CN1128040C - Intelligent in-situ machine tool cutting flutter controlling method and system - Google Patents

Abstract

An online intelligent control method for machine tool cutting chatter is applicable to the field of precision flexible manufacturing and processing. The system control method is composed of three parts: system initialization, chatter identification and chatter control. The present invention can quickly predict the omen of cutting chatter according to the cutting vibration signal, and adjust the dynamic characteristics of the cutting system online in time according to the information contained in the acceleration signal, so as to suppress the cutting chatter in its budding state, without producing machining vibration marks on the surface of the workpiece, and without affecting The normal progress of the cutting process ensures the quality and efficiency of processing. In this method, the CDM method is used to identify chatter signs online, and the chatter signs that appear can be identified within 50ms. The online adjustment of the dynamic characteristics of the processing system mainly depends on the mechanical structure containing electrorheological materials. The invention enhances the ability of the mechanical processing equipment to adapt to different processing objects and processing conditions, and greatly improves the flexibility of the mechanical processing equipment.

Description

Intelligent in-situ machine tool cutting flutter controlling method

Technical field:

Intelligent in-situ machine tool cutting flutter controlling method belongs to automation of machinery manufacture and intelligent manufacturing field, the processing stability that mainly belongs to cut for the guaranty money, avoid occurring in the process autovibration-flutter, guarantee processing parts quality and cutting tool life-span.This technology is applicable to the accurate flexible manufacturing system of processing that automaticity is high, can be widely used in precision machined Aeronautics and Astronautics of needs and automobile and other industries.

Background technology:

Cutting-vibration belongs to autovibration, is ferocious relative vibration between the cutter that produces in working angles of Metal Cutting Machine Tool and the workpiece.The rule of reason that it produces and generation, development and cutting process itself and Metal Cutting Machine Tool dynamic perfromance all have inherent essential connection, and influence factor is a lot, is a very complicated mechanical oscillation phenomenon.Along with the factory automation development, the flexibility of machining requires cut to carry out to different workpiece with under the different operating condition, therefore the appearance that often can not fundamentally stop the flutter phenomenon by the method for flutter prevention and control is so carry out the in-service monitoring forecast and control becoming a guardian technique that improves cutting system stability to it.

Because it is sudden and uncertain that cutting system generation flutter has, very of short duration from normally being cut to the time history that flutter takes place, generally within the hundreds of millisecond.So cutting-vibration is carried out in-service monitoring and controls is very difficult.Comparatively successful method can be classified as two classes at present: class employing is carried out Flutter Control to the method that cutting system carries out system modelling, and class employing is carried out online adjustment to cut parameter (speed of mainshaft, the amount of feeding, cutting depth etc.) and suppressed flutter.But, the complicacy of cutting processing system sets up accurately very difficulty of system mathematic model because making, the serious hysteresis of cut machinery system response simultaneously makes above-mentioned two class methods all can not carry out On-line Control to cutting stability well, can not fully the flutter phenomenon be eliminated.

Summary of the invention:

Fundamental purpose of the present invention just is aimed at flexible manufacturing cell, develops a kind of method of the general cutting-vibration On-line Control that does not influence manufacturing cell's flexibility degree, to improve crudy and production efficiency.The present invention is primarily aimed at and overcomes the obstacle that the technology of flutter On-line Control in the past is difficult to break through, and utilizes intellectual material according to the On-line Control of sensor acquisition cutting vibration information regulation and control physical construction dynamic perfromance with the realization cutting-vibration.

Design philosophy of the present invention is based on theory that control cutting system time change step response improves cutting stability as its theoretical foundation, and is very little to the change of original machine cut system when specifically implementing.Original cutting system mainly is made up of lathe, workpiece and tooling system three parts, the present invention is just in the design of the key components and parts that is influencing cutting system stability (for example boring bar, handle of a knife etc.), adopted the method for embedded intellectual material, the characteristics of utilizing the intellectual material dynamic perfromance to regulate and control real-time, the dynamic perfromance of whole system of processing is carried out online quick regulation and control, and the condition of eliminating the flutter generation is to avoid the generation of cutting-vibration.

Control method of the present invention is referring to process flow diagram 1, and CDM ONLINE RECOGNITION flutter prediction is referring to CDM information flow Fig. 3 and CDM ONLINE RECOGNITION flutter omen process flow diagram 2, this method is characterised in that the cutting vibration acceleration signal of gathering according in real time, the dynamic perfromance of automatically controlled rheological characteristics rapid adjustment cutting vibration system of utilizing er material in the cutting system is to avoid cutting the generation of autovibration one flutter, and it may further comprise the steps successively:

It may further comprise the steps successively:

(1) system initialization is set: the initialization electric field strength E ₀,

Sample frequency F _s

(2) flutter identification: acceleration signal input flutter recognizer module CDM, circulation is differentiated and operation, until identify the flutter omen;

(3) Flutter Control:

(a) in a single day identify the flutter omen, calculate and get flutter frequency F this moment _C, order: F _Post=F _C, F _Post: the flutter frequency when not applying control signal;

(b) apply control electric field intensity: E=E ₀KV/mm;

(c) again acceleration signal is imported above-mentioned CDM module, whether differentiation flutter this moment omen still exists, disappear as the flutter omen, and electric field strength E=0 that order applies, program changes step (2) continuation operation over to then.Still exist as the flutter omen, program is moved downwards;

(d) calculate flutter frequency F this moment _C

(e) differentiate F _CWhether greater than F _Post

If F _C＞F _Post, then making E=E-0.1 KV/mm, measuring acceleration moves by (c)～(d) again;

If F _C＜F _Post, then making E=E+0.1 KV/mm, measuring acceleration moves by (c)～(d) again;

(4) said procedure flow process circular flow always in cutting process finishes up to cutting process, and this control program finishes;

Above-mentioned CDM module is carried out following steps successively:

(1) input sample acceleration signal r _i

(2) with r _iInput LO-RBF type neural network structure program module, operation according to the following steps successively:

(a) calculate transducing signal seasonal effect in time series probability density function at r _iEstimated value f (the r at place _i) and probability density function single order differentiation function at r _iThe estimated value at place

Promptly calculate: $(f\left(r_i\right),\frac{\∂}{\∂r_i}f\left(r_i\right))=\mathrm{LO}-\mathrm{RBF}\left(r_i\right),$

(b) structure new signal sequence G _n, contrast imports transducing signal sampled value r with each _iThe time get G _nMiddle element g _iFor: $g_i=f\left(r_i\right)/\frac{\∂}{\∂r_i}f\left(r_i\right)$

(3) with g _iInput Fuzzy ARTMap flutter omen identification module, operation according to the following steps successively:

(a) whether differentiate i greater than 128, differentiate and circular flow,

(b) be under the 1000Hz in sample frequency, differentiate i＞128 o'clock, judge every 50 milliseconds whether a flutter takes place, promptly fmod (i, 50)=0 o'clock carries out a flutter omen differentiation,

(c) take-off time burst: promptly take out preceding 127 points in the G sequence, with currency g _iConstitute 128 timing signal sequence G ₁₂₈, i.e. G ₁₂₈={ g _I-127, g _I-126..., g _I-1, g _i,

(d) with G ₁₂₈Input fast fourier transform subroutine, promptly the FFT subroutine obtains G ₁₂₈Fourier transform sequence F ₁₂₈, F ₁₂₈=FFT (G ₁₂₈),

(e) get F ₁₂₈The mould of preceding 64 elements of sequence constitutes new sequence S ₆₄, S ₆₄In the energy density S of each element representative each Frequency point in analyzing frequency range _i=| F _i| (i=1,2,3 ... 64),

(f) 64 dimension sequence S ₆₄Input Fuzzy ARTMap neuroid subroutine,

(g) differentiate the bivector C that Fuzzy ARTMap neuroid is exported ₂:

C ₂＝Fuzzy?ARTMap(S ₆₄)

If: C ₂=0,1} represents no flutter omen,

If: C ₂=1, and 0}, expression has the flutter omen,

In the above-mentioned FuzzyARTMap supervised learning stage, the M that when the cutting-vibration omen exists, collects 64 dimensional signal S ₆₄With the M that represents the flutter omen two-dimensional vector C ₂, be set at 1,0}, the ART of fan-in network respectively simultaneously _aSubmodule and ART _bSubmodule; The 64 dimensional signal S that collect when equally, N flutter omen do not exist ₆₄N two-dimensional vector C with the no flutter omen of representative ₂, be set at 0,1}, the ART of fan-in network respectively simultaneously _aSubmodule and ART _bSubmodule; Here, { 0,1} is corresponding to no flutter omen, and { 1,0} is corresponding to the flutter omen is arranged.By such supervised learning, at ART _aSubmodule and ART _bForm with weights between the submodule is set up mapping relations, guarantees that like this network can be online according to ART _aThe input of submodule differentiates in the acceleration sampled signal whether have the flutter omen.

The present invention introduces intellectual material in the On-line Control of metal cutting processing flutter, use the instant response ability of intellectual material inherent characteristic to electric control signal, not only overcome the shortcoming of cutting system physical construction to the control signal low-response, and this system carries out the online inhibition of flutter according to the heat transfer agent that obtains, and overcome the cutting system complicacy and is difficult to set up the difficulty of controlling models accurately.Therefore this invention has strengthened machining equipment and has adapted to the different processing objects and the ability of processing conditions, has improved the flexibility degree of machining equipment widely.

CDM ONLINE RECOGNITION flutter Application in Prediction mainly shows that based on lot of experiments the evolution of flutter has following characteristics in the process among the present invention:

(1) the flutter waveform is similar to resonance wave, and the growth of amplitude is the process of a gradual change, can be divided into initial flutter stage, flutter developing stage and abundant flutter stage.

(2) cutting vibration frequency is stabilized to the natural frequency near system gradually with the development of flutter.Vibrational energy distributed in frequency domain and transferred the arrowband energy distribution to by the broadband distribution this moment.

(3) in the initial flutter stage, vibration frequency has been stabilized to the natural frequency place of system, and this moment, vibration amplitude did not reach the maximum amplitude of flutter as yet.Reaching abundant flutter stage precontract in the flutter amplitude has 400 to 600 milliseconds or longer, and this just provides the quality time of identification and FEEDBACK CONTROL to supervisory system.

By These characteristics as can be known: can be used as the important omen that flutter takes place by the resonance wave characteristic of flutter and the arrowband feature of frequency domain.Flutter forecasting technique among the present invention just is based on above-mentioned analysis, adopts local optimum signal detection technology and neuroid technology in flutter developing stage identification flutter omen, carries out the flutter forecast, to reach purpose of the present invention.

Description of drawings:

Fig. 1: control method process flow diagram among the present invention, F _S=sample frequency, E=electric field intensity, E ₀=initial electric field intensity, AS=acceleration signal, CDM=flutter recognizer module, F _C=flutter frequency, the flutter frequency when Fpost=does not apply control signal;

Fig. 2: CMD ONLINE RECOGNITION flutter omen process flow diagram;

Fig. 3: CMD information flow chart;

Fig. 4: system chart of the present invention, 1, computing machine, 2, mainboard PCI slot, 3, analog/digital digital-to-analog transition card, 4, analog input mouth, 5, the simulation delivery outlet, 6, charge amplifier, 7, acceleration transducer, 8, voltage changer, 9, intelligent boring bar (containing er material);

Fig. 5: system principle diagram of the present invention, a, interference, b, dynamic cutting force, c, steering order, d, vibratory response, 10, controller, 11, the mechanical vibrating system formed of lathe-workpiece-cutter, 12, working angles signal module;

Fig. 6: intelligent boring bar synoptic diagram, 13, support set, 14, positive electrode, 15, O shape circle, 16, insulation sleeve, 17, boring bar, 18, boring cutter, L ₁, the boring bar length that is installed, L ₂, the boring bar length that overhangs;

Fig. 7: acceleration transducer is settled and the boring system schematic, and 19, chuck, 20, the cutter head anchor clamps, 21, workpiece, 22, boring tool holder;

Fig. 8: surveillance signal transmission configuration schematic diagram of the present invention, the flat adapter of 23,8 passages;

Fig. 9: utilize intelligent boring bar to carry out the system and device synoptic diagram of boring processing;

Figure 10: information input synoptic diagram during the training of Fuzzy-ARTMAP neuroid off-line learning;

Figure 11: the information input and output synoptic diagram during Fuzzy-ARTMAP neuroid online forecasting flutter omen.

Embodiment:

The present invention can use in boring processing.Because boring is endoporus processing, so boring bar generally is designed to elongated cantilever beam structure, and poor rigidity is easy to stressed occuring bending and deformation, and flutter often can't be avoided when being subjected to dynamic cutting force.The present invention is in order to overcome the weakness that boring bar rigidity can't improve at all, when its structural design, added a kind of intellectual material---er material, by er material being applied the dynamic perfromance of the online change boring bar of electric field integral body, regulate and control the boring bar dynamic perfromance to avoid the generation of flutter according to transducing signal in conjunction with flutter online forecasting technology is online.

System wherein adopts accessory and mutual relationship to be described below according to Fig. 4 technology assembling routinely in the present embodiment:

(1) the computer CPU model is PII233, and mainboard has 3 PCI slots, in system as the carrier of data acquisition, flutter forecasting controlling software.

(2) analog/digital, the digital-to-analog transition card: model is HY-6070, inserts in the computer PCI slot, and its analog input end links to each other with the voltage output end of charge amplifier, and analog output links to each other with the Input voltage terminal of high-voltage variable parallel operation.

(3) charge amplifier: model is YE5858, and the quantity of electric charge of its input end degree of will speed up sensor output inserts, and is output as the voltage signal of representing acceleration signal, and output inserts A/D, the analog input end of D/A card.

(4) acceleration transducer: model is YE14103, and acceleration transducer is fixed on the cantilever end of boring bar by dowel screw, and output links to each other with the electric charge input end of charge amplifier.

(5) voltage changer: model is GYW-010, its function be 0～10000 volt high voltage for 0～5 volt low voltage transition will being exported by computing machine, its Input voltage terminal and A/D, the analog output of D/A card links to each other, and the two poles of the earth of its high-voltage output terminal link to each other with the both positive and negative polarity of er material in the boring bar respectively.

Of the present inventionly in the structural design of cutting system, er material is introduced the machining system critical component, develop a kind of intelligent boring bar that can directly control mechanical Structure dynamic characteristics by external electric signal.The structural design of intelligent boring bar is at changed the limited shortcoming of variation range that obtains based on the laminated girder construction of er material or the dynamic perfromance of hollow beam structure with control electric field intensity in the past, utilize er material to change the local support rigidity of boring bar root, improve the variation range of Structure dynamic characteristics greatly, do not influenced normal boring processing simultaneously.Boring bar as shown in Figure 6, positive electricity is the Steel Thin-Wall circle very, the support set part relative with positive electrode is as the negative pole (ground connection just) of er material, the electrode axial length is 20 millimeters, two electrode gaps are 0.5 millimeter.Between positive electrode and boring bar insulation sleeve is arranged, er material is fed in the cavity between positive electrode and negative electrode, and the sealing of er material guarantees by 2 O type circles.Support set and boring bar are fixed from orthogonal both direction by four hexagon socket head cap screws.Support set is installed in the knife rest in the boring system, L ₁Equal 100 millimeters, L ₂Equal 180 millimeters, boring bar overhang the part L/D ratio be 6: 1.Boring cutter is installed in end at boring bar.Adopt vibrator to boring bar exciting test shows the control electric field intensity variation range of er material during at 0 to 2000 volt/every millimeter the boring bar natural frequency 30 hertz variable quantity is arranged.

Boring system, acceleration transducer are settled as shown in Figure 7.Why acceleration transducer all adopts horizontal direction, mainly be to consider that cutting-vibration mostly is the flutter of regeneration type just, can know to have only just remarkable influence cutting force of the relative vibration displacement component of cutting depth direction cutter with workpiece by regeneration type flutter mechanism of production perpendicular to this direction of cutting surface.The evolution of the reflection cutting-vibration that proof cutting depth direction vibratory response and cutting force dynamic component can be sensitiveer than the transducing signal of other direction in the enforcement.Acceleration transducer is installed on the end of boring bar, because obvious in the vibratory response of end.

The digital signal flow process figure that computing machine is directly handled as shown in Figure 8.In process, the acceleration sensing signal is converted to voltage signal through the YE5853 charge amplifier with charge signal through the collection of YE14103 accelerometer, through 12 A/D transition cards of 8 passage flat cable adapters input, becoming can be by the direct digital signal of handling of computing machine then.

The configuration of er material considers that mainly er material is operated in the room temperature range, material system will have high dispersion stabilization, material will have big viscosity, elastic modulus change scope, and the electric rheological effect stability of material is high, and material will have low electric conductivity in addition.The basic configuration process is: starch and pumping fluid by mixing, are added an amount of rosin derivative and additional additives again, at room temperature stir minute with electric blender, stirring the back is uniform brown suspension liquid.This kind material is placed under blow-by, 15～40 ℃ ambient temperature range, layering, deposited phenomenon do not occur, and electric rheological effect is obvious.

Cutting-vibration control system device synoptic diagram is shown in 9.The boring system is based upon on the CA6140 lathe, and the holding of workpiece cantilever is on main shaft, and boring cutter is installed on the knife rest.Supervisory system is the PII233 of an association computing machine of being furnished with the HY6070 data collecting card.The vibration signal of gathering is the acceleration signal of boring bar end horizontal direction.According to vibration signal and the cutting-vibration On-line Control strategy gathered, the output channel of capture card is sent control signal and is given the GYW-010 voltage changer, voltage changer is applied to the electric field of a certain size electric field intensity between both positive and negative polarity in the boring bar according to control signal, regulates and control the dynamic perfromance of boring bar with this by changing er material performance between both positive and negative polarity.Such configuration can guarantee that adjusting electric power output voltage according to the cutting vibration signal very easily carries out the online inhibition of cutting-vibration with the dynamic perfromance of control boring bar.

The operating process of system control method as shown in Figure 1.System is made up of three parts: system initialization, flutter identification and Flutter Control.

System initialization part mainly be with comprise sample frequency, by the parameter input systems such as initial electric field intensity of experience decision.Sample frequency is according to required signal frequency range and sampling thheorem decision, and initial electric field intensity is selected, determined by experiment according to the cut condition.Arrive in case the flutter omen is predicted, initial electric field intensity is applied in to er material.Like this, can the saving system be used to search for the time of optimum electric field intensity.

Giving of flutter identification division involving vibrations signal (for example acceleration, vibration displacement etc.) handled and two sport technique segments of the seizure of flutter omen in spectrogram.Giving of transducing signal handled along with working angles carries out in real time.The spectrogram of vibration signal is transfused to the network to Fuzzy ARTMap every 50 milliseconds, to be used for the identification of flutter omen.System's continuous monitoring vibration signal in working angles, in case the flutter omen is identified, flutter On-line Control program will be activated.

Third part is finished the function of flutter On-line Control, is divided into two modules.In first module, at first a peak value searching program begins to find out flutter frequency F in spectrogram _C, make F _PostEqual F _CThen, computing machine sends steering order and allows the control power supply of er material apply initial electric field intensity to give er material.Although initial electric field intensity is according to machining condition, by the optimum value that experiment is obtained, because the complicacy of cutting system, the on-line search optimum electric field intensity is still necessary sometimes.Therefore, when the flutter omen still existed after initial electric field intensity applies, second module was activated the continuous online adjustment electric field intensity of beginning and disappears up to the flutter omen.If the flutter omen disappears, program jumps to flutter ONLINE RECOGNITION part, and the electric field intensity that imposes on er material simultaneously reverts to zero electric field intensity.

In second module, adopted a feedback control strategy to adjust the generation of electric field intensity, inhibition flutter.Control strategy is as shown in Figure 7: as the flutter omen still exists after applying initial electric field intensity, then at first calculates flutter frequency F _C, as flutter frequency F this moment _CGreater than the flutter frequency F that applies before the electric field intensity _Post, then reduce to put on the electric field intensity of er material, if F _CBe not more than F _Post, then increase the electric field intensity that puts on er material.And then gather the cutting vibration signal, and judge whether the flutter omen exists, still have the possibility of generation as there being the explanation flutter, calculate flutter frequency so once more, according to said process repeated calculation, adjusting electric field intensity, disappear up to the flutter omen.

The information flow of flutter identification forecasting technique as shown in Figure 3.At first, degree of will speed up signal is adopted into computing machine by the A/D conversion, and it is because the boring flutter mainly occurs near the boring bar natural frequency that sample frequency is set at 1000Hz, is generally between 150～400Hz, so in order to reach satisfied time-frequency transformation result, sample frequency is taken as 1000Hz.

Then, degree of will speed up burst gives processing through the LO-RBF detection techniques, produces new burst.Adopting LO-RBF detection techniques purpose is to increase flutter omen signal---the intensity of resonance signal in the ground unrest.When the burst of LO-RBF detection techniques reconstruct is being done fast fourier transform, can get less sampling number and can manifest its arrowband feature among the signal spectrum figure in the flutter starting stage.Can reduce a large amount of sampling times that data are got that is used for like this.The present invention can reach the arrowband energy distribution feature that manifests the flutter omen in the acceleration frequency spectrum in the boring flutter starting stage through testing the fast fourier transform that determine to adopt at 128.Then, utilize Fuzzy Artmap network model that fuzzy theory and adaptive resonance neuroid technology combine, come the stability of working angles is judged as pattern classifier.When the supervised learning of network, the input of network be respectively the analysis frequency range that obtains after the FFT conversion (the mould vector A (64 dimension) of 1～500Hz) each Frequency point and represent the appearance of flutter omen and do not have flutter bivector B (0, and 1} with 1,0}).Behind learning success, the mapping relations of determining between vectorial A and vectorial B, have been set up.In the online forecasting stage, the real-time online acquired signal that is input as of network is carried out mould vector A after FFT changes after treatment, is output as the result that the flutter omen is forecast.

This technology is made accurate forecast can reaching on the flutter forecast speed in back 50ms appears in the flutter omen.

What adopted among the present invention is Fuzzy ARTMap ARTOICAL NEURAL NETWORK MODEL.The Fuzzy-ARTMap network that adopts in the supervised learning stage is (to be respectively ART by a pair of Fuzzy-ART neuroid _a, ART _b) and ART _a, ART _bBetween the mapping controller form.Here, ART _aInput signal be 64 dimensional signal S ₆₄, the frequency range (energy density of each Frequency point in 1～500Hz) is analyzed in representative respectively.ART _bBe input as 2 dimensional signal C ₂, be respectively 0,1} and 1,0}, they have represented flutter omen and no flutter omen respectively.

The study of network is carried out under off-line state, referring to Figure 10.By the cutting experiment under different machining conditions, obtain flutter by not having to the sample sequence that acceleration signal in the process is arranged, it is become G through the LO-RBF network reconfiguration _nSequence is to G _n128 FFT conversion are carried out in the sequence segmentation, with M 64 dimensional vector S of gained ₆₄As ART _aThe input vector group.Judge the moment that the flutter omen occurs according to the depth of surface of the work cutting chatter mark then, the input vector component is become two parts: preceding K is steadily cutting, ART when promptly not having the flutter phenomenon _aThe input vector group; The ART that (M-K) is individual in the back when existing for the flutter omen _aInput vector.With above-mentioned two parts ART _aThe input vector group corresponding, ART _bInput vector group C ₂Be respectively 0,1} and 1,0}.Fuzzy-ARTMAP adopts fuzzy minimax learning rules to increase progressively study, behind the learning success, passes through ART _a, ART _bAnd ART _a, ART _bBetween the weight vector W that couples together of mapping controller _j ^a, W _k ^bAnd W _j ^AbAt ART _aAnd ART _bBetween set up mapping relations.

The online forecasting stage is referring to Figure 11.This stage, ART _aInput signal be the 64 dimensional signal S that online in real time is obtained ₆₄, ART _bWhen no flutter omen export C as output terminal this moment ₂So that 0,1}, output C when the flutter omen ₂So that 1,0}.

Claims (1)

1. An online intelligent control method for machine tool cutting chatter, characterized in that according to the cutting vibration acceleration signal collected in real time, the dynamic characteristics of the cutting vibration system are quickly adjusted by using the electronically controlled rheological properties of the electrorheological material in the cutting system to avoid cutting self-control. Excited Vibration—Occurrence of flutter, which in turn includes the following steps: (1) System initialization, setting: initialize the control electric field strength E ₀ of the electrorheological material, Sampling frequency F _s ; (2) Flutter recognition: input the acceleration signal into the flutter recognition program module CDM, and perform circular discrimination and operation until the flutter omen is identified; (3) Flutter control: (a) Once the flutter omen is identified, calculate the flutter frequency F _C at this time, set: F _post = F _C , F _post : flutter frequency when no control signal is applied; (b) Apply a control electric field intensity to the electrorheological material: E=E ₀ KV/mm; (c) Input the acceleration signal into the above-mentioned CDM module again to judge whether the omen of flutter still exists at this time, if the omen of flutter disappears, make the applied electric field strength E=0, then the program goes to step (2) to continue running, if the omen of flutter disappears, The vibration omen still exists, and the program runs downward; (d) Calculate the flutter frequency F _C at this time; (e) Determine whether F _C is greater than F _post . If F _C > F _post , then set E=E-0.1KV/mm, re-measure the acceleration, and run according to (c)~(d); if F _C <F _post , then Make E=E+0.1KV/mm, re-measure the acceleration, and run according to (c)~(d); (4) The above-mentioned program flow runs continuously during the cutting process until the end of the cutting process, and the control program ends; the above-mentioned CDM module executes the following steps in sequence: (1) Input sampling acceleration signal r _i ; (2) input r _i into the LO-RBF type neural network structure program module, and run according to the following steps successively: (a) Calculate the estimated value f(r _i ) of the probability density function of the sensor signal time series at r _i and the estimated value of the first-order differential function of the probability density function at r _i

i.e. calculate: $(f\left(r_i\right),\frac{\&PartialD;}{{\&PartialD;r}_i}f\left(r_i\right))=\mathrm{LO}-\mathrm{RBF}\left(r_i\right),$ (b) When constructing a new signal sequence G _n , contrasting with each input sensing signal sampling value r _i , the element g _i in G _n is: $g_i=f\left(r_i\right)/\frac{\&PartialD;}{\&PartialD;r_i}f\left(r_i\right)$ (3) Input g _i into the fuzzy adaptive resonance neuron network, that is, the FuzzyARTMap neuron network, to identify flutter omens, and run according to the following steps in turn: (a) judge whether i is greater than 128, make a judgment and run in a loop, (b) When the sampling frequency is 1000Hz, when i>128 is judged, it is judged whether the chatter occurs every 50 milliseconds, that is, when fmod(i, 50)=0, the cutting chatter is judged once, (c) Take out the time signal sequence: take out the first 127 points in the G sequence, and form a time signal sequence G ₁₂₈ of 128 points with the current value g _i , that is, G ₁₂₈ ={g _i-127 , g _i-126 ,...,g _i-1 , g _i }, (d) G ₁₂₈ is input into the fast Fourier transform subroutine, i.e. the FFT subroutine, to obtain the Fourier transform sequence F ₁₂₈ of G ₁₂₈ , F ₁₂₈ =FFT(G ₁₂₈ ), (e) Taking the modules of the first 64 elements of the F ₁₂₈ sequence to form a new sequence S ₆₄ , each element in S ₆₄ represents the energy density of each frequency point in the analysis frequency band S _i =|F _i |(i=1, 2, 3, ... 64), (f) input the 64-dimensional sequence S ₆₄ into the FuzzyARTMap neuron network subroutine, (g) Discriminate the two-dimensional vector C ₂ output by the FuzzyARTMap neuron network: C ₂ =Fuzzy ARTMap(S ₆₄ ) If: C ₂ = {0, 1}, it means there is no sign of flutter, If: C ₂ = {1, 0}, it means there is a sign of flutter, In the above-mentioned FuzzyARTMap supervised learning stage, M 64-dimensional signals S ₆₄ and M two-dimensional vectors C ₂ representing chatter signs collected when the cutting chatter sign exists are set as {1, 0}, and input ART _a sub-module and ART _b sub-module of the network; similarly, the 64-dimensional signal S ₆₄ collected when N flutter signs do not exist and N two-dimensional vectors C ₂ representing no flutter signs are set to {0 , 1}, while inputting the ART _a submodule and ART _b submodule of the network respectively; here, {0, 1} corresponds to no flutter omen, and {1, 0} corresponds to flutter omen; through such supervised learning , establish a mapping relationship in the form of weights between the ART _a sub-module and the ART _b sub-module, so as to ensure that the network can judge whether there is a flutter omen in the acceleration sampling signal based on the input of the ART _a sub-module online.

Images