 099137664 Form No. A0101 Page 13 / 28 Page 0992065649-0 201220010, the initial value of "K" is numerical control The device is estimated in advance; the actual performance indicator change amount "\" is equal to the index of the current measurement minus the previous measurement. Since the actual measurement's performance indicator change is different from the estimated performance indicator change, the actual performance indicator change amount is subtracted from the estimated performance indicator change, and the performance indicator change amount can be defined.

[0031] V % Ml ei a Hi =: _e«_ ^ ... Μ a ··· Ak. 9

^:^···Λ1:„” -1 X

Eq( 5 ) 099137664 [0032] The correction amount of each element of the Jacobian matrix obtained by Eq(5) is updated with the Jacobian matrix for the next use: Form No. A0101 Page 14 of 28 k and 0992065649-0 201220010 (K)r+l = (K)r + ΑΚ Ο

Eq(6) [0034] After correcting the Jacobian matrix, the parameter correction sub-unit 35 is entered. If the flag "L" is 0, it means that the target improvement amount "Ρ^" of the previous total performance indicator is too large, resulting in invalid learning. First, reduce the target improvement amount. The learning goal is to pursue the extreme value of the total performance indicator "P". When the direction of the parameter modulation is parallel to the gradient of the total performance index, the learning efficiency is the highest, so Eq ( 7): 099137664 Form No. A0101 Page 15 of 28 0992065649-0 201220010 [0035]

ΔΡ=Υ ——ΔΧ.^δχ; J

dP dP

dP a

•C

Eq(7) wherein [0037]

099137664 will improve the target amount ", solve the unknown number "C" Form No. A0101

p" is substituted with the element "k" of the Jacobian matrix 〇 J

, and each parameter correction amount "xj" Page 16 / Total 28 I Corrected 0992065649-0 [0038] 201220010 After parameter "X j", for next learning or processing [0039] 〇) r+1 — ) Rj

Eq( 8 ) [0040] 〇 Parameter learning sets the convergence condition, for example, the upper limit of the number of learning times or the lower limit of the target improvement amount of the total performance indicator. When the convergence condition occurs, that is, the parameter learning function ends, then the processing is performed using the parameters of the learning completion, and the new parameters are no longer updated or learned. Obviously, the controller with parameter learning function of the invention and the learning method thereof can adjust the parameters of the mechanical device through the parameter learning method according to the user's preference for the processing characteristics, thereby shortening the processing time, improving the processing precision, and improving. The effectiveness of mechanical processing stability. The above description of the present invention is intended to be illustrative, and is not intended to limit the precise application of the invention. Therefore, the technical idea of the present invention will be determined by the following patent application scope. 099137664 Form No. A0101 Page 17 of 28 0992065649-0 201220010 [Simplified Schematic] [0042] Figure 1 is a block diagram of the numerical control of a known CNC machine; [0043] Figure 2 is a numerical control file of the machining program FIG. 3 is a block diagram of a numerical control of a CNC mechanical device of the present invention; [0045] FIG. 4 is a block diagram of a fast quasi-stable index calculation module of the present invention; [0046] FIG. Learning function module block diagram. [Main component symbol description] [0047] 10 Digital control module [0048] 11 Machining program [0049] 12 Path planning unit [0050] 13 Motion planning unit [0051] 14 Interpolation unit [0052] 15 Driver [0053] 16 Motor [0054] 17 Position sensing element [0055] 18 Numerical control parameter [0056] 20 Fast quasi-stable index calculation unit [0057] 22 Fast index calculation sub-unit [0058] 23 Quasi-indicator calculation sub-unit [0059] 24 Stable indicator Calculation subunit form number A0101 099137664 Page 18 of 28 Page 0992065649-0 201220010

[0060] 25 Calculate the corresponding fast indicator message [0061] 26 Quasi-indicator message [0062] 27 Stable indicator message [0063] 28 Fast quasi-stable indicator message [0064] 30 Learning function unit [0065] 31 Fast quasi-weighting weight Message [0066] 32 Total performance indicator calculation sub-unit [0067] 33 Learning effect check sub-unit [0068] 34 Jacobian matrix correction sub-unit [0069] 35 Parameter correction sub-unit [0070] w.: 1 User-defined The weight of the performance indicator [0071] qi : various performance indicators; ' 乂 ' [0072] Q .: actual performance indicator change; > 1 .---.=· [0073] q.': estimate [0074] e : prediction error of performance indicator change; [0075] P: total performance indicator; [0076] P ^: target improvement amount of total performance indicator; [0077] X.: j Various parameters; [0078] X.: correction amount of each parameter; 099137664 Form No. A0101 Page 19/Total 28 Page 0992065649-0 201220010 [0079] [0081] [0083]: r The parameters used in the second study; [Jac〇bian matrix of performance indicators and parameters; K. Jacobian (jac〇bian) moment Correction amount; (K) r 1 * Di - al Science & use of a Jacobian (jac〇bian) matrix: Jacobi (the Jacobian) matrix of individual elements. 099137664 Form nickname A0101 Page 20 of 28

Claims (1)

201220010 VII. Patent application scope: 1. A controller with parameter learning function, comprising: a path planning unit for receiving a processing program and planning a processing path message according to the processing program; a motion planning unit, an input thereof The end is connected to the path planning unit and receives the processing path message, and the other input end is connected to a digital control parameter, and the processing path message is planned according to the processing path information and the digital control parameter, and a motion message is output; an interpolation unit, Connecting with the motion planning unit, and performing interpolation calculation on the motion information output by the motion planning unit, and then outputting a control command to a motor driver for driving a motor; a fast quasi-stable index calculation The unit has an input end connected to a position sensing component, and receives a feedback signal input by the position sensing component, and calculates a fast performance indicator, a quasi-performance indicator and a stable performance indicator according to the feedback signal. To output a fast quasi-stable indicator message; a learning function unit having a first input terminal Receiving a fast quasi-steady weight message set by the user, a second input end is connected with the fast quasi-stable index calculation unit and receiving the fast quasi-stable indicator message, recalculating the new digit control parameter, and outputting ^ to the movement Planning unit.控制器 2. The controller of claim 1, wherein the processing path information output motion message comprises: speed and acceleration. 3. The controller of claim 2, wherein the feedback signal comprises a position, a speed and an acceleration of the motor, etc. 4. The controller of claim 3, wherein the fast quasi-stable index The calculation unit further includes: a fast indicator calculation sub-unit, a quasi-index calculation sub-unit and a stable index calculation sub-unit, and according to the feedback signal, respectively calculate the fast performance indicator, the quasi-performance indicator and the quasi-performance indicator The 099137664 form number A0101 page 21 / 28 pages 0992065649-0 201220010 stable performance indicators. The controller of claim 4, wherein the fast-precision finger-arranging unit performs the recording of the single-segment in the step of performing the parameter learning function. The required time is added, and the fast performance indicator (FI) is calculated as Tt, and FI==aF^Ti is a fast indicator constant. For example, the control (4) described in claim 4, wherein the fast quasi-stable index calculation unit cuts the quasi-pure calculation sub-unit to calculate the relative trajectory error by calculating the relative command trajectory position and the feedback position signal. The calculation index of the quasi-performance indicator (Pi) Ei 099137664 Form No. A0101 Page 22 of 28 0992065649-0 201220010 The formula is α^α, /Υ^Ε, which is the quasi-indicator constant. The controller of claim 4, wherein the steady index calculation subunit in the fast quasi-stable index calculation unit performs a differential operation on the motor acceleration value of the feedback signal to Obtain an acceleration change value. 8. The controller of claim 1, wherein the learning function unit comprises: a total performance indicator calculation sub-unit, based on the fast-accurate re-scoring information and the fast performance indicator, the quasi-performance indicator or At least one performance indicator of the stable performance indicator is used to calculate a total performance indicator; a learning effect test sub-unit is connected with the total performance indicator calculation sub-unit, and receives the total performance indicator to perform a judgment; a Jacobian matrix Correcting the subunit, connecting with the learning effect inspection subunit, and correcting the performance index pair parameter in the Jacobian matrix according to the learning result inspection sub-unit judgment result; a parameter correction subunit, and the Jacobian matrix correction subunit Connect and set a parameter learning convergence condition. 9. A controller parameter learning method, comprising: providing a path planning unit, 099137664 Form number Α0101, page 23/28 pages 0992065649-0 201220010, for receiving a processing program and planning a processing path message according to the processing program; a motion planning unit, wherein an input terminal is connected to the path planning unit and receives the processing path message, and another input end is connected to a digital control parameter, and the processing path information is planned according to the processing path information and the digital control parameter. Outputting a motion message; providing an interpolation unit, connecting the motion planning unit and performing an interpolation operation on the motion message output by the motion planning unit, and then outputting a control command to the motor driver for driving a motor; a fast quasi-stable index calculation unit, wherein an input end is connected to a position sensing component and receives a feedback signal input by the position sensing component, and respectively calculates a fast performance indicator, a quasi-performance indicator according to the feedback signal A stable performance indicator to output a fast-stabilized indicator message; provide a learning The first input end of the energy unit receives a fast quasi-station weight message set by the user, and a second input end is connected with the fast quasi-stable index calculation unit and receives the fast quasi-stable index message, and recalculates the new digit Control parameters and output to the motion planning unit. 10. The controller parameter learning method according to claim 9, wherein the calculating step of the learning function unit comprises: providing a total performance indicator calculation sub-unit according to the fast quasi-station weight information and the fast performance indicator And the at least one performance indicator of the quasi-performance indicator or the stable performance indicator to calculate a total performance indicator; providing a learning effect test sub-unit, connecting with the total performance indicator calculation sub-unit and receiving the total performance indicator to perform a judgment; Providing a Jacobian matrix correction subunit, connecting with the learning effect inspection subunit and verifying the judgment result of the subunit according to the learning effect, correcting the performance index pair parameter in the Jacobian matrix; providing a parameter correction subunit, and the ya Comparable matrix correction sub-unit connection, and set a parameter learning convergence condition 〇099137664 Form No. A0101 Page 24 / Total 28 Page 0992065649-0 201220010 11 . The parameter learning method of the controller according to claim 10, wherein The parameter modification step of the parameter correction subunit includes: determining a total performance indicator The target improvement amount, if the previous learning is invalid, the target improvement amount is reduced, if it is effective, the target improvement amount is not changed; the direction of the modulation parameter is determined, and a parallel total performance index is selected to adjust the parameter direction gradient direction; After the corrected numerical control device parameters, according to the results of the first two steps, the correction amount of each parameter is obtained and added to the current parameter to obtain a corrected parameter. Ο 099137664 Form No. A0101 Page 25 of 28 0992065649-0