CN101716730A - Method for improving numerical control machine feed movement precision in batch production by learning - Google Patents

Abstract

The invention provides a method for improving numerical control machine feed movement precision in batch production by studying. A learning device is installed between an original numerical control system and an actuator without regulating and changing the original numerical control system and the actuator, which is convenient and practical. When one part is repeatedly processed, a control signal needed for processing the next part is obtained according to a control signal and an error signal of a mass storage memory numerical control system and a certain learning algorithm, and movement control is carried out; a new control signal can reduce a kinematic error when processing the previous part; by multiple remembering-learning processes, the kinematic error can be reduced so as to improve the outline kinematic accuracy of the numerical control machine tool. The control device comprises a microprocessor, a mass storage, numerical control system feed movement control command signals, control output signals and position detection signals after learning and the like.

Description

Improve the method for numerical control machine feed movement precision in the batch process by study

Technical field

The present invention relates to a kind of method, belong to Digit Control Machine Tool control technology field by numerical control machine feed movement precision in the study raising batch process.

Background technology

When Digit Control Machine Tool repeated to process certain part along certain track, its motion control need be carried out reruning along a track.The repeat property of this type games is not considered in the control of digital control system at present, and kinematic error all repeats to produce each time.The thought of study control is the error of at every turn moving by learning, and controlled quentity controlled variable is revised, thereby improved the precision of motion when moving next time.It does not need accurate system model, and the not modeling characteristic of system is had certain robustness, and amount of calculation is little in real time.Machine tool numerical control system generally adopts the position command pulse that each feed shaft is controlled at present, and each shaft-driven tracking error influences machining accuracy of NC machine tool.Owing to the reasons such as complexity of closure, computer hardware and software structure and the control algolithm of digital control system, it does not all have the feed motion learning functionality; And the user can't add and realize this learning functionality in existing digital control system.

Summary of the invention

Goal of the invention: the object of the present invention is to provide a kind of method that improves numerical control machine feed movement precision in the batch process by study.Add this method and device between conventional machine tool numerical control system and the feed shaft driver, when digital control system controller edge of bed certain track repeats processing parts, learning device can be according to certain learning rules, constantly control signal is revised, reduce kinematic error, thereby reach the purpose that improves the Digit Control Machine Tool part processing precision.

A kind of method by numerical control machine feed movement precision in the study raising batch process is characterized in that comprising following process:

(1), produces in batches, learning device is installed between existing digital control system and the motion driver, the control signal of digital control system output flows to the motion driver after handling through learning device again, wherein learning device utilizes the every 0.1-10ms of interrupt routine to carry out once, finishes corresponding learning process;

(2), in the process of carrying out first part, utilize learning device to read the control signal u of digital control system output ₀(k) and the error signal e of control signal and physical location ₀(k), wherein k is sampling instant, with above-mentioned control signal u ₀(k) and error signal e ₀(k) be stored in the memory;

(3), second and after part processing the time, utilize learning device according to learning strategy, calculate the control signal u of corresponding sampling instant _J+1(k) and flow to the motion driver, learning algorithm is:

u _j+1(k)＝Q(q)[u _j(k)+L(q)e _j(k)] (1)

In the formula: k is the sampling time; J is the number of times of repetitive study control; u _jBe control signal last time; e _jBe error signal last time; u _J+1Be this control signal; Q (q) is for guaranteeing the stable wave filter of learning process; L (q) is a learning strategy.

The above-mentioned wave filter Q (q) that states chooses zero and differs wave filter, because zero differs wave filter and has the advantage of only signal amplitude being handled and not influenced signal phase.

Above-mentioned learning strategy is the ratio learning strategy, i.e. L (z)=Γ, and wherein Γ is a scale factor, its corresponding learning algorithm is

Wherein Γ is more little, and the process that error reduces is mild more, but the speed that error reduces is slow more; Γ is big more, and the process that error reduces is fast more, but error may produce concussion.

Above-mentioned learning strategy is the proportion differential strategy, i.e. L (z)=z[K _p+ K _d(1-z ^-1)], K wherein _P, K _dBe ratio and differential divisor, its corresponding learning algorithm is

u _j+1(k)＝Q(q){u _i(k)+K _pe _j(k+1)+K _d[e _j(k+1)-e _j(k)]}

K wherein _PGet 1; K _dMore little, the process that error reduces is mild more, but the speed that error reduces is slow more; K _dBig more, the process that error reduces is fast more, but error may produce concussion.

The selection of learning strategy is main to require height according to the requirement of error convergence speed is chosen as the convergence rate to error, then need adopt the proportion differential strategy.Not high to the error convergence rate request, but require algorithm simply can select the ratio learning strategy for use.

A kind of learning device of realizing learning method, it is characterized in that: be installed between digital control system and the motion driver, comprise the microprocessor that is used to realize learning algorithm and control function, link to each other with microprocessor be used to store the mass storage of control signal and error signal in the past; Wherein microprocessor and digital control system command pulse interface, study back control impuls interface, position detection signal interface, keyboard interface, LCD display interface link.

Should note when using method and apparatus of the present invention: therefore the path accuracy of study control has proposed higher requirement to electric power system to the electric source disturbance sensitivity; The uncertainty of load disturbance also can have certain influence to the convergence rate of error in addition.

Beneficial effect: technology provided by the invention need not former digital control system and driver are made any adjustment and change, as long as this device is installed between former digital control system and the driver, repeat processing when (batch process) what carry out a kind of part, control and error signal by mass storage memory digital control system, and according to certain learning algorithm, obtain the required new control signal of next part processing and carry out motion control, the kinematic error when new control signal will reduce a last part processing.Through repeatedly remembering---the process of study, kinematic error is reduced greatly, thereby improve the kinematic accuracy of Digit Control Machine Tool, it uses simply, is easy to realization, with low cost.

Description of drawings

Fig. 1 is the drive control scheme of conventional digital control system.

Fig. 2 is for adopting the control scheme of this learning device.

Fig. 3 digital control system movement instruction curve.

Fig. 4 is along with the increase of study number of times, the variation of error IAE value.

The typical case of the concrete technical application of Fig. 5

The circle arc error curve of the 3rd iteration of Fig. 6

The circle arc error curve of the 6th iteration of Fig. 7

The circle arc error curve of the 12nd iteration of Fig. 8

Fig. 9 is control software flow schematic diagram.

Figure 10 is the learning process schematic diagram.

The specific embodiment

Fig. 1 is the drive control scheme of conventional digital control system; Fig. 2 is for adopting the control scheme of this learning device.As can be seen the present invention in one's power device be installed in the conventional scheme between the digital control system and driver, do not need former digital control system and driver are done change, convenient and practical.

Fig. 4 be motion control axle to formula 2 model descriptions along Fig. 3 curvilinear motion, adopt the proportion differential strategy to carry out the result that theoretical simulation is analyzed, pass through iterative learning as can be seen 3-5 time, the IAE of error (integration of Error Absolute Value) reduces fast.

$P\left(z\right)=\frac{{\mathrm{<figure-callout\; id="3"\; label="z"\; filenames="H2009100349435E00022.png"\; state="\{\{state\}\}">z</figure-callout>}}^3}{333.3\&CenterDot;z^4-1173\&CenterDot;z^3+1596\&CenterDot;z^2-991.8\&CenterDot;z+236.7}---\left(2\right)$

Fig. 5 has provided the typical case of concrete this technology of enforcement.Wherein microprocessor is used to realize learning algorithm and control function.Can adopt MCU, DSP, ARM or other microprocessors.Mass storage: be used to store control signal and error signal in the past, these signals are foundations of learning; Its amount of capacity determines that according to the length of Digit Control Machine Tool processing parts required time the time, long more required canned data amount was then big more, needs the memory span of outfit then big more.Show and keyboard portion: be used to set information such as its mode of learning, learning parameter.Other: mainly comprise and accept and send the required interface of various types of signal.Wherein, the TMS 320F2812 that selects for use is 32 high-speed dsps, has powerful operation capacity, is fit to the characteristics that this technology realizes; CY7C1041 is the 4Mbit mass storage, can adopt more jumbo memory as required; Adopt Altera FPGA to realize the Digital Logic that other are required; RA8335 can realize the control of 320*240 LCD display; Simultaneously, three input/output signal interfaces all adopt differential acceptance and transmission, can improve the antijamming capability of signal and increase transmission range.

The proportion differential strategy is: L (z)=z[K _p+ K _d(1-z ^-1)], K wherein _P, K _dBe ratio and differential divisor, its corresponding learning algorithm is

u _j+1(k)＝Q(q){u _j(k)+K _pe _j(k+1)+K _d[e _i(k+1)-e _j(k)]}

The experiment X that adopts, Y-axis feeding drive and are MITSUBISHI MR-J2S-70A AC servo, and it is installed on the XK0816 CNC milling machine.Selection percentage differential strategy K _p=1, K _d=6.Carry out the circular motion experiment in experimental system, arc diameter is 50mm, and motion angular speed is 0.4.After adopting learning algorithm, Fig. 6, Fig. 7, Fig. 8 are respectively and learn the 3rd, 6,12 time circular motion error curve.The initial circularity profile errors of motion is about-0.8mm-+0.9mm for the first time; Through 3 iteration, the circularity profile errors is decreased to pact-0.4mm-+0.4mm; Through 12 iteration, the circularity profile errors is decreased to pact-0.05mm-+0.08mm.Along with the increase of study number of times, the contour motion error obviously reduces as can be seen.

Fig. 9 is control software flow schematic diagram.Main program after the start reads the instruction of digital control system, and is stored in the memory, as the foundation of follow-up error calculating.After the motion beginning, the every 5ms of interrupt routine (under the situation that CPU allows, break period is short more good more, generally is no more than 10ms) carries out once, finishes the learning process in all moment.

Figure 10 is the learning process schematic diagram.It has described the relation of the k time study and the k+1 time study, and it has reflected the implementation procedure of formula (1) learning strategy in fact.

Claims (4)

1. one kind is passed through the method that study improves numerical control machine feed movement precision in the batch process, it is characterized in that comprising following process:

(1), produces in batches, learning device is installed between digital control system and the motion driver, the control signal of digital control system output flows to the motion driver after handling through learning device again, wherein learning device utilizes the every 0.1-10ms of interrupt routine to carry out once, finishes corresponding learning process;

(2), in the process of carrying out first part, utilize learning device to read the control signal u of digital control system output ₀(k) and the error signal e of control signal and physical location ₀(k), wherein k is sampling instant, with above-mentioned control signal u ₀(k) and error signal e ₀(k) be stored in the memory;

(3), second and after part processing the time, utilize learning device according to learning strategy, calculate the control signal u of corresponding sampling instant _J+1And the control signal u in the refresh memory (k) and flow to the motion driver, _J+1(k) and error signal e _J+1(k), wherein learning algorithm is:

u _j+1(k)＝Q(q)[u _j(k)+L(q)e _j(k)] (1)

In the formula: k is the sampling time; J is the number of times of repetitive study control; u _jBe control signal last time; e _jBe error signal last time; u _J+1Be this control signal; Q (q) is for guaranteeing the stable wave filter of learning process; L (q) is a learning strategy.

2. the method by numerical control machine feed movement precision in the study raising batch process according to claim 1, its feature exists: above-mentioned wave filter Q (q) is the zero wave filter that differs.

3. according to claim 1 improve to produce in batches by study in the method for numerical control machine feed movement precision, its feature exists: above-mentioned learning strategy is the ratio learning strategy, i.e. L (z)=Γ, wherein Γ is a scale factor, the learning algorithm that it is corresponding

Wherein Γ is more little, and the process that error reduces is mild more, but the speed that error reduces is slow more; Γ is big more, and the process that error reduces is fast more, but error may produce concussion.

4. the method by numerical control machine feed movement precision in the study raising batch process according to claim 1, its feature exists: above-mentioned learning strategy is the proportion differential strategy, i.e. L (z)=z[K _p+ K _d(1-z ^-1)], K wherein _P, K _dBe ratio and differential divisor, the learning algorithm that it is corresponding

u _j+1(k)＝Q(q){u _j(k)+K _pe _j(k+1)+K _d[e _j(k+1)-e _j(k)]}

K wherein _pGet 1; K _dMore little, the process that error reduces is mild more, but the speed that error reduces is slow more; K _dBig more, the process that error reduces is fast more, but error may produce concussion.

Images