Method and device for adjusting an element

Abstract

The invention relates to a method and a device, whereby a galvanometer actuator (7) that is connected to the element and a control unit are provided. Said unit generates a control signal for the actuator for adjusting the element according to a desired position time history that can be freely predetermined, whereby the control unit detects the control signal by means of the desired position time history and by considering a model (4) consisting of the unit of the actuator and the element. Said model is a time discrete state space model and is selected in such a way that said model can predict the position time history of the element for a time history of the control signal and with predetermined exactness, whereby said time history of the control signal can be freely predetermined. The predetermined desired position time history is verified by the control unit and by means of a preliminary filter and is modified with regard to the feasibility thereof if required. A position sensor is provided which generates a position signal according to the position of the element and supplies said signal to the control unit that considers said position signal during detection of the control signal. Detection of the control signal by means of the model is only carried out at predetermined control times of the desired position time history. The value of the control signal is detected at a control time by requesting that the summed-up, weighted difference between the desired position and the position predicted by the model becomes minimal for the next control times for said value of the control signal. The thus detected value of the control signal is applied at least until the next control time.

Description

[0001] The present invention relates to a method and an apparatus for positioning a movable element, in particular a rotating mirror, according to a predetermined desired time-position history using a galvanometer actuator to which a control unit provides a control signal.

[0002] In the prior art the control unit has been designed as a PID (proportional plus integral plus derivative) controller. Such a controller determines the control signal for the element from the position signal of a position sensor, in which the deviation between the intended position and the actual measured position, the integral of such deviation, and the time derivative are used as input parameters for the controller. The disadvantage of this way of controlling is often inadequate precision in effecting rapid changes in the desired position.

[0003] U.S. Pat. No. 5,450,202 discloses a system to position a mirror with a galvanometer actuator operated in resonance and in which the control unit regulates the current through the actuator so that the mirror occupies a precisely predetermined position at control times that differ by half of the resonance period. The sequence of positions recurs cyclically, i.e., is the same in every run. The current flow between two consecutive control points is held at a constant level calculated by the control unit. The algorithm is designed to be adaptive in the sense that it records the deviation between the actual position and the predetermined desired position for every run from information provided by a position sensor and uses this deviation to calculate the value of the current for the desired position in the next run and minimize deviations. This correction is accomplished with a parameter derived from a model that considers the mirror and the actuator as damped harmonic oscillator. This model is not adaptive. The disadvantage of this system is that the control strategy only works for systems running in a resonant mode in which a certain predetermined desired position-time history is repeated cyclically. Moreover, on the one hand the position can only be specified at certain predetermined resonance times and on the other hand the system does not provide an adequately precise control for a position time history that runs only once.

[0004] Patent publication JP-A 10023777 discloses a system for positioning a mirror the control unit of which uses a motion model of the mirror actuator to calculate the control signal using a filter for the angular acceleration and the current of the motor.

[0005] Patent publication DE-A-43 44 283 discloses a control method for an electric motor that uses a model with an adaptive parameter determination.

[0006] Patent publication U.S. Ser. No. 233,512 discloses a control method for an electric motor that uses a model of the electric motor to control the error rate of the control method.

[0007] Patent publication U.S. Pat. No. 5,519,605 discloses a control method using a model based on a step response, that is, an impulse response.

[0008] The task of the present invention is to provide a method and an apparatus for positioning an element which provide the possibility of an exact freely determined desired position time history of the positionable element using minimum calculation but nevertheless being suitable for high dynamic demands.

[0009] According to this invention, this task is performed by a method according to claim 1 and a corresponding apparatus according to claim 19. The advantage of this invention is that the selected model, the selected control strategy, the provision of a preliminary filter, and if necessary the optimization of the desired position time history allow an exact realization of a predetermined desired position time history within the bounds of the system's possibilities while using little calculation effort and satisfying high dynamic requirements.

[0010] The preferred embodiments of this invention can be found in the dependent claims.

[0011] The single FIGURE (FIG. 1) shows a block diagram of a form of implementation of the control unit according to the invention.

[0012] “Galvanometer actuator” is used herein to mean all kinds of electromechanical equipment using magnetic fields produced by current flow to position a movable element either cyclically or randomly. “Position sensor” means here every kind of device which tracks such a positioning and provides a signal from which the course of this motion can be reconstructed unambiguously to the desired accuracy. The term “control” herein refers in particular to feedback control, whereby the feedback is used either in the control algorithm as one of several variables in calculating the instantaneous position of the system from measured values or as input for adjusting model parameters so as to produce a better model prediction at subsequent times or for both purposes. The term “control unit” refers especially to a digital regulator based on a model, wherein every device or system is understood to operate on the basis of a mathematical model for the galvanometer actuator, which is connected to the element to be positioned, which model is able to determine a control signal for the actuator so that a predetermined position-time history can be executed as exactly as possible, including the actuator being in specified positions. The control unit receives as input signals preferably at least the position signal provided by the position sensor. “Model” as used in this invention is to be understood as any collection of calculation rules that permits prediction of the actuator's position-time history according to an at least piecewise constant variation of the control signal within an appropriate accuracy range. Preferably this will be the discrete state space model introduced below. The accuracy range is appropriate when the control signal determined by the controller causes the actuator to attain a desired position at the intended time with the average accuracy established for the model-based control of the particular application in use at that time.

[0013] The model parameters can if necessary be adjusted later in order to guarantee conformity to reality, whereby the adaptation results from a comparison of the model prediction with the actual measured position-time history. In addition to adapting to the actual time variation of the system, the later adjustment of model parameters can also enable the system to register the non-linearity of the actuator. Accordingly, one usually first chooses for the actuator model a simple linear model in comparison to a non-linear model and then captures the non-linearity by subsequent adjustment of model parameters dependent on position and state. Proceeding this way makes for a better adaptation of the predictive capacity of the model to reality.

[0014] Moreover, the control unit applies a pre-filter to the intended position-time history that identifies a predetermined, physically impossible or impracticable position-time history according to the model and calculates a new model-based position-time history that can actually be achieved by the actuator and is as close as possible to the desired position time history. This pre-filtering is, for the selected control strategy, especially necessary when the position signal has sharp discontinuities or demands unrealistic speed changes. By means of the pre-filter the algorithm, which determines whether a given position-time history can be run exactly enough, can operate more-easily, that is, with less computational intensity, because it does not (within the present state of the art) have to produce optimal performance with arbitrary boundary conditions.

[0015] In addition to the position signal from the position sensor the control unit can use as input value the current flow inside the actuator or the actual speed of the motor shaft or both.

[0016] To improve the accuracy to which a fixed position can be achieved and held, the system can use an error integrator according to the integral segment of a PID controller, which can be either analog or digital, although an analog implementation is possibly advantageous in lowering digital noise.

[0017] The model used in this invention comprises a discrete time state space model, which produces a prediction for the actuator's behavior using input parameters.

[0018] The galvanometer actuator can be for example a so-called galvanometer scanner that has either a fixed coil and a rotating magnet or a fixed magnet and a rotating coil. The galvanometer actuator, however, can also be a linear motor with either a coil or a magnet which can be linearly displaced. The galvanometer actuator can comprise the active part of a head positioning system for hard disks, which for example, works with a movable coil and a fixed permanent magnet, while the position information is coded on the magnetic surface of the disk.

[0019] Hereafter an example of the implementation of the invention is explained in detail. In this example the element to be positioned is a rotating mirror of an optical system, which is powered by a galvanometer actuator that consists of a rotating magnet turned by the magnetic field of an electrical current driven fixed coil. The model is based on a differential equation that takes into account the electric resistance of the circuit R, the inductance L of the coil, and the voltage U applied to the coil, as positioning signals for the actuator, taking into consideration the torsion T (torque/current flow) and the moment of inertia J of the moving parts (magnet and mirror), for example in this form: 1$\begin{array}{cc}U=T\frac{d\vartheta }{dt}+L\frac{dI}{dt}+\mathrm{IR}& \left(1\right)\\ \mathrm{IT}=J\frac{d^2\vartheta }{d{t}^2}& \end{array}$

[0020] U: voltage applied to the coil

[0021] T: torsion constant

[0022] L: inductance of the electromagnet

[0023] R: circuit resistance

[0024] J: moment of inertia of the moving systems (permanent magnet, shaft, mirror)

[0025] θ: time-dependent function of the angle the mirror has to pass

[0026] I: time-dependent function of the current flow through the electromagnet

[0027] Such a differential equation can also be described as a state space model in the form:

{right arrow over (Z)} (t)=A{right arrow over (Z)}(t)+{right arrow over (b)}U(t) and X(t)={right arrow over (c)}T{right arrow over (Z)}(t)  (2)

[0028] where A is a matrix, b and c are appropriate vectors, U(t) is the input value, X(t) the output value, and Z(t) the state space vector. The parameters R, L, T, J appear generally in the matrix and in the vectors c and b. The components of the state space vector need not have an obvious physical meaning. However, it is possible to choose the matrix in such a way that the resulting state space vector is Z(t)=(X(t), V(t), I(t)), where X(t) represents the position, V(t) the velocity, and I(t) the time dependent current. The solution of such a differential equation for X(t) and a continuous voltage U(t) produces a function X(R, L, T, J, X0, V0, I0, U(t), t), which, in addition to the physical parameters R, L, T, J of the scanner uses also the starting conditions X0, V0, I0 and the input value U(t). For discrete time systems governed by a digital system the solution follows from the above mentioned state space model:

{right arrow over (Z)} (k+1)=Ad{right arrow over (Z)}(k)+{right arrow over (b)}d U(k) and X(k)={right arrow over (c)}dT{right arrow over (Z)}(k)  (3)

[0029] in which the state space vector, with an appropriate choice of Ad, uses the values of Xk, Vk, and Ik that specify the position, the speed, and the current at the time tk, which is the sampling point at the kth interval. The equation describes the transition from state Z(k) into state Z(k+1) brought about in the kth interval by the input value U(k) (which is constant in that interval) and its peculiar dynamics AdZ(k). If a specific state Z(i) and the input factors U(i), U(i+1), . . . , U(k−2), U(k−1) are known the whole dynamic of the system from Z(i) to the scanning interval k is predictable. As set out in equation (3), a solution for X(k) proceeds from the scalar product of Z(k) with the vector C. This solution for X(k), where the vector Z(k) is obtained recursively from a known state Z(j) by the application of U(i) with j≦i≦k−1, is hereafter denoted as X(R,L,T,J,Z(j),k). Its dependence on the sequence {Uk} of all system input values Uj up to the kth sample, starting with U1 is suppressed. Z(0) is the initial state and must be initialized according to the system's beginning boundary condition.

[0030] To this point no measured data, such as the actual measured position or the measured current, have been used. However, if Ad is selected in the right way the state vector Z will contain these values. Thus it is possible to replace the calculated values in part with measured values after every iteration step k according to Equation (3) in the calculated state vector Z(k+1). The present invention uses such an approach, such that a pure steering is turned into regulation. According to the single FIGURE this iteration task and the subsequent replacement of calculated with measured values is implemented by component 8.

[0031] Measured data can be filtered before use to increase the signal to noise ratio. For that purpose of course a filter can be added to the model.

[0032] In general a function Y(t) could be predetermined for the angular trajectory with time that is supposed to be run so that the actuator moves the mirror as accurately as possible. This can usually be managed by choosing U(t) in an appropriate way so that Y(t)=X(R,L,T,J,X0,V0,I0,U(t),t) is valid.

[0033] In the present state of the art, U(t) is usually produced by a PID controller that in general miscalculates as input values the error Y(t)−X(t), the error integral, and the angular velocity as well as the first derivative of the error value. In case of comparatively slow motions or reduced requirements on dynamic accuracy there are optimizing controllers based on impulse response. If the impulse response can not be restricted to a limited contribution by a few samples over long times (U.S. Pat. No. 5,519,605), the choice of this model adversely affects the performance of the controller, the model's adaptation to changing boundary conditions, and the incorporation of nonlinearities.

[0034] An object of the present invention is to provide a controller that can achieve a good sequence of desired values even under high dynamic requirements.

[0035] Therefore a controller is provided that contains an explicit model of the actuator and uses it to optimize the control voltage proceeding from the instantaneous state of the actuator and the desired future position-time history without inclusion of specified limitations for optimization.

[0036] The response times of such an actuator normally lie in the millisecond range. To guarantee only a minimal deviation from the desired value time history even during rapid changes of the signal the controller has to record the instantaneous position as often as possible and correct the control voltage possibly every 20 microseconds. For faster actuators even higher sampling rates must be contemplated.

[0037] To be able to use such high sampling rates, several modifications of conventional optimizing controllers, such as the ones that are used for process control and monitoring in the chemical industry, must be adopted.

[0038] In the preexisting state of the art, a model is usually provided having a step response or impulse response. Such a step response cannot be used in galvanometer actuators because the actuator would exceed its specifications.

[0039] Generally also an impulse response controller for a galvanometer actuator sends significant contributions at later points in time, which contributions cannot be suppressed in an exact model description. Accordingly, the usual method of limiting the impulse response to the part where it changes significantly leads still to a rather high data volume of 200 to 300 values for the highest accuracy requirements. Since these data are used in every calculation step, the allowable sampling rate declines and therefore the achievable control accuracy diminishes as well at high dynamic requirements. Further, all these values have to be corrected if one desires to incorporate slowly varying parameters (R, T, . . . ) or nonlinearities into the specification of the model.

[0040] The system presented here is different from the usual approach in that it uses a discrete state space model (3), which is able to store the actuator's entire system behavior in, for example, 6 to maximum 15 parameters.

[0041] To realize the desired value-time history with maximum accuracy it is principally necessary to actually determine the above identified function X(R, L, T, J, X0, V0, I0, U(t), t) and calculate a U(t) for a specified Y(t). In a digital control system with physically realistic curves Y (no voltage U(k) is available) can therefore only be required that the interval between desired and actual values per sample k shall be small:

|Y(tk)−X(R,L,T,J,0,V0, I0, U(tk),tk)|≦ε  (4)

[0042] The requirement (4) as set out above does not have to refer to individual available samples, but also can be understood in reference to possible over-sampling.

[0043] There are many possibilities for a solution of the requirement (4). Usually it is required that a functional of goodness of fit in the form: 2$\begin{array}{cc}J=\sum _{i=n+1}^{n+m+1}{{\alpha }_1\left(Y\left(i\right)-X\left(R,L,T,J,Z\left(n\right),i\right)\right)}^2+{\beta }_1{U\left(i\right)}^2& \left(5\right)\end{array}$

[0044] be minimized, where αi, βi, are two weighting sequences. This is an optimization exercise that can be solved through a group of procedures in addition to the least squares method.

[0045] The disadvantage of this common method for determination of the output voltage U(n) by optimization is the high calculation effort and the therefore limited suitability for use with high dynamic requirements. Accordingly, in the existing state of the art, the output voltage of controllers is usually determined by solving a Diophantine equation or a applying a method of linear optimization. This state of affairs becomes further aggravated when during optimization boundary conditions have to be maintained. Steps that seem to be useful in processes which take a minute or a second turn out to be impossible in the 10 microsecond range.

[0046] To respond to the increased scanning rate that the present invention allows, only a short interval of the future position-time history is optimized such that the interval is small compared to the step response and shall not exceed 10 samples (sampling at control time points), for example 3 to 5 samples. Explicit inclusion of the voltage values into the optimization does not take place, that is, all βi are set equal to zero. In addition to that the boundary conditions do not have to be observed.

[0047] To save even more calculation time, in the present invention only a single voltage is specified for optimization, and this voltage is taken as a constant during the optimizing process over the selected samples (future sample and control times). The optimization is carried out as if there were a constant voltage during the relevant portion of the position-time history interval. Ideally the so specified voltage is only passed on to the following sample and the algorithm is repeated at the next sample time. The result of this procedure is a discrete controller algorithm in form of a sum of two scalar products:

U=< Reg1|Sposn>+<Reg2|Zn>

[0048] Reg1 and Reg2 are two vectors, which are specified through the parameters that the galvanometer actuator is based on (R,L,T, . . . ), the system's internal values according to a specified formula. This formula flows naturally from the optimization rule. Sposn is the maximal 10-dimensional vector, which is formed by the future desired values yn+i based on the current desired value yn+i, and Zn is the current state of the system. The first scalar product is formed between two vectors having at most ten dimensions and the vectors of the second system have the dimensions of the order of the system. The vectors of a model without oscillator are 3dimensional. If an oscillator formed by the magnet, the motor shaft, and the mirror is taken into account, the whole model is 5-dimensional.

[0049] Hereafter an example is described in which the optimization is performed using only two samples, that is, the two directly following samples. It is required that the sum 3$\sum _{i=n+1}^{n+2}{\left(Y\left(i\right)-X\left(R,L,T,J,Z\left(n\right),i\right)\right)}^2={\left(Y\left(n+1\right)-c^T{A}_d\overrightarrow{Z}\left(n\right)+c^T\overrightarrow{b}U\left(n\right)\right)}^2+{\left(Y\left(n+2\right)-c^T{A}_d^2\overrightarrow{Z}\left(n\right)+c^T{A}_d\overrightarrow{b}U\left(n\right)+c^T\overrightarrow{b}U\left(n\right)\right)}^2$

[0050] be minimized, where Z(n) is provided according to item 8 in the FIGURE. An explicit controller logic is established by solving the above expression for U(n) and setting it to zero. In this way a voltage is derived that brings the movements of the scanner as closely as is possible to the two following time points that must be reached. Then the calculated voltage is preferably used only for one sample and the calculation is repeated for i=n+2 and so on, or is actually used for the calculated time and the calculation is repeated for i=n+3. As described above this method can be used for more than two, but preferably less then 10 points.

[0051] However, this efficient controller logic has a disadvantage as well. It sets the specific voltage that the system must hold in the neighborhood of the desired value sequence without any compromise. If the optimization horizon is chosen too small (*5), for example during maximum acceleration a dynamic accuracy of up to 14 bits is still possible. But the controlling law fails if a voltage for the desired position time history gets specified that is not attainable.

[0052] For this reason the desired position-time history curve has to be examined for physical feasibility very rapidly and if necessary replaced by a best fit position-time history curve which does not violate limitations.

[0053] This assurance is provided by a pre-filter, which is shown in further detail in the FIGURE. Input position-time histories are checked for their feasibility and corrected if necessary. The corrected position-time history curve is supposed to trace the desired course as closely as possible without violating any boundary conditions.

[0054] For that reason the control signal is delayed in an m+1 graduated shift register (or a cyclic buffer) 1 while the desired position-time history values in the buffer are analyzed. The actual control loop of the system, which consists of the controller 6, the galvanometer actuator 7, and the model 8, delays the desired position time history by m steps.

[0055] For analysis the complete control loop including the galvanometer actuator 7 can be simulated and checked for failures by using an input supervisor 5. For that reason a simulator that consists of a simulated controller 3, the model 4, and the input supervisor 5 receives the desired position-time history value sequence through a tap of the shift register 1 and provides states that are correlated with the desired values back to the shift register. If a violation with predetermined conditions occurs a local optimizer 2 with a certain optimizing strategy according to the violation begins. This optimizer brings about a new position time history within the shift register 1 that does not violate any limitations. It tries to pick up the predetermined position time history on both sides of the part within the shift register and to maintain it as long as possible not violating any limitations. This test takes place iteratively and delivers as a result the position time history curve that had to be replaced. The central element of the local optimizer is a matrix equation that is developed from the state space model. With this matrix equation n input values U are calculated based on an initial state and a sequence of n to-be-attained final values to be calculated from n input values situated over n variable time intervals. The variable n should be of the same magnitude as the system order. Preferably the simplified galvanometer model from equation (1) is used to obtain n equal to 3. The special advantage of determining the optimization through voltage values can be found in the fact that a few voltage values can describe curve courses over a relatively long time period of 2-3 milliseconds. If the input supervisor detects a problem the whole control process is stopped and more time for calculation is made available. This calculation time is given to the optimizer, which tries during long time intervals to pick up at desired positions at the beginning of the shift register, which were recorded long before the problem occurred but are stilt existing within the state of the shift register. If the voltage values do not reach Umax the time intervals will be shortened. In this manner the system works itself up to the optimum time interval. For this case the method of successive approximation is helpful, as it is used in analog/digital converters, because it enables the system to find the optimum time window with an accuracy of {fraction (1/256)} with only 8 tries. When the voltage values are determined after a few iterations—preferably eight—the local optimizer 2 calculates the curve segment by using the discrete state space model. This curve segment bridges the error spot and attaches to both sides of the curve course that is to be found in the shift register. The system state, which the local optimizer was using in its last optimizing step, is now used as the input state for the model. This curve production does not require an additional calculation effort, since the optimizer has already done the work. The curve course does not violate any limitations because it was constructed under these limitations. After this curve course is produced the simulator of the control loop is initialized again and runs the reconstructed curve, whereby the initial state of the pre-filter is reestablished.

[0056] The main system limitation is the maximum voltage to be used. This can be tested well during the optimizing routine. It only has to be checked if the voltage values of the actual iteration step are smaller than Umax. If that is the case, the next optimizing step follows. If not the voltage values are selected from the last optimizing step or the system refines the time interval to reach the optimum as described above.

[0057] The maximum voltage for galvanometer actuators is a global optimizing parameter. If it is lowered the maximum current flow and the specified power output are lowered as well. That means that by lowering Umax it is possible to save a galvanometer actuator from overheating without losing any of its performance capability because it happens in full compatibility with the optimizing routine. During this procedure the supervisor can determine the average specified power, since the simulated states of the galvanometer are available to it. It also determines whether Umax has to be reduced to save the galvanometer from overheating.

[0058] It is also possible to do without a complete simulation of the control loop, so the components 3, 4, and 5 can be omitted. Then the monitoring possibilities are limited, and the local optimizer 2 would have to take care of the monitoring and the optimization. In this case the results of the above mentioned matrix equation, the voltage values, have to be routed into a discrete state space model, which then calculates the state that is needed for the equation in the next step. In these circumstances calculation effort can not be saved.

[0059] Once a specific curve course is launched the local values Xi, the input values for the actuator U(i), and if necessary I(i), which are provided by the feedback routine and belong to the intervals i can be stored in memory and one can continuously adapt the parameters R,L,T, and J as well as the whole model by minimizing the expression 4$\sum _{i=n+1}^{n+m}{\left(X_i-X\left(R,L,T,J,Z\left(n\right),i\right)\right)}^2$

[0060] with only a few parameters. During this optimization the state Z(n) is either known or predetermined as unknown and then established during optimization. Thus a resistance value fluctuating because of temperature changes can be taken into account. This adaptive procedure can be established to increase the prediction accuracy of the model, but it is not necessary in every case.

[0061] The parameter T usually describes an angular dependence T=T(X). In this instance the differential equation generally becomes non-linear. Introducing T later can alleviate this problem. In this approach the model is expanded to

X=X (R,L,Ti−1,J,Z(i−1),i) with Ti−1=T((Xi+Xi−1)/2),

[0062] to name only one possibility of discretizing T(X). Generally one applies a method of treating non-linearity by including it as stepwise changes.

[0063] The model constructed by the differential equation described above does not make allowance for the naturally occurring friction, the angular dependence of the torsion, the oscillator's natural resonance (which is created by the magnet, the shaft of the mirror mounting, and the mirror) as well as effects (which result from the fact that the moving permanent magnet magnetizes its environment). Also, the position detector is assumed to be ideal, that is, that it delivers the true position information. The model can be extended in the case of higher accuracy requirements.

[0064] An extended model, which considers the oscillator's motion because of the limited stiffness of the shaft between actuator and mirror, can be described with the following differential equation: 5$=T\frac{d\vartheta }{dt}+L\frac{dI}{dt}+\mathrm{IR}$$\mathrm{IT}-K\left(\vartheta -\phi \right)=J\frac{d^2\vartheta }{d{t}^2}$$K\left(\vartheta -\phi \right)=J\frac{d^2\phi }{d{t}^2}$

[0065] where K is the stiffness of the connecting rod in regard to the rotation.

[0066] For a linear motor, which operates vertically, the following differential equation is valid: 6$=B\frac{dX}{dt}+L\frac{dI}{dt}+\mathrm{IR}$$\mathrm{IB}+\mathrm{gm}=m\frac{d^{2}X}{d{t}^2}$

[0067] In this equation:

[0068] U=voltage on the electromagnet;

[0069] B=acceleration constant

[0070] L=inductance of the electromagnet

[0071] R=resistance opposed to the current

[0072] m=mass of the moving system (permanent magnet, mounting rod, mirror)

[0073] X=time dependent function of the location the mirror is supposed to pass through

[0074] I=time dependent current flow through the electromagnet.

[0075] If the linear motor is operated horizontally the term gm is omitted in the second equation.

Claims

1. Method for positioning an element, whereby a galvanometer actuator that is connected to the element and a control unit are provided, said control unit generating a control signal for the actuator for positioning the element according to a predetermined desired position-time history, in which the control unit determines the control signal from the desired position time history by means of a model of the unit comprising the actuator and the element, said model being a discrete time state space model selected so that said model can predict the position-time history of the element for a control signal position-time history with predetermined accuracy, in which said position-time history of the control signal can be freely predetermined, the predetermined desired position time history being verified by the control unit and by means of a preliminary filter and being modified taking into account the feasibility of the desired position-time history as required, a position sensor being provided which generates a position signal according to the position of the element and supplies said signal to the control unit that considers said position signal during detection of the control signal, detection of the control signal by means of the model only being carried out at predetermined control times of the desired position time history, the value of the control signal being determined at a control time by the condition that the summed-up, weighted difference between the desired position and the position predicated by the model becomes minimal for the next control times for said value of the control signal, the thus determined value of the control signal being applied at least until the next control time.

2. The Method according to claim 1 in which the maximum subsequent ten, preferably the maximum subsequent five, but at least the subsequent two control times are used for minimization.

3. The Method according to claim 2 in which the minimization is effected using the least squares method.

4. Method according to claim 1 in which within the pr-filter a certain limited part of the desired position-time history is analyzed.

5. Method according to claim 4 in which the preliminary filter contains a shift register in which a part of the desired position time history is located, whereby the memorized part is replaced by a modified and feasible desired position-time history when the original part of the desired position time history is detected to be not feasible, and whereby the control unit is given the content of the shift register successively as desired position-time history that has to be passed during the detection of the control signal with a delay according to the length of the shift register.

6. Method according to claim 5 in which during the feasibility test of the part of the desired position time history the control signal time history is detected as necessary to realize the part of the desired position time history and the detected control signals are checked for feasibility.

7. Method according to claim 6, which is characterized by the fact that the preliminary filter covers a simulation of the control loop including the actuator to determine the control signal time history.

8. Method according to claim 6 in which control signals are determined and optimized in case of a violation of the system restrictions with respect to their feasibility and the modified desired position time history is calculated by means of the optimized control signals.

9. Method according to claim 8 in which the detected part of the desired position-time history is adjusted to both sides of the original time history within the shift register during optimization.

10. Method according to claim 8 in which the optimization procedure is repeated iteratively.

11. Method according to claim 1 in which the parameters used in the model are variable because of the compromise between the model prediction and the position signal for optimization of the prediction of the model.

12. Method according to claim 11 in which a linear model is selected and the parameters used within the model are followed-up to detect a nonlinearity of the unit between actuator and element.

13. Method according to claim 1 in which the current flowing within the actuator is measured and provided as a significant input factor of the control unit during the detection of the control signal.

14. Method according to claim 1 in which the control unit contains an integrator that integrates the difference between the desired control function and the position signal and considers the starting signal during the detection of the control factor.

15. Method according to claim 1 in which the element is a rotating mirror of an optical system and the galvanometer actuator is manufactured with a fixed coil and a rotating magnet or a fixed magnet and a rotating coil.

16. Method according to claim 1 in which the galvanometer actuator is a linear motor.

17. Method according to claim 1 in which the galvanometer actuator is the movement mechanism of a main positioning system for hard disks and the position information is coded on the disks to produce the position signal.

18. Method according to claim 1 in which the control signal is voltage driving the galvanometer actuator.

19. Device for adjusting an element, whereby a galvanometer actuator and a control unit are provided, which generates the control signal for the actuator for adjusting the element according to a desired position time history that can be freely predetermined, daid control unit detecting the control signal by means of the desired position time history and by considering a model consisting of the unit of the actuator and the element, said model being a time discrete state space model and is selected in such a way that said model can predict the position time history of the element for a time history of the control signal and with predetermined exactness, whereby said time history of the control signal can be freely predetermined, the predetermined desired position time history being verified by the control unit and by means of a pre-filter and being modified with regard to the feasibility of the predetermined desired position-time history, a position sensor being provided which generates a position signal according to the position of the element and supplies said signal to the control unit that considers said position signal during detection of the control signal, detection of the control signal by means of the model being only carried out at predetermined control times of the desired position-time history, the value of the control signal being detected at a control time by requesting that the summed-up, weighted difference between the desired position and the position predicated by the model becomes minimal for the next control times for said value of the control signal, the thus detected value of the control signal being applied at least until the next control time.