FIELD OF THE INVENTION [0001] The invention relates to a method of determining the speed of rotation of a motor and to a computer software product according to the method. The device the speed of rotation of which is to be determined is one that comprises a rotating shaft, such as an electric motor, or a similar device comprising a rotating shaft.

BACKGROUND OF THE INVENTION [0002] The invention relates to monitoring the condition of a motor.

Motor condition monitoring based on the vibration produced by the motor and a condition analysis made on the basis of the vibration is widely known in the art. There are several motor faults, such as unbalance, looseness, resonance, bearing failures, and the like, which cause abnormal vibration of the motor.

[0003] A commonly applied condition indicator is the root mean square (rms) value of a vibrational velocity signal in the region of 10-1000 Hz.

The allowed limits are given in ISO standards, such as ISO-10816. The root mean square value allows motor faults to be detected, but it cannot be used for identifying the type of fault concerned. An experienced serviceman, how- ever, is often able to analyse the spectrum of the vibration signal and thereby evaluate also the type of fault.

[0004] It is known art to carry out measurements relating to condi- tion monitoring of motors in the following way, for example. When a fault is suspected in a motor, or in connection with a routine maintenance of the mo- tor, the vibration of the motor is measured using a suitable vibration measuring sensor. The measurement is carried out by coupling a speed sensor or an ac- celeration sensor connected to a data collection means, such as a PC, to the motor, the measurement being then carried out by taking samples of the motor vibration in axial, horizontal and vertical directions at the driven end of the motor shaft and also in one direction at the opposite shaft end. The measure- ment data obtained from the measuring sensor is stored in the memory of the data collection means and processed by applying software which uses the vibration data to produce a vibration spectrum which is then analysed visually.

[0005] Depending on the speed of rotation of the motor's rotor, the direction and magnitude of the forces causing vibration in the motor vary. For this reason mechanical motor faults often cause motor vibration which has a

cycle length which is inversely proportional to the speed of rotation of the mo- tor's rotor. The faults thus cause periodic vibrations in the motor, the frequen- cies of which can be found out by spreading the vibration signal measured from the motor into frequency components. The form of the vibration signal and the periodic frequencies it comprises depend on the motor type, the speed of rotation and the type of malfunction.

[0006] For the above reasons, accurate determining of speed is of primary importance in fault analysis. In prior art measurement solutions, rota- tion speed of motors is measured using tachometer or stroboscope measure- ments.

[0007] There are, however, major drawbacks in the prior art. When a tachometer is used for measuring speed of rotation, there must be one mounted in the motor, or one must be mounted for the measurement. Corre- spondingly, when a stroboscope is used, the motor must contain the means for carrying out the measurement. Motors do not usually have built-in speed measurement devices, but the motor must be halted for mounting one. How- ever, motors used in industrial processes cannot usually be halted without causing undue harm for the process in which the motor is one component.

[0008] A fault analysis method in which measurement data ob- tained from a measuring sensor is stored in the memory of a data collection means and processed using software which produces from the vibration data a vibration spectrum for visual analysis requires a discrete Fourier transform (DFT) of the signal. In DFT the measurement time and the frequency resolu- tion (the distinction between two consecutive frequency points) are interrelated in that the better the desired resolution, the longer is the measurement time required. A long measurement time is a problem, because the motor load, and thereby its speed of rotation, should remain constant during the measurement to allow accurate and reliable measurement data to be collected. When the motor is used in an industrial process, this is not, however, usually possible without causing undue harm for the process. Secondly, a set of frequency points where the calculation is to be carried out is determined in advance in the DFT method on the basis of the measurement time and sampling fre- quency. If the speed of rotation is not exactly the same as the frequency at any of the predetermined frequency points, error will occur in the estimation of fault frequencies, which are proportional to the speed of rotation, and, conse- quently, amplitude estimates will also be erroneous.

[0009] As illustrated above, drawbacks that often appear in con- nection with the prior art include the need to halt the motor for the mounting of the speed measurement device, the need for a plural number of measurement devices and, thereby, the need to carry out various measurements to allow an analysis to be made. Moreover, an accurate analysis requires a long meas- urement time, during which a constant speed of rotation of the motor is re- quired. This naturally slows down and complicates the measurement, and im- pairs its accuracy and reliability.

BRIEF DESCRIPTION OF THE INVENTION [0010] It is an object of the invention to alleviate the drawbacks of the prior art, and to provide an improved method of determining speed of rota- tion and computer software implementing the method.

[0011] This is achieved by a method and computer software of the present invention which comprise the characteristics set forth in the claims.

More precisely, a method according to a preferred embodiment of the inven- tion is primarily characterized by what is stated in the characterizing part of claim 1.

[0012] An underlying idea of the invention is that speed of rotation of a motor is determined by measuring the mechanical vibration of the motor with a suitable measuring sensor. Measurement data is collected from the vi- bration at a suitable frequency for a predetermined measurement period. The measurement data is stored in the memory of a data carrier as measurement data of a fixed format. The speed of rotation of the motor is then determined from the measurement data by determining the cycle length of a periodic vi- bration signal in the time domain, and not in the frequency domain, by using a maximum likelihood estimate (MLE) calculated by maximizing a maximum like- lihood function (MLF) of the measurement data. The speed of rotation is thus at the frequency point where the maximum likelihood function (MLF) obtains its highest value.

[0013] Before the ML function is calculated, the vibration signal is filtered using a bandpass filter on the passband of which the rotation fre- quency estimated in advance is located. An advantage of this is that interfer- ence in the measurement data, if any, can be attenuated.

[0014] The speed of rotation is determined using the maximum like- lihood estimate of the time domain. This provides an advantage in that the

measurement time needed for determining the frequency is now significantly shorter than in the conventional DFT method. The reason for this is that in the maximum likelihood method, frequency is produced as a continuous variable and not as separate values in which the minimum difference between two fre- quency values, i. e. their resolution, would be determined by the measurement time and the sampling frequency, as in the commonly used DFT method of the frequency domain. In the maximum likelihood method the only factor having an effect on the accuracy at which the speed of rotation can be determined is the magnitude of interference in the measurement signal.

[0015] Practice has shown that for the impact of the interference that is in the measurement signals to be eliminated, the length of the meas- urement period must be more than 50 times the cycle length of the vibration signal. As a result, the required measurement time is shorter at higher rotation speeds than at lower rotation speeds. A measurement time which is about 50 times the cycle length of the vibration signal is, however, advantageously short in proportion to the variation in the speed of rotation caused by variations in the motor load. For example, motors having a speed of rotation of 25 Hz re- quire a measurement time of 2 seconds in the maximum likelihood method, whereas in DFT methods a measurement time of about 30 seconds is needed to obtain the required resolution.

[0016] For sufficiently reliable measurement results to be obtained, a sampling frequency about three times the synchronous speed of the motor is preferred.

BRIEF DESCRIPTION OF THE DRAWINGS [0017] In the following the invention will be described with reference to the accompanying drawings, in which Figure 1 is a schematic view of a measurement arrangement of the invention for determining a speed of rotation of a motor; Figure 2 illustrates a principle of determining the speed of rotation of a motor by using a maximum likelihood function adjusted to measurement data; and Figure 3 illustrates mechanical vibration measured from an unbal- anced motor with an acceleration sensor and shown in the form of a voltage signal represented as a function of time.

DETAILED DESCRIPTION OF THE INVENTION [0018] Figure 1 shows an arrangement of a preferred embodiment of the invention for measuring speed of rotation. The embodiment illustrates, by way of example, the determining of the speed of rotation of an electric mo- tor by means of a method of the invention.

[0019] The speed of rotation is measured from an electric motor 1 comprising a shaft 3. The shaft 3 of the electric motor 1 transmits rotational motion to a process to which the motor 1 is connected. The rotation of the shaft 3 of the electric motor 1 causes mechanical vibration in the motor 1, the vibration being measured by a measuring sensor 2 connected to the electric motor 1.

[0020] The measuring sensor 2 is an acceleration sensor or a speed sensor the implementation of which is based on capacitive or piezoe- lectric operation. To the measuring sensor 2 is connected a data collection means 4, such as a PC 4, or a similar device provided with a memory means suitable for data collection.

[0021] For collecting data on the mechanical vibration of the electric motor 1 for the rotation speed measurement, the measuring sensor 2 coupled to the data collection means 4, such as the PC 4, is connected to the electric motor 1. The measurement is taken in axial, horizontal and vertical directions at the driven end of the shaft 3 of the motor 1 and also in one direction at the opposite end of the shaft 3. Figure 3 illustrates mechanical vibration measured from an unbalanced motor with an acceleration sensor and shown in the form of a voltage signal represented as a function of time.

[0022] The data collection means 4 comprises computer software which is used for collecting the vibration data and for storing it in the memory of the data collection means 4 to provide processable measurement data of a fixed format. Before the measurement is initiated, the software is provided with the initial data, to be described below, needed for carrying out the measure- ment.

[0023] On the basis of the characteristics 5 of the electric motor 1, a synchronous speed fn can be determined for the electric motor 1. The de- scribed preferred embodiment assumes that the electric motor 1 rotates at a synchronous speed fn of 1500 rotations per minute, i. e. 25 rotations per sec- ond, or at a frequency of 25 Hz.

[0024] Next, the length of the measurement period T used in the data collection is set. The preferred length of the measurement period T is 50 times the synchronous speed fn of the electric motor 1. When the synchronous speed fn of the electric motor 1 is 25 Hz, a measurement period T of 2 seconds is needed.

[0025] The next step is to set a sampling frequency fs to be applied in the data collection. To provide a reliable measurement result, the sampling frequency fs must be at least three times the synchronous speed f, of the elec- tric motor 1. When the synchronous speed of the motor 1 is 25 Hz, a sampling frequency fs of at least 75 Hz is required.

[0026] The measurement is then initiated by collecting samples x [n] at the sampling frequency fi, set on the basis of the mechanical vibration of the electric motor 1, for the duration of the measurement period T. The data col- lection means 4 comprises computer software 6 for collecting the vibration data and for storing it in the memory of the data collection means 4 as processable measurement data of a fixed format.

[0027] After the measurement data have been collected and stored in the memory of the data collection means, the measurement data are proc- essed using the computer software 6 determining the speed of rotation of the electric motor 1.

[0028] The speed of rotation of the electric motor 1 is determined from the measurement data by determining the cycle length of the periodic vibration signal in the time domain by means of a maximum likelihood estimate (MLE) calculated by maximizing a maximum likelihood function J. The speed of rotation is at a frequency point where the maximum likelihood function J obtains its highest value.

[0029] The maximum likelihood function to be used for determining the speed of rotation of the electric motor 1 takes the following form: in which formula ABS = the absolute value of a complex number

J = a maximum likelihood function (MLF) to be maximized N = the number of samples n = the order number of a sample x (n) = an individual sample u = a digital frequency of a periodic vibration signal, and u=f/fs, in which formula f = the frequency of the periodic vibration signal 5 = sampling frequency.

[0030] In the following, the operation of the method will be de- scribed with reference to Figure 2 which illustrates a principle for determining the rotation speed of the motor 1 from the maximum likelihood function J ad- justed to the measurement data.

[0031] In the first step, a rotation speed search interval 21 is deter- mined, i. e. an initial frequency point fini 20 and an end point fend 21 of fre- quency. The point of origin of the search interval20, 21 is obtained on the ba- sis of the synchronous speed fn of the electric motor 1. If the synchronous speed fn of the electric motor 1 is 1500 rotations per minute, i. e. 25 Hz, then the initial point 20 of the search interval20, 21 is selected such that fini = 23 Hz and the end point 21 such that fend = 25 Hz, because the slip of the motor 1 causes the speed of rotation to be lower than the synchronous speed fn.

[0032] In the second step, the vibration signal in the measurement data is filtered with a bandpass filter, such as a FIR (Finite Impulse Response) filter of the 50t order, the coefficients of which have been calculated using an optimal Parks-McClellan algorithm (Alan V. Oppenheim, Ronald W. Scafer : "Discrete-Time Signal Processing"Prentige Hall Inc., New Jersey 1989, pp.

476-479). The search interval20, 21 is located on the passband of the band- pass filter.

[0033] In the third step, the search interval20, 21 is divided into sub-intervals 10,11,12,13,14 of an equal length such that fk = fini + k (fend-fini)/N, where k = 0,1,..., N.

[0034] As shown in the example of Figure 2, the number of sub- intervals is selected such that N = 5, whereby the end points of the sub- intervals are denoted in sub-interval 10 by points 10a = 23 Hz and 10b = 23.4 Hz, in sub-interval 11 by points 11a = 23.4 Hz and 11b = 23.8 Hz, in sub- interval 12 by points 12a = 23.8 Hz and 12b = 24.2 Hz, in sub-interval 13 by points 13a = 24.2 Hz and 13b = 24.6 Hz, and in sub-interval 14 by points 14a = 24. 6 Hz and 14b = 25 Hz.

[0035] In the fourth step, the highest value of the maximum likeli- hood function J is determined in each sub-interval : J max = max (J (fkmaX)) where k = 1, 2,..., N fkmaX E fk_, fk1 [0036] As shown in Figure 2, in the second 11, third 12 and fifth 14 sub-intervals the maximum likelihood function J obtains its highest value at end point 11a, where 2maX = 23. 4 Hz, end point 12b, where = 24. 2 Hz, and end point 14a, where 5maX = 24.6 Hz. In the first 10 and fourth 13 sub-intervals the maximum likelihood function J has a local maximum value Jkmax. In Figure 2 the local maximum values are at points 10max, where f = 23. 33 Hz and 13max, where f4maX = 24.22 Hz. The highest value of the maximum likelihood function J is searched for using for example the Brent method (William H.

Press, Brian P. Flannery, Saul A. Teukolsky, William T. Wetterling :"Numerical Recipies", Cambridge University Press 1986, pp. 283-286).

[0037] In the fifth step, the speed of rotation of the electric motor 1 is selected to be an fk, a, value 13f corresponding to the highest value 13max of maximum likelihood function Jkmax. In the example of Figure 2 the speed of ro- tation thus arrived at is value 4maX = 24. 22 Hz 13f which corresponds to the highest maximum likelihood function value J4 max 13max of the calculated fkmaX values 13f. Consequently, the actual speed of rotation of the electric motor 1, determined with the method of the invention, is at point 13f f4max = 24.22 Hz, i. e. 1453.2 rotations per minute.

[0038] It is to be understood that the above specification and the accompanying Figures are only intended to illustrate the present invention.

The invention is thus not restricted to the embodiment described above or dis- closed in the claims, but a person skilled in the art will find it apparent that di- verse variations and modifications can be made to the invention within the in- ventive idea disclosed in the accompanying claims.