Vector control refers to a manner of controlling an AC motor which allows flux linkage and torque of the motor to be controlled independently like in a DC motor. In the DC motor, direct currents influencing the flux linkage and the torque are controlled, while in the AC motor both the amplitude and the phase angle of the currents have to be controlled. Thus, current vectors are controlled, from which comes the term vector control.

To implement vector control, the flux linkage and the current of the motor have to be known. The flux linkage of the motor is generated by the ac- tion of stator and rotor currents in the inductances of the machine. In an asyn- chronous machine, the rotor current has to be estimated and the estimation requires information on rotation speed of the rotor. This requires measured or estimated rotation speed of the rotor. In the synchronous machine, a magneti- zation current independent of stator magnetization is applied to the rotor, or the rotor magnetization is implemented with permanent magnets and its influ- ence, seen from the stator, shows in the direction of the rotor position angle.

To know the flux linkage caused by the position angle, the position angle of the rotor has to be measured or estimated.

When vector control of the AC motor employs a measured rotation speed or position angle of the rotor, the control method is known as the closed-loop vector control. If the rotation speed or the position angle is esti- mated, the control method is known as the open-loop vector control. Depend- ing on the implementation method, a variable to be estimated can also be the stator flux linkage, apart from the rotor angle or angular speed.

When the synchronous machine is started by vector control, the machine's stator flux linkage has the initial value vS0 which is dependent on the rotor magnetization yf and the rotor position angle Or as follows: When voltage us influences the stator flux linkage, the stator flux linkage changes in accordance with the equation It appears from the equation that when integrating the stator flux linkage a previous value of the stator flux linkage is required, apart from the voltage and current values. Thus, to start the machine controllably, information on the initial position angle of the rotor is required. When employing the closed-loop vector control, the initial angle is measured, whereas when em- ploying the open-loop vector control, the initial angle has to be defined by es- timation. When the rotor rotates, the rotor flux linkage generates an electro- motive force which can be utilized in vector control in a normal operating situa- tion, but at a rotor standstill there is no electromotive force.

In a salient pole synchronous machine, such as a separately ex- cited synchronous machine or one with permanent magnet magnetization or in a synchronous reluctance machine, the stator inductance L, in stationary co- ordinates varies as a function of the rotor angle Or as presented in the following equation Ls = Lso + Ls2 Cos2Br and Figure 1 shows a graphic illustration of the equation. It appears from the figure that the inductance varies around the basic value Lso at twice the rotor angle in a magnitude indicated by the inductance coefficient L, 2. The inductance coefficients Lso and Ls2 are defined as follows where the inductances Lso and Lsq are the direct-axis and quadrature-axis tran- sient inductances of the synchronous machine.

To utilize the above equation for defining the initial angle of the rotor is known per se and it is set forth, for instance, in the articles by S. Ostlund and M. Brokemper"Sensorless rotor-position detection from zero to rated speed for an integrated PM synchronous motor drive", IEEE Transactions on Industry Applications, vol. 32, pp. 1158-1165, September/October 1996, and by M. Schroedl"Operation of the permanent magnet synchronous machine

without a mechanical sensor", Int. Conf. On Power Electronics and Variable Speed Drives, pp. 51-55,1990.

According to the article by M. Leksell, L. Harnefors and H.-P. Nee "Machine design considerations for sensorless control of PM motors", Pro- ceedings of the International Conference on Electrical Machines ICEM'98, pp.

619-624,1998, sinusoidally altering voltage is supplie to a stator in the as- sumed direct-axis direction of the rotor. If this results in a quadrature-axis cur- rent in the assumed rotor coordinates, the assumed rotor coordinates are cor- rected such that the quadrature-axis current disappears. The reference states that a switching frequency of the frequency converter supplying the synchro- nous machine should be at least ten times the frequency of supply voltage.

Thus, the supply voltage maximum frequency of a frequency converter capa- ble of 5 to 10 kHz switching frequency, for instance, is between 500 and 1000 Hz. This is sufficient for an algorithm to function. Switching frequencies as high as this are achieved by IGBT frequency converters, but frequency converters with GTO or IGCT power switches, required at higher powers, have the maxi- mum switching frequency of less than 1 kHz. The maximum frequency of the supply voltage in the initial angle estimation remains then below 100 Hz. At such a low frequency the machine develops torque and the algorithm be- comes considerably less accurate.

In the reference by M. Schroedl, 1990, the initial angle is calculated directly from one inductance measurement, or, if more measurements are em- ployed, the additional information is utilized by eliminating the inductance pa- rameters. A drawback with the method is that an error, which is inevitable in measuring, has a great influence. One example of actual inductance meas- urement with a permanent magnent machine at rotor angles Or = [0,..., 27ruz is shown in Figure 2. The figure shows theoretically great deviations from the sine curve. The inductance measurement is effected such that a stator is fed with a current impulse which causes flux linkage on the basis of which the in- ductance is calculated. Errors may arise from an error in current measurement or from the fact that the measuring current produces torque that swings the rotor.

From the inductance expression in the stationary coordinates it is possible to derive an expression for a rotor angle B = 1 arccos LS Lso + n 2 lez where n is an integer. The influence of the error in the measured Ls can be studied by differentiating 0, with respect to Ls :

This allows calculating an error estimate for the angle It is observed that It is observed from the above that, when the inductance difference between the direct-axis and quadrature-axis directions is small, the error estimate of the angle approaches infinite. In other words, initial rotor angle definition based on inductance measurings becomes the more unreliable, the closer to one an- other the magnitudes of the direct-axis and quadrature-axis inductances of the rotor.

In the method presented in the reference by S. Ostlund and M. Brokemper the rotor angle is not calculated directly, but the minimum inductance is searched by starting the measuring of inductances in different directions first at long in- tervals and when approaching the minimum by reducing the angular difference of successive measurings. Even though it is not mentioned in the article, the method easily catches fictitious minima resulting from measuring errors, and therefore, an error value may be extremely high.

On the basis of the above, the influence of the inductance measur- ing errors should be reduced somehow. One method could be to employ sev- erai measurings in each direction and to calculate the average from the meas- ured inductances, yet this procedure does not eliminate the influence of a systematic error.

BRIEF DESCRIPTION OF THE INVENTION The object of the present invention is to provide a method which avoids the above-mentioned disadvantages and enables starting of an open- loop vector control in a synchronous machine in a reliable manner. This is achieved by the method of the invention which is characterized by comprising the steps of arranging measured stator inductances in a determined stator in- ductance model in order to form model parameters giving the minimum error, checking magnetization polarity of a rotor in order to verify the direction of the rotor magnetization, initializing flux linkages of the open-loop vector control according to the formed model parameters and the direction of the rotor mag- netization, and starting the synchronous machine by the vector control method.

The method of the invention is based on the idea that the magni- tude of the stator inductance is measured in a plurality of directions and the inductance values obtained as measurement results are arranged in the in- ductance model of the machine. As a result of the arrangement, very accurate information on the initial rotor angle of the synchronous machine is obtained.

In addition, by utilizing the method of the invention, information on the initial value of the rotor magnetization in the stationary coordinates is obtained, and consequently the machine can be started in a reliable manner without tran- sients or jerking startup.

BRIEF DESCRIPTION OF THE DRAWINGS In the following the invention will be described in greater detail in connection with preferred embodiments with reference to the attached draw- ings, wherein Figure 1 is a graph of a stator inductance Ls as a function of an an- gle; Figure 2 shows measured magnitudes of stator inductance as a function of an angle; Figure 3 shows the relations between the coordinates ; Figure 4 shows the order of stator inductance measurings; and Figure 5 is a flow chart of the starting procedure in accordance with the invention.

DETAILED DESCRIPTION OF THE INVENTION According to the invention, starting of an open-loop vector control in a synchronous machine first comprises the step of determining a stator in- ductance model of the synchronous machine L, = Lso + Ls2cos2dr. The equation shows how the inductance depends on the rotor angle in stationary coordi- nates. The equation thus proves how the inductance measured in the direction of x-axis of the stator coordinates changes when the rotor is rotated a degree of an angle 0,.

When it is desired to determine the initial rotor angle, it is not possi- ble to rotate the rotor at different rotor position angles for measuring the in- ductance. Instead, the stator coordinates are rotated to the angle, in the direc- tion of which the inductance is measured. These rotatable stator coordinates are referred to as virtual stator coordinates. Figure 3 illustrates various coordi- nates and they are indicated such that the stator coordinates are coordinates xy, the virtual stator coordinates are coordinates x'y'and the rotor coordinates are dq.

Then the inductance model in the virtual stator coordinates may be written as where Or is the rotor angle in the stationary coordinates, A is an angle in the virtual coordinates and 0,-A is the rotor angle in the virtual coordinates. By introducing a parameter ç =-20ru it is possible to simplify the equations under study. So the desired rotor angle in the stationary coordinates is given by <BR> <BR> <BR> <BR> (<BR> <BR> <BR> <BR> <BR> 2 With reference to the flow chart of Figure 5, in accordance with the invention, after initializing 2 the inductance measuring the stator inductance of the synchronous machine is measured 3 in a plurality of directions. The stator inductance measurement is advantageously effected such that the stator is fed with a voltage pulse generating a current pulse which causes flux linkage on the basis of which the inductance can be calculated. The stator inductance

measurement can be performed, for instance, in six directions. These six volt- age vectors can readily be implemented with inverters, since the vectors cor- respond to inverter switch combinations, in which a positive and a negative voltage is generated between each two poles of the three-phase stator. Figure 4 illustrates six current vectors and their preferable mutual order in connection with measuring. For improved accuracy, a plurality of measurings can be per- formed in each direction. In order to reduce the possibility that the rotor turns, the measurings are performed in the order shown in Figure 4, i. e. measuring starts in the direction of 0°, thereafter in the direction of 180°, thereafter in the direction of 60° and so on, until all the directions are measured. Figure 2 shows measured values of the stator inductance with different rotor angle val- ues. In the flow chart of Figure 5, it is checked 4 after the inductance meas- urement 3 if the inductances of all directions determined to be measured are measured, and if not, a new direction is initiale 5 for measuring the induc- tance in accordance with Figure 4.

According to the invention, the measured stator inductances are ar- ranged 6 in the determined stator inductance model in order to form model parameters giving the minimum error. Measurement data is preferably ar- ranged in the model by using the method of least mean squares (LMS). In the LMS method, a model is formed for a measurable variable, in which model the data to be measured is arranged such that the square sum is minimized. The square sum refers to a sum of squares of the difference of the measured val- ues and corresponding model values. If the model is linear, the parameters of the model can be solved in a closed form, but in a nonlinear case the question is about a numerically solvable nonlinear optimization task.

In a nonlinear LMS method the following function is minimized where m is the number of samples (measurements), fj=yj-M (tj, a) is the difference of a nonlinear model M and the measured data (tj, yj), are the model parameters, is a vector formed by functions f"

(,) refers to an inner product.

A gradient of the function F (a) needed in a numerical solution is VF (a) = 2 J (a) f (a) where J (a) is the Jacobian matrix of the function f (a).

In the method, the inductance model is indicated by M and its pa- rameters by a,, a2and a3 M(t,a)+#+a1+a2cos(st+a3)+#,y= wherey = L, is an inductance measured in virtual coordinates, t =/I is an angle in the virtual coordinates, i. e. a direction of measuring in the stationary coordinates, g is a term for measuring error and is a parameter vector of the model.

To solve the parameters of the model, it is possible to use a known conjugate gradient method. However, implementation of the method by com- monly employed DSP processors is difficult, so it is worth while utiiizing the knowledge that where inductances Lsd and Lsq are the direct-axis and the quadrature-axis tran- sient inductances.

The inductance model is simplifie by assuming that the transient inductances Lsd and Lsq are previously known. The transient inductances can be measured when introducing the machine by turning the rotor first in a di- rect-axis position supplying direct current to the stator. After the rotor is turned in the direct-axis position, the stator is supplie in the direct-axis and the quad- rature-axis directions with step-like voltage pulses that cause currents on the basis of which the inductances are calculated. Another option is to use the starting procedure such that, instead of solving the angle giving the minimum error, parameters Lgo and Ls2 giving the minimum error are solved.

A simplifie inductance model is given by y = M (t, a) + c = Lao + Ls2 cos (2t + a) + E.

Only parameter a remains to be solved, and consequently the algo- rithm used for the solution can be simplified considerably. The above- mentioned known gradient needed for the solution is then simplifie to a com- mon derivative according to the following equation fj = yj-M (t"a) being the difference of the model M and data (tj, yj) and m being the number of data.

Minimization of a function with one variable is concerned, which can be implemented simply by a processor, for instance, by reducing the value of the target function M with a given step and by selecting a which produces the minimum value. In said simplifie case, the a producing the minimum value is the solution to the initial angle, i. e. the information on the angular position of the synchronous machine rotor.

Since the stator inductance is a function of twice the rotor angle, the above solution is only a candidate for a rotor angle. It is also possible that the solution is said candidate plus 180°. Thus it is not known for sure, whether the direction found is the north or the south pole of rotor magnetization or the permanent magnet.

To find out the polarity, the invention utilizes the fact that the rotor magnetization saturates the direct-axis inductance. On counter magnetization with a stator current, flux density decreases and saturation decreases, on for- ward magnetization vice versa. Thus, the direct-axis inductance has different magnitudes in the direction of 0° and 180°. After finding a solution candidate, the polarity of the rotor magnetization can be found out by measuring the in- ductance once more in the direction 8 of the solution candidate and in the di-

rection 10 of 180° therefrom. The inductances of both directions having been measured 9, the lower one of these inductances is selected 11,12,13 to be the correct rotor angle.

If the synchronous machine used is a synchronous reluctance ma- chine 7, there is no need to find out the polarity of the machine poles due to the structure of the machine.

The above-described inductance models do not take the saturation of the direct-axis inductance into account, so the polarity of the permanent magnet has to be found out separately. However, it is possible to take the po- larity into account in the model, as a result of which it is possible to find out the position angle of the rotor without any separate step of polarity checking. By assuming that the direct-axis inductance changes linearly as a function of the direct-axis current, it is possible to write for the direct-axis inductance Lsd = Lsdo + k i5d where k is the slope of the inductance and Lsd0 is its value, when isd = 0. By supplying current into the stator in the x-axis direction of the stator coordi- nates, whereby i5y = 0, said current in the rotor coordinates is given by isd = isd cose.

The stator inductance is then In the above-described manner, a replacement #r # #-#r is made, which re- sults in an improved inductance model taking the saturation into account y = M, a) + = a1 + a2 cos(t + a4/3)+(a3 + a2 cos(t + a4/3))cos(2t + a3) +, where parameters an are Depending on the number of previously known parameters, the solution can be simplifie in the above-described manner.

When the direction of the rotor is found out in accordance with the invention, flux linkages are initialized 14 according to the rotor direction. The objective of the initialization is to adapt the inverter control systems to the con- ditions in the synchronous machine. After initialization, the synchronous ma- chine can be started in a reliable manner by utilizing any known open-loop vector control method.

It is obvious to a person skilled in the art that as technology pro- gresses the basic idea of the invention can be implemented in a variety of ways. Thus, the invention and its embosdiments are not restricted to the above-described examples but they may vary within the scope of the claims.