CN101398672A - Learning method for enhancing positioning accuracy of folding mould mechanism - Google Patents

Abstract

The invention discloses a learning method used for improving the positioning accuracy of a mold opening and clamping mechanism of a plastic jetting-molding machine. The position preact of the mold opening and clamping motion is learned by the iterative learning law to compensate for the stop position error value of the moving mold caused by the pressure flow moderating process, the response time lag of a hydraulic control system, and the inertia effect of the mold opening and clamping mechanism when the motion of the mold opening and clamping mechanism is stopped. The position preact of the mold opening and clamping motion is determined, the stop error of the mold opening and clamping mechanism is reduced, and the positioning accuracy of the mold opening and clamping mechanism is improved. A variable learning gain pk method is proposed to accelerate the algorithm convergence rate; and a forgetting factor gamma method and a learning law error tolerance delta control method are applied to improve the robustness of the learning algorithm. Under the precondition of ensuring the stability of the mold opening and clamping motion by the learning method, the mold opening and clamping mechanism can reach higher accuracy under the normal working conditions, and meet the related process requirements of the plastic jetting-molding technology.

Description

A kind of learning method that improves positioning accuracy of folding mould mechanism

Technical field

The invention belongs to field of intelligent control technology, relate to injection machine clasp mould mechanism and relevant hydraulic control system, relate in particular to the method that a kind of iterative learning improves the injection machine positioning accuracy of folding mould mechanism based on electrohydraulic proportion technology.

Background technology

Modern injection moulding machine is a system that collection is mechanical, electrical, liquid integrated.Because this equipment has characteristics such as moulding the part with complicated structure, back processing capacity plastics kind little, processing are many, therefore, obtains widespread usage.Now, the engineering plastic product more than 80% all adopts injection molding process processing.

The injection machine course of work generally can be divided into following steps: matched moulds → injection pressurize → melten gel feeds in raw material → cooling and shaping → die sinking → thimble holder mould.Overwhelming majority injection machine has all adopted hydraulic system to realize the control of step switch and the control of pressure and flow, thereby finishes above-mentioned working cycle.The hydraulic control system that is applied to injection machine is broadly divided into three classes: conventional hydraulic control system, servo hydraulic control system and ratio hydraulic control system.Conventional hydraulic injection molding machine control system was popularized last century 50, the sixties, form by common hydraulic pump, surplus valve etc., pressure, flow, direction to hydraulic system can only be the control of switching regulator break-make, pressure and speed regulate what has only, be difficult to satisfy complicated plastic product forming technological requirement, and power consumption is big, therefore adopts the injection machine of conventional hydraulic system to belong to obsolete product.The servo-hydraulic system is meant that the single or multiple servo-valves of configuration come the accurately pressure and the flow of controlled hydraulic system in oil circuit, has excellent stable state and dynamic property, and its response frequency can reach more than the 100HZ.But the hydraulic system that adopts electrohydraulic servo valve exists the cleanliness to element precision and work fluid to require height, thereby causes shortcomings such as cost height, oil rub resistance ability, is not used widely in the hydraulic injection molding machine system.In actual applications, pressure and flow that Shooting Technique is crossed the range request hydraulic system obtain successive control, and its control accuracy do not need to require very high, and hydraulic system need have oil rub resistance ability preferably simultaneously, and cost is low.Adopt the hydraulic system of electro-hydraulic proportional valve can satisfy above-mentioned requirements, therefore modern hydraulic injection molding machine control system almost not the employing of exception electrohydraulic proportion technology.Electrohydraulic proportion technology is by the proportional action of electromagnetic force with the spring force of the opening amount of controlling spool or spool, the ratio of flow system flow and pressure that realizes is amplified control, carries out single-stage, Multistage Control thereby reach injection speed, screw speed, folding mould speed, dwell pressure, screw rod torque, injection seat pressure, ejecting force etc.

The electric-hydraulic proportion pressure flow combination valve that uses in the hydraulic system can be divided into open loop control ratio valve and closed-loop control proportioning valve two classes by control mode.When adopting the hydraulic work system of open loop control ratio valve, provide control signal with the corresponding pressure of setting value, flow by controller, proportioning valve is exported corresponding pressure and flow by this signal controlling spool opening degree.Because in the control procedure, to spool position is open loop control, the control accuracy of its control and repeatable accuracy can only be carried out accuracy guarantee by proportioning valve, and control procedure can be subjected to the influence of various changing factors such as oil temperature, greasy dirt and the fluctuation of load, therefore do not reach very high control accuracy, can only satisfy the control requirement of conventional products.Improve control accuracy, solve control problem and mainly contain two kinds of approach: the control system or the proportion control system of closed ring that adopt full servo-valve.Wherein fairly simple way is to adopt the closed loop proportional valve, and the closed-loop control of the spool position of passing ratio valve itself improves control accuracy.Also there is the injection machine that adopts open loop proportional variable pump and closed loop proportional variable output pump technology in the existing market, the control principle of pump control system is the same with switching ring proportioning valve, but the purpose that can reach injection machine energy-saving control by the output pressure and the flow of direct control hydraulic pump.

In China, by 2003, the injection machine total production reached 5.1 ten thousand, ranked first in the world, and the market share accounts for 60% of Gross World Product.At present in the home market, what account for the hydraulic injection molding machine system employing of total production more than 95% is electrohydraulic proportion technology, and what wherein the overwhelming majority adopted is the proportioning valve (position control is not the close-loop control mode that adopts position error signal to control) of open loop control mode.This type systematic all disposes the real-time computer control system and controls the injection machine action.At first on operation pages, set pressure, flow and the action switching position (electronic ruler or travel switch mode) of injection machine actions at different levels by the user, according to the user parameter is set by controller then, carry out in certain sequence and change action, thereby finish the injection moulding technological process of production.

Concerning injection machine, folding mould device is to guarantee that the injection machine mould locks and realize mould open/close reliably and eject the parts of product.Two main variables of folding mold process control are mold clamping force and position.The repetition of mold clamping force is a necessary condition of stablizing molding cycle, and the hydraulic system of employing electrohydraulic proportion technology can better satisfy output and open the mold clamping force accuracy requirement.Injection machine is to the clasp mould mechanism status requirement: refer to that on the one hand advanced injection machine constantly pursues the raising of efficient, shifting formwork speed directly influences molding cycle, therefore requires shifting formwork speed fast as far as possible; On the other hand, because the continually developing and promoting of special process, the control accuracy of clasp mould mechanism location requires more and more higher.

At present, open loop proportioning valve+solenoid directional control valve control is all used in the action of the most of injection machine folding of China mould.Owing to the output of ceasing and desisting order in the action of folding mould arrives the moderating process that also there is a pressure flow in the moving platen actual stop position, therefore add that there is reaction time lag in hydraulic system, the physical location that because of inertia effect moving platen is stopped when stopping at the matched moulds position of departing from its setting.Die sinking stop position error is approximately several millimeter～tens millimeters, and what influence is such positional precision do not have to half general/fully-automatic production, we also discover less than.But along with plastic product scale, automated production develop gradually, for enhancing productivity, in process of production, increasing production firm begins to be equipped with robot device to injection machine.When using mechanical arm to carry out fully-automatic production taking-up goods, because the movement locus of mechanical arm is certain, if the position of die sinking is excessive, mechanical arm may be got less than product, causes mechanical arm frequently to be reported to the police, and interrupts the fully-automatic production process.The automatic mold inner poster technology of rising gradually in the present Shooting Technique also split cavity bearing accuracy has also proposed requirements at the higher level, its manufacture process is: mechanical arm picks up the label that has printed, be placed in the mould, high temperature when melting by the plastics melten gel melts the thermosol at the in-mold labels back side, so that label and container combine together.The stationarity and the bearing accuracy of the action of folding mould that controller of plastic injection molding must guarantee, it is accurate and can not be moved in mould could to guarantee to approach the label pad pasting.In the matched moulds process, for preventing extruding and the impact of mechanical system to mould, require moving platen according to slowly → fast → slow mode moves, and the matched moulds stop position wants accurately, thereby reach the effect of protection mould.

It is relevant that folding mould positioning error is set the action flow/pressure with injection machine clasp mould mechanism size and user, the injection machine of different model, because clasp mould mechanism rigidity and the difference of required pressure flow of moving, so folding mould positioning error has nothing in common with each other.Even to the injection machine of unified model, because the inconsistency of each parts, its folding mould bearing accuracy also has nothing in common with each other.At present, for fear of the clasp mould mechanism positioning error, solution mainly is to adopt mechanical system to limit folding mould position, thereby arrives the purpose of accurately locating and protecting.The employing of die sinking stop position error is installed the die sinking mechanical positioner additional and is prevented, and matched moulds stops stop position and protected by matched moulds termination limitation travel switch.The method that employing adds mechanical positioner or travel switch guarantees folding mould bearing accuracy, because the action of folding mould arrives when setting stop position, pressure and flow may not shed yet fully in the oil circuit, the hard closing action can cause the physical shock phenomenon, thereby causes not steadily also acceleration mechanical wearing and tearing of folding mould action.

Summary of the invention

At the horizontal present situation of prior art injection machine control technology, the objective of the invention is to improve the bearing accuracy of clasp mould mechanism in the injection machine course of work.The present invention utilizes injection machine working in reciprocating mode characteristic, has proposed a kind of method of raising injection machine positioning accuracy of folding mould mechanism of practicality.This method is applicable to and adopts open loop classification action control and utilize Linear displacement transducers such as electronic ruler or scrambler to carry out the injection machine system or the similar hydraulic control system of the positioning control of folding mould.

In order to realize described purpose; the present invention proposes a kind of learning method step that improves positioning accuracy of folding mould mechanism and comprises: learn the position lead that the action of folding mould stops by iterative learning control law; when being used to compensate folding mould action and stopping, because of the response time lag of pressure flow moderating process, hydraulic control system and the moving platen stop position error amount that causes at the inertia effect of clasp mould mechanism.

According to embodiments of the invention, described law of learning comprises the steps:

Step a: the stop position W of given setting _d, choose the lead u (1) of learning process for the first time;

Step b: in the positioning control of folding mould, add lead u (1) input; Sampling obtains dynamic model actual stop position w according to Linear displacement transducer simultaneously ₁, calculate output stop position error e (1) after the die sinking action is finished;

Step c: utilize iterative learning control law to learn the lead u (2) of the 2nd learning process;

Steps d: the process that repeating step b, step c are identical can obtain u (3), u (4) successively ... u (k), u (k+1) ... and corresponding e (2), e (3) ... e (k) ..., converge within the desired scope up to folding mould positioning error;

Step e: finish learning process.

According to embodiments of the invention, utilize described folding mould action repetitive cycling characteristic to learn the action of folding mould and stop the anticipated future position amount.

According to embodiments of the invention, the law of learning form of the k time study lead u (k) is:

u(k+1)＝Q(z)u(k)+p _ke(k)，

Wherein, k is an iterations; E (k) is the k time folding mould stop position error; Q (z) is a low-pass filter, is used to strengthen the robustness of learning algorithm; p _kBe variable learning gain.

According to embodiments of the invention, wave filter Q (z) need be chosen as a cause and effect low-pass filter, and Design of Filter is Q (z)=(1-γ), and γ is a forgetting factor, its essence is the low-pass first order filter along the iteration axle, and forgetting factor γ is used to improve the robustness of learning algorithm.

According to embodiments of the invention, variable learning gain p _kSatisfy formula | 1-p _kG _u|＜1, be used to improve the learning algorithm speed of convergence, become learning gain p _kSatisfy:

$p_k=\left\{\begin{array}{cc}0.15+\mathrm{proj}\left(\frac{\left|e\left(k\right)\right|}{E_m}\right)& \left|e\left(k\right)\right|>\mathrm{\δ}\\ 0& \left|e\left(k\right)\right|\≤\mathrm{\δ}\end{array}\right.$ And | 1-p _kG _u|＜1,

E wherein _mFor the die sinking of corresponding injection machine type under typical die sinking parameter setting stops the error expectation value; δ is an error margin; $g_u=\∂g/\∂u,$ G () is the transport function between folding mould stop position w (k) and the lead u (k), and proj () is a limiter.

According to embodiments of the invention, described change learning gain p _kDie sinking under typical die sinking parameter setting stops to exceed the flat expectation value E of error with corresponding injection machine type or each type maximum clamping force _kRelevant.

According to embodiments of the invention, the flat expectation value E of described error _kBy setting up a look-up table, and use interpolation method and calculate, determine the E of various different type of machines _kValue is used for learning gain P _kCalculating.

According to embodiments of the invention, control learning gain by setting limiter proj () and error margin δ, be used to improve the robustness of iterative learning algorithm, limiter is used to limit the maximal value of learning gain, guarantees | 1-pg _u|＜1 condition satisfies, and certain redundancy is arranged, and is stable in order to guarantee learning algorithm.

According to embodiments of the invention, after the clasp mould mechanism positioning error reached error margin δ, learning gain was 0, promptly stops iterative learning, when the clasp mould mechanism positioning error surpasses error margin δ, began iterative learning again.

Good effect of the present invention, under injection machine action open loop step control mode, the clasp mould mechanism positioning error mainly is to arrive the setting stop position by hydraulic control system perception dynamic model to cause to the moderating process of carrying out output and response time lag.For reaching the purpose of accurate positioning control, before stop position, just should slow down and turn-off direction valve during certain distance at the action hydraulic cylinder, in the time of time lag, compensate this distance by the inertia motion of dynamic model own.According to injection machine folding mould action reciprocating action characteristic, in case positioning accuracy of folding mould mechanism reaches requirement, then this lead is determined, therefore available iterative learning method is learnt to estimate to this lead.

Iterative learning control law in the learning process of the law of learning of folding mould action stop position lead can effectively be controlled the non-linear controlled device with the motion of repetition period property, and the design of control algolithm does not rely on the dynamic system precise math model, only need less priori to get final product, we can say that iterative learning control is a kind of basic non-model control method.The present invention is by the lead of iterative learning method study folding mould action stop position, and certain position begins action and stops before arriving folding mould stop position, thereby reaches higher folding mould bearing accuracy.The control algolithm design needs priori few, and algorithm is simple, and calculated amount is little, and adaptability is strong, and under the prerequisite that does not change existing folding mould method of controlling operation at different levels, can easily be embedded in the existing control algolithm and go.

The present invention is by a kind of P type iterative learning method study folding mould stop position lead, and folding mould stop position lead, a preceding folding mould stop position deviation and a learning gain are relevant in front of this folding mould stop position lead.

Through a large amount of experiments, by analyzing the speed of convergence of folding mould positioning error under the different learning gain, the present invention proposes a kind of choosing method of P type law of learning, propose a kind of improvement P type iterative learning control method on this basis, thereby accelerate folding mould positioning error speed of convergence based on variable learning gain.

Be the interference that exists in the compensation real system, initial error and measurement noise, the present invention carries out Filtering Processing by adding a wave filter to this action controlled quentity controlled variable, also reaches the purpose that strengthens the control learning algorithm robustness.

Be the robustness of enhancement algorithms, the present invention also adopts the method for positioning error tolerance limit to control law of learning.

By reading following detailed description and, just can understanding principle of the present invention and the characteristics thereof of characterizing with reference to relevant drawings.

Description of drawings

Fig. 1 is a prior art toggle rod type clasp mould mechanism electrohydraulic control system schematic diagram

Fig. 2 is a prior art injection moulding machine mould open classification action synoptic diagram

Fig. 3 is the die sinking action control block diagram of prior art

Fig. 4 is the pressure flow variation diagram of the die sinking classification action of prior art

Fig. 5 is the die sinking action control block diagram of the present invention's one example embodiment

Fig. 6 is the die sinking classification operating pressure fluctuations in discharge figure of the present invention's one example embodiment

Fig. 7 is that the die sinking of the present invention's one example embodiment stops the lead learning process

Fig. 8 is the die opening mechanism positioning error change procedure under the different iterative learning gain of the present invention p

Embodiment

Describe each related detailed problem in the technical solution of the present invention in detail below in conjunction with accompanying drawing.Be to be noted that described embodiment only is intended to be convenient to the understanding of the present invention, and it is not played any qualification effect.

The method that is used to improve positioning accuracy of folding mould mechanism that the present invention proposes is not subjected to the restriction of clasp mould mechanism type.What use always in the existing market is the toggle rod type clasp mould mechanism; and the toggle rod type clasp mould mechanism is because its mechanical property; the matched moulds bearing accuracy is higher; and die sinking and mould assembling action flow process are similar; therefore following die sinking action with injection machine toggle rod type clasp mould mechanism is embodiment, and the present invention is set forth:

The clasp mould mechanism kind of injection machine is more, by the principle of work branch, mainly contains fluid pressure type (direct press type) and toggle rod type (mechanical type) two major types.They all are made up of template, pull bar, clamping and other auxiliary equipments.Toggle rod type clasp mould mechanism essence is a mechanical type force-increasing mechanism, leans on the self-locking action of this mechanism to can be used for setting up the needed mold clamping force of mould.Because the toggle rod type clasp mould mechanism is after setting up mold clamping force; promptly no longer need the effect of hydraulic coupling; hydraulic system can be used for controlling other topworks; and has folding mould quick action; energy consumption is low; hydraulic system is simple, and therefore low cost and other advantages is obtaining widespread use in the injection machine market at present.

Fig. 1 is the electrohydraulic control system schematic diagram of injection machine toggle rod type clasp mould mechanism, and clasp mould mechanism also comprises die cylinder 101, mode transfer nut 102, rear pattern plate 103, what adopt is parts such as five hole deviations row hyperbolic elbow clasp mould mechanism 104, moving platen 105, guide pillar 106, solid plate 107, electronic ruler 108, real-time controller 109 fixed displacement pumps 110, four-way valve 111, flow valve 112, pressure pilot stage 113 and flowrate control valve 114.The hydraulic control system oil sources adopts fixed displacement pump 110 to supply with, and flow, pressure adopt composite electromagnetic ratio pressure flowrate control valve 114 to carry out stepless control, and direction of action control adopts electromagnetic switch three-position four-way valve 111 to control.

Solenoid-operated proportional pressure flow operation valve 114 is on the basis of level pressure difference overflow type threeway proportional flow control valve 112, increases ratio pressure pilot stage 113 and combines.This class combination valve both can carry out proportional control to delivery rate, can regulate system pressure realization ratio again.Proportional control valve can make things convenient for rapidly, realize accurately injection machine action working cycle action control, and the acceleration and deceleration transient process between switching by control action, avoid spike pressure, the life-span of prolonged mechanical and Hydraulic Elements, proportional control valve is less demanding to the fluid clean level simultaneously, therefore is used widely in the hydraulic injection molding machine control system.

The actions at different levels of injection machine folding mould are controlled by real-time controller 109, controller is by solenoid directional control valve 111 control action directions, control ratio valve 114 is by the pressure and the flow of setting value output respective action then, and move switching according to the operating position that sampling electronic ruler 108 obtains the real time position (the electronic ruler measuring accuracy is 0.1mm, and sample frequency is 2kHz) of moving platens 105 and user's setting.

Fig. 2 is injection moulding machine mould open classification action synoptic diagram.107 represent cover half among the figure, and 105 represent dynamic model.At present, adopt the controller of plastic injection molding of electrohydraulic proportion technology all to adopt the open loop control mode of action step control to realize the injection machine motion flow, whole die sinking action generally can be divided into the level Four action: die sinking one is slow 203, one 202 of die sinking, two 201 of die sinkings, die sinking two slow 200.In the die sinking action, die sinking one slow effect is that the mould that will lock draws back, and for fear of drawing bad product, so pressure is wanted height, flow is low, and actuating length is shorter, is advisable mould can be drawn back.Die sinking one fast stroke will be grown, and makes pressure flow no change in most strokes of die sinking, and fast and die sinking two constitutes a level and smooth curve that stops slowly jointly by die sinking two at last, finishes die sinking and moves.The desirable speed-change process of die sinking action be movement velocity by slowly → fast → change slowly, can make the injection machine even running like this, reduce mechanical shock.Action reasonable parameter methods to set up at different levels, to make approximate mild curve of change curve of pressure, flow exactly, the user can and be provided with operating pressure flows at different levels and the action switching position by continuous adjustment by above-mentioned rule, reach the die sinking action smoothly, purpose rapidly.

Finishing of die sinking action, at first need pressure and flow and the at different levels action switching position (A, B, C, D) of user, the typical die sinking action parameter example of a cover that table 1 is provided with for reach HD120 type injection machine in the sea in the actions at different levels of controller of plastic injection molding host computer setting.

Die sinking operating pressure flows at different levels and position that table 1 user sets

Action	Die sinking two is slow	Two of die sinkings	One of die sinking	Die sinking one is slow
					Pressure (bar)	30	45	60	50
Flow (%)	30	40	50	50
					Action stop position mm	260.0	220	150	50

Fig. 3 is the die sinking action control block diagram of prior art.Before the die sinking action beginning; controller opens solenoid directional control valve 111 is communicated with the die sinking oil circuit; sending into comparer 211 according to the action switching positions at different levels 210 of user's setting with the dynamic model physical location 214 that records by Linear displacement transducer 108 then compares; judge and carry out that classification action; and corresponding pressure and the flow 212 of output, clasp mould mechanism 213 is carried out corresponding actions.After dynamic model arrives die sinking two slow 200 action stop position A, i.e. phase down pressure flow output, and then close solenoid directional control valve 111, thus finish whole die sinking action.

Fig. 4 is the pressure flow change curve of corresponding die sinking actions at different levels.By the method for action step control, when actions at different levels are switched, there is the situation of pressure and flow sudden change unavoidably.If directly switch, will cause action to be impacted, influence the stationarity of die sinking action, therefore when operating pressure and flow sudden change, need change by a certain percentage.Pressure and flow uphill process are called pressure oblique ascension 204 and flow oblique ascension 205, and the decline process is called pressure oblique deascension 206 and flow oblique deascension 207.Oblique ascension, oblique deascension parameter stipulated when pressure flow from current output valve when new output valve changes, the size that the DAC output valve changes in the unit interval, it has directly determined the flatness of mechanical part action.The parameter of the oblique ascension of die sinking actions at different levels and oblique deascension pace of change also is to set in advance in controller by the user.

As shown in Figure 4; when the action of folding mould stops; owing to also there is the moderating process (206,207 indication dash area) of pressure and flow behind the arrival stop position; and the response time lag of hydraulic system; thereby make the die sinking meeting exceed the position of its setting because of action of inertia makes the moving platen actual stop position, general die sinking position error range is several millimeters～tens millimeters.At present; adopt electromagnetic direction valve+proportioning valve to realize under the action control mode in the injection machine action; because the existence of electromagnetic switch and proportioning valve response itself is than large time delay; clasp mould mechanism rigidity is big in addition, response frequency is low; and the action of folding mould is unidirectional, therefore adopts the closed loop position control method infeasible.For satisfying plastic product scale, automated production development need gradually, the solution of split cavity bearing accuracy mainly is to adopt to install the die sinking mechanical stop limiter additional and limit the die sinking stop position at present, arrives positioning accuracy request.The method that employing adds mechanical positioner guarantees folding mould bearing accuracy, because the action of folding mould arrives when setting stop position, pressure and flow may not shed yet fully in the oil circuit, will cause the physical shock phenomenon, and the die sinking action steadily also can the acceleration mechanical wearing and tearing.

Fig. 5 is the die sinking action control block diagram of the lead 215 that embeds law of learning study die sinking action and stop.The present invention adopts the method for iterative learning control law 216 to determine the lead 215 that the die sinking action stops, and this lead 215 is embedded in the open loop classification die sinking action control goes, when actual dynamic model position 214 arrives when setting die sinking stop positions 210 and deducting die sinking and stop the position of lead 215, be the output of closing presure and flow, thereby reach the purpose that improves the die sinking bearing accuracy.

Fig. 6 improves the die sinking operating pressure fluctuations in discharge figure of die sinking bearing accuracy for adopting the iterative learning method.The A position is for setting the die sinking stop position.After adding the iterative learning algorithm, after the die sinking action enters die sinking two slow 200 actions, shift to an earlier date certain distance 208 at arrival die sinking stop position A and promptly begin to reduce operating pressure and flow.The iterative learning algorithm is this lead 208 of repetition learning in the die sinking action that repeats, and makes the progressive die sinking stop position that is tending towards setting of final die sinking stop position.

Fig. 7 has represented 222 processes up to the k time 223 iterative learning the 1st time 221, the 2nd times for die sinking stops the learning process of lead u among the figure, the concrete steps of law of learning 216 are as follows:

Step a: the stop position 220 of given setting is designated as W _d, choose the lead 225 of learning process for the first time, be designated as u (1), the general mould action lead u of folding for the first time (1) is set to 0;

Step b: in the positioning control of folding mould, add lead u (1) input; Sampling obtains dynamic model actual stop position w according to Linear displacement transducer simultaneously ₁, calculate output stop position error e (1) after the die sinking action is finished;

Step c: utilize iterative learning control law to learn the lead u (2) of the 2nd learning process;

Steps d: the process that repeating step b, step c are identical can obtain u (3), u (4) successively ... u (k), u (k+1) ... and corresponding e (2), e (3) ... e (k) ..., converge within the desired scope up to folding mould positioning error;

Step e: finish learning process.

Law of learning 216 utilizes the lead 226 of the k time 223 die sinking action, is designated as u (k) and this die sinking stop position error 224, is designated as e (k), obtains the action of die sinking next time by certain law of learning and stops lead 227, is designated as u (k+1).By iterating, make die sinking stop position error reduce gradually, up to satisfying positioning accuracy request.

What law of learning 216 adopted among the present invention is p type iterative learning control law, and it is the Iterative Learning Control Algorithm that has only kept the proportional error gain coefficient, also is the most frequently used law of learning, and the law of learning form that is used to learn the lead u that die sinking stops is

u(k+1)＝u(k)+pe(k)， (1)

In the formula, u (k+1) is that the die sinking of the k+1 time iteration stops lead; U (k) is the k time a lead; P is a learning gain; E (k) is the k time die sinking stop position error, e (k)=w _k-w _d, w _kBe the k time die sinking actual stop position, w _dFor setting the die sinking stop position.

Suppose that injection moulding machine mould open action is a desirable reciprocating action process, then in theory, under the constant situation of injection moulding machine mould open action setup parameter and system, repeat the die sinking action and will cause identical system responses.Investigation formula (1) this simple control algolithm, getting in touch between having set up this die sinking action and last time die sinking being moved.Can intuitively imagine: fully approach required value when last time die sinking action stops lead u (k), then u (k) can be competent at as the ingredient that next die sinking stops lead u (k+1).If this error signal is if can not ignore, then pe (k) will play the part of the role of error correction, and revised die sinking stops lead and is formula (1).Angle from study, iterative learning with the control experience u (k) in past as the control priori, add that still there is the correction of error decision in control this time, this process is that the similar mankind pass through constantly to repeat and revise a certain action, grasps the learning process of correct way.

If there are following funtcional relationship: w (k)=g[u (k) in die sinking stop position w and die sinking stop position u lead], and satisfy:

0<a ₁≤g _u≤a ₂<∞

If the p type iterative learning control law of formula (1) satisfies:

|1-p·g _u|<1， (3)

Then die sinking positioning error w _kTo increase and decay with iterations.By formula (2), formula (3) needs to know g as can be known _uBound a ₁, a ₂, just can be by choosing suitable learning gain p, make the iterative learning algorithm convergence, setting value is progressively approached in the die sinking position location.

Learning gain p determines.Because die opening mechanism is the nonlinear system of an open loop input and output stable (BIBO), therefore there is a ₁, a ₂, make funtcional relationship g between die sinking stop position w and the die sinking stop position u lead _u(2) satisfy condition.But, therefore be difficult to determine g because actual die opening mechanism response characteristic is non-linear _uBound.In the practical application, for assurance formula (3) is set up, can be by experiment, the method that adopts examination to gather is selected the OK range of learning gain p.

Fig. 8 is the change procedure of the different learning gain situation lower open die of employing mechanism positioning error, and the experiment type is that the sea reaches HD120 type injection machine.Convergence process according to die sinking positioning error among Fig. 8 can find out that for iterative learning algorithm (1), learning gain p is more little, and it is slow more then to learn error convergence speed, and the study number of times is many more.Increase learning gain p, can reduce error and accelerate response speed, but be subjected to the restriction of the condition of convergence (3), choosing of learning gain p can not increase arbitrarily.Among Fig. 8, if p excessive (p=1.4), the fluctuation of die sinking positioning error is bigger, is not easy to converge to ideal value on the contrary.If p continues to increase, will cause formula (3) not satisfy, the die sinking positioning error is dispersed in the learning process.

In order to accelerate the learning algorithm response speed, guarantee the positioning error asymptotic convergence simultaneously, can accelerate the iterative learning algorithm the convergence speed by the method that becomes learning gain p.Sum up the experimental data rule, when if die sinking stops lead u (k) away from desirable range of control, adopt big learning gain p, can make it faster near desirable controlling value, when die sinking stops lead u (k) near desirable range of control, adopt less learning gain p, can make the desirable controlling value of its more stable convergence.Comprehensive above the analysis proposes a kind of p type iterative learning algorithm that becomes learning gain, and learning gain p Changing Pattern is as follows:

$p_k=0.15+\frac{\left|e\left(k\right)\right|}{E_m},---\left(4\right)$

In the formula, E _mBe the different injection machine types of correspondence, under typical die sinking parameter setting, die sinking stops to exceed the error expectation value.The size of injection moulding machine mould open mechanism is directly proportional with required maximum clamping force.The injection machine model generally is expressed as brand+mold clamping force, reaches HD120 as the sea, and the expression maximum clamping force is 1200KN, and three along SHE500, and the expression mold clamping force is 5000KN to the maximum.The clasp mould mechanism equivalent load of different injection moulding types is inequality, and required oil circuit maximum pressure and flow are also inequality, so its folding mould positioning error difference.Under the action of die sinking at present open loop step control mode,, find that the die sinking stop position exceeds error becomes the approximate reverse ratio with injection machine type maximum clamping force nonlinear relationship through measuring.The injection moulding type little to mold clamping force, clasp mould mechanism rigidity is little, causes die sinking stop position error bigger on the contrary, and along with the increase of mold clamping force, the rigidity of the clasp mould mechanism of injection machine becomes greatly gradually, and it is more little that corresponding die sinking stop position exceeds error.

E in the formula (4) _mCan inquire about respective value by setting up a sequential search table by the ordering of plastic injection molding machine clamp force size.Die sinking in the look-up table under the various different injection moulding type canonical parameters of record stops positioning error, and its data can be by experimentizing definite to various injection moulding types.For the injection moulding type of the unregistered mold clamping force of look-up table, can obtain corresponding E by interpolation method _mValue.In the iterative learning algorithm, use variable learning gain p _k, need the final user before controller uses, import the corresponding maximum clamping force of the supporting injection machine of controller in advance.

In the real system, because the existence of interference and noise signal makes the iterative learning algorithm of formula (1) in theory no longer have complete convergent character, and can only converge within certain error margin.Simultaneously in the learning process because the accumulation of some interfering noise signal, error signal may occur descended earlier at the iterative learning initial stage, the phenomenon that rises again with the increase of iterative learning number of times then, therefore be to guarantee the performance of iterative learning algorithm and improve robustness, can be by an error margin δ be set, when the die sinking positioning error less than this scope δ, promptly stop iterative learning.In addition, learning gain can not be too big as can be known according to learning algorithm stability condition (3), therefore in formula (4), needs to add limiter (being designated as proj ()), with the size of restriction learning gain.Learning gain behind adding error margin and the limiter is:

$p_k=\left\{\begin{array}{cc}0.15+\mathrm{proj}\left(\frac{\left|e\left(k\right)\right|}{E_m}\right)& \left|e\left(k\right)\right|>\mathrm{\&delta;}\\ 0& \left|e\left(k\right)\right|\&le;\mathrm{\&delta;}\end{array}\right.---\left(5\right)$

In the formula, the qualification output amplitude of limiter proj () is 1.1.

The selection of error margin δ.The random disturbance signal that taking into account system may exist, Select Error tolerance limit are that 2/3 of bearing accuracy is advisable.Through experiment repeatedly, employing formula (5) becomes the method for learning gain, the final bearing accuracy of die sinking can reach ± 0.6mm within, desirable error margin δ=0.4mm.Actual experiment shows, employing formula (5) becomes the learning gain method, average iterations 3～4 times, and the die sinking bearing accuracy can reach the error margin of setting, under the prerequisite that satisfies the die sinking bearing accuracy, the iterative learning algorithm of error convergence speed relative fixed learning gain is much improved.

Adopt the method for error margin δ, the influence that still can not eliminate original state error, system interference fully and measure problems such as noise, they not only can influence the performance of learning algorithm, even cause the learning algorithm instability.To this type of interference and noise, can suppress by in learning algorithm, adding a wave filter, thus the robustness of enhancing learning algorithm.Learning algorithm formula behind the adding wave filter is revised as:

$u(k+1)=\tilde{u}\left(k\right)+\mathrm{pe}\left(k\right),$

$\tilde{u}\left(k\right)=Q\left(z\right)u\left(k\right),$ (6)

In the formula, Q (z) is a causal filter, to preceding n time in advance controlled quentity controlled variable carry out after the filtering again substitution learning algorithm formula and estimate this control lead.Because interference and error signal generally have the characteristic of high frequency and zero-mean, and control signal generally is positioned at low-frequency range, therefore general Q (z) is designed to a low-pass filter, keeps low-frequency range, the attenuate high frequency section.Owing to the adding of Q (z), strengthened the antijamming capability of learning algorithm, but the learning algorithm of formula (6) can't converge to ideal value in theory fully, the study error can only converge within certain error range.In the practical application, split cavity stops bearing accuracy, and we also just require can satisfy application request in certain error range, and the sacrificial section performance is worth for the robustness that strengthens learning algorithm.

Wave filter Q (z) is a moving average weighting filter in essence, and original controlled quentity controlled variable is carried out smothing filtering, eliminates radio-frequency component wherein.Q (z) is typically chosen in the FIR cause and effect low-pass filter with linear phase, and conventional design method is first according to IIR low-pass filter of performance of filter index Design, obtains the coefficient of FIR low-pass filter then by the method for time-domain windowed.

For the sake of simplicity, the present invention adopts the method that adds forgetting factor γ to strengthen the robustness of iterative learning algorithm.The adding of forgetting factor can be decayed the instability frequency signal and be forgotten that it is along the low-pass first order filter on the iteration axle in essence.Because the effect of forgetting factor is also forgotten useful learning signal in the lump in the learning process, thereby is caused learning algorithm can't converge to ideal value fully in theory.We point out the front, and the robustness of control algolithm and performance often can not get both, and need take all factors into consideration between them, through checking, select for use forgetting factor γ=0.15 can reach better effects.

In conjunction with variable learning gain p _kAnd forgetting factor γ, final, the iterative learning algorithmic formula that adopts in the die sinking position fixing process is unified as follows:

$u(k+1)=\tilde{u}\left(k\right)+p_{k}e\left(k\right),$

$\tilde{u}\left(k\right)=(1-\mathrm{\&gamma;})u\left(k\right),$ (7)

In the formula, $p_k=\left\{\begin{array}{cc}0.15+\mathrm{proj}\left(\frac{\left|e\left(k\right)\right|}{E_m}\right)& \left|e\left(k\right)\right|>0.4\\ 0& \left|e\left(k\right)\right|\&le;0.4\end{array}\right.,\mathrm{\&gamma;}=0.15.$

Practical application shows that the iterative learning algorithm that the present invention proposes can significantly improve injection machine folding mould bearing accuracy.In electric-hydraulic proportion control system for shot machine based on open loop control; embed the learning algorithm that the present invention proposes, generally through behind 3 to 4 iterative learnings, can reach ± positioning accuracy of folding mould mechanism of 0.6mm; bearing accuracy can reach the needs of actual use fully, and strong robustness.Clasp mould mechanism guarantees folding mould stable action in normal operation because the raising of bearing accuracy can be avoided physical shock, and satisfies relevant injection moulding processing technology requirement.But; control learning algorithm can't be avoided the fault location that causes because of reasons such as electrical failure; such as causing measuring position drift or measuring position to have the phenomenon of saltus step because electronic ruler wears out; the display position that causes moving platen is different with physical location; thereby cause the inefficacy of control algolithm; therefore adopt iterative learning algorithm controls folding mould bearing accuracy, need while cooperative mechanical locating device to carry out hardware protection, increase the reliability of folding die positioning mechanism work.

The above; only be the embodiment among the present invention; but protection scope of the present invention is not limited thereto; anyly be familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected; all should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (10)

1, a kind of learning method that improves positioning accuracy of folding mould mechanism; it is characterized in that; learn the position lead that the action of folding mould stops by iterative learning control law; when being used to compensate folding mould action and stopping, because of the response time lag of pressure flow moderating process, hydraulic control system and the moving platen stop position error amount that causes at the inertia effect of clasp mould mechanism.

2, the method for study lead as claimed in claim 1 is characterized in that, described law of learning comprises the steps:

Step a: the stop position w of given setting _d, choose the lead u (1) of learning process for the first time;

Step b: in the positioning control of folding mould, add lead u (1) input; Sampling obtains dynamic model actual stop position w according to Linear displacement transducer simultaneously ₁, calculate output stop position error e (1) after the die sinking action is finished;

Step c: utilize iterative learning control law to learn the lead u (2) of the 2nd learning process;

Steps d: the process that repeating step b, step c are identical can obtain u (3), u (4) successively ... u (k), u (k+1) ... and corresponding e (2), e (3) ... e (k) ..., converge within the desired scope up to folding mould positioning error;

Step e: finish learning process.

3, learning method as claimed in claim 1 is characterized in that, utilizes described folding mould action repetitive cycling characteristic to learn the action of folding mould and stops the anticipated future position amount.

4, learning method as claimed in claim 2 is characterized in that, the law of learning form of the k time study lead u (k) is:

u(k+1)＝Q(z)u(k)+p _ke(k)，

Wherein, k is an iterations; E (k) is the k time folding mould stop position error; Q (z) is a low-pass filter, is used to strengthen the robustness of learning algorithm; p _kBe variable learning gain.

5, learning method as claimed in claim 4, it is characterized in that, wave filter Q (z) need be chosen as a cause and effect low-pass filter, Design of Filter is Q (z)=(1-γ), γ is a forgetting factor, its essence is the low-pass first order filter along the iteration axle, and forgetting factor γ is used to improve the robustness of learning algorithm.

6, learning method as claimed in claim 4 is characterized in that, variable learning gain p _kSatisfy formula | 1-p _kG _u|＜1, be used to improve the learning algorithm speed of convergence, become learning gain p _kSatisfy:

$p_k=\left\{\begin{array}{cc}0.15+\mathrm{proj}\left(\frac{\left|e\left(k\right)\right|}{E_m}\right)& \left|e\left(k\right)\right|>\mathrm{\&delta;}\\ 0& \left|e\left(k\right)\right|\&le;\mathrm{\&delta;}\end{array}\right.$ And | 1-p _kG _u|＜1,

E wherein _mFor the die sinking of corresponding injection machine type under typical die sinking parameter setting stops the error expectation value; δ is an error margin; $g_u=\&PartialD;g/\&PartialD;u,$ G () is the transport function between folding mould stop position w (k) and the lead u (k), and proj () is a limiter.

7, learning method as claimed in claim 6 is characterized in that, described change learning gain p _kDie sinking under typical die sinking parameter setting stops to exceed the flat expectation value E of error with corresponding injection machine type or each type maximum clamping force _kRelevant.

8, learning method as claimed in claim 7 is characterized in that, the flat expectation value E of described error _kBy setting up a look-up table, and use interpolation method and calculate, determine the E of various different type of machines _kValue is used for learning gain P _kCalculating.

9, learning method as claimed in claim 6 is characterized in that, controls learning gain by setting limiter proj () and error margin δ, is used to improve the robustness of iterative learning algorithm, and limiter is used to limit the maximal value of learning gain, guarantees | 1-pg _u|＜1 condition satisfies, and certain redundancy is arranged, and is stable in order to guarantee learning algorithm.

10, learning method as claimed in claim 6 is characterized in that, after the clasp mould mechanism positioning error reached error margin δ, learning gain was 0, promptly stops iterative learning, when the clasp mould mechanism positioning error surpasses error margin δ, begins iterative learning again.

Images