Method for detecting electric machine fault during transient operation

Abstract

The invention relates to a device and a method and a device for detecting a fault of an electric machine during transient operation of the electric machine. The invention: senses at least one motor signal, identifies a fundamental frequency of the motor signal, the fundamental frequency being a frequency of a voltage waveform driving a stator of the motor, performs heterodyne incoherent demodulation of the motor signal over a frequency band, a center frequency of the frequency band and a bandwidth of the frequency band having a fixed proportionality coefficient with the identified fundamental frequency, -determining a level of a fault in the electric machine by comparing the demodulated signal with a signal derived from the motor signal.

Description

Technical Field

[0001] The present invention generally relates to a method for detecting electric machine faults during transient operation. Background Art

[0002] Several motor faults may occur in rotating machines, for example, when the stator and rotor rotation axes are misaligned, when bearings are damaged, when rotor bars in the stator are broken, or in case of winding insulation damage, etc. Such faults lead to wear of the motor bearings, unexpected oscillations in the generated torque, and undesirable vibrations, which may damage the equipment.

[0003] In the prior art, motor faults can be detected using vibration sensors or current sensors or a combination of both. Each motor fault has a specific frequency signature that can be detected from the measured current waveform. MCSA (Motor Current Signal Analysis) technology is very popular because it can reuse the current sensors required for closed-loop control of the motor. The level and frequency of the fault harmonics will vary with the rotation speed, applied torque, and the type and level of the fault.

[0004] In an asynchronous machine, the rotor frequency is different from the stator frequency, and this difference (rotor slip) is caused by the application of torque. Since many applications are not equipped with a position sensor, it is difficult to estimate the rotor frequency and / or rotor slip.

[0005] The popular MCSA method uses a time-frequency transform (e.g., Fourier, wavelet, ...) of the input (e.g., phase current) waveform to estimate the presence of a motor fault. The Fourier method decomposes the input waveform in a base of equally spaced gyrators. The SFFT or STFT method can track the evolution of the gyrator components over time in both steady-state and transient applications.

[0006] The prior art suffers from a number of problems. The FFT method requires the acquisition, storage and processing of waveforms of relatively long duration (eg 10s for 0.1 Hz accuracy). This processing is then hardly compatible with the implementation of a motor controller.

[0007] The measured waveform includes a strong component at the stator frequency. This basic signal is superimposed on the fault characteristic signal and is several orders of magnitude larger than the fault characteristic signal.

[0008] The fault frequency varies with the rotor frequency, which is usually unknown due to rotor slip. Although an FLL (frequency lock loop) can be used to estimate the frequency of the fault signature, the FLL is more likely to lock on a strong fundamental signal rather than on the fault signature, and the FLL has a tendency to easily make locking errors, especially for noisy signals. The machine may also have more than one fault type.

[0009] This technique cannot detect motor faults during transient operation of the machine. In many applications, the operating conditions (speed and torque) are not static and cannot be maintained for a long time. Summary of the invention

[0010] The present invention aims to provide a method for detecting a motor fault during transient operation of the motor, characterized in that the method comprises the following steps:

[0011] - sensing at least one motor signal,

[0012] - identifying the fundamental frequency of the motor signal, which is the frequency of the voltage waveform driving the stator of the motor,

[0013] - performing heterodyne incoherent demodulation of the motor signal over a frequency band whose center frequency and whose bandwidth have a fixed proportionality factor to the identified fundamental frequency,

[0014] - determining the level of a fault in the motor by comparing the demodulated signal with a signal derived from the motor signal.

[0015] The invention also relates to a device for detecting a fault in an electric machine during transient operation of the electric machine, characterized in that the device comprises:

[0016] - means for sensing at least one motor signal,

[0017] - means for identifying the fundamental frequency of the motor signal, said fundamental frequency being the frequency of the voltage waveform driving the stator of said motor,

[0018] - means for performing heterodyne incoherent demodulation of said motor signal over a frequency band, the centre frequency of said frequency band and the bandwidth of said frequency band having a fixed proportionality factor to the identified fundamental frequency, and

[0019] - means for determining a fault level in said motor by comparing the demodulated signal with a signal derived from said motor signal.

[0020] Therefore, as long as the fault characteristic signal has a frequency included in the frequency band, the fault level can be estimated. The fault level (such as the eccentricity level for an eccentricity fault) can be determined based on the level of the detected fault characteristic signal. Even when the exact frequency of the fault characteristic signal is not completely known (for example, due to the presence of a slip between the stator frequency and the rotor frequency in an induction machine), the level of the fault characteristic signal can be estimated.

[0021] Even without implementing complex time-frequency signal analysis (such as SFFT), the level of the fault characteristic signal can be estimated. Detection of motor faults can be achieved with low complexity and can be implemented in a general inverter without additional sensors and / or computing power.

[0022] The fault level is estimated in real time across continuous signal inputs. The method does not require storage of data over a large analysis window. The fault level is estimated in a short time and is robust to rapid changes in the operating conditions (torque and speed) of the machine.

[0023] Since the frequency of the motor fault is always determined as a weighted sum of the stator frequency and the rotor frequency, the process is simple and effective for all rotor speeds. The ratio of bandwidth to stator frequency can be selected so that the modification of the rotor frequency due to slip always keeps the fault signature within the selected frequency band. When the selected frequency band is indexed to the stator frequency, the method can track the fault level during transient speed conditions.

[0024] This approach is effective in applications where long-term steady-state conditions are rarely or never met.

[0025] According to particular features, the non-coherent demodulation comprises the following steps:

[0026] - mixing the motor signal with a sine waveform and a cosine waveform oscillating at the center frequency,

[0027] - low pass filtering the mixed signal using a cut-off frequency equal to the bandwidth of said frequency band, and

[0028] - Combine the low pass filtered signals.

[0029] Therefore, as long as the frequency of the fault characteristic signal belongs to the frequency band, that is, as long as the distance from the frequency of the fault characteristic signal to the center frequency is less than the cutoff frequency, another reference signal located at the center frequency can be used to demodulate the fault characteristic signal. There is no need to estimate the exact frequency of the characteristic fault signal. The method does not require a systematic search across all frequencies such as in the time-frequency analysis method. It does not require an estimate of the rotation frequency, reducing the risk and cost associated with estimating or sensing the rotor position.

[0030] Because energy is collected in both the in-phase and quadrature components of the reference signal, the level of the fault signature signal is estimated without energy loss or unwanted energy oscillations.

[0031] Interfering signals, such as the second-order harmonics generated by a mixer, are suppressed as long as their frequency location is far from the frequency band. A fixed ratio can be set strictly to suppress all unwanted harmonics that may interfere with the fault estimation. Different types of faults can be distinguished because they are located in different segments of the frequency spectrum.

[0032] According to particular features, the method further comprises the following steps, performed before the non-coherent demodulation:

[0033] - performing a notch filter on the motor signal, wherein the center frequency of the notch filter is equal to the identified base frequency, the width of the notch filter has a fixed proportional coefficient to the base frequency, and the center frequency of the frequency band is outside the frequency band of the notch filter.

[0034] Therefore, when the frequency band is adjacent to a strong current signal (such as current flowing at the motor fundamental frequency), the demodulation within the frequency band is robust to the detection of weak fault characteristic signals. Cross-channel interference is minimized, and the method can detect weak fault levels. The method is suitable for condition monitoring of motors, where the condition of the machine is regularly monitored from an initial zero fault condition to a severe fault condition until a severe fault condition, before which predictive maintenance can be planned to repair the fault, thereby without the risk of suddenly stopping the application in the presence of an accident caused by a severe fault.

[0035] According to particular features, for a given fault type, the center frequency of the frequency band, the bandwidth of the frequency band and the proportionality coefficient of the center frequency of the notch filter to the base frequency are predetermined.

[0036] Therefore, by simply replicating the method using a different set of scaling factors, several fault types can be monitored at once. The fault signature exhibits harmonics at the locations where an integer number of stator frequencies and an integer number of rotor frequencies are combined. Although the rotor frequency is usually unknown, it differs from the stator frequency only by the frequency slip, which can have an upper limit. In practice, the maximum slip is related to the maximum torque of the motor. As a result, the frequency range in which the fault harmonics are located can be strictly determined.

[0037] According to particular features, the method also comprises the following steps, performed before heterodyne incoherent demodulation and / or notch filtering:

[0038] - angularly resampling the motor signal, the resampled signal having equal phase increments of the signal rotating at the identified fundamental frequency.

[0039] Thus, the number of signal samples to be processed is reduced and the numerical complexity of the method is reduced. Because the center frequency and bandwidth of the frequency band have a fixed proportionality factor to the fundamental frequency and because the signal is resampled at equal phase increments of the signal rotated at the fundamental frequency, the subsequent processing for detecting motor faults becomes independent of the fundamental frequency and thus robust to changes in the stator frequency. Therefore, the method is suitable for tracking fault levels during transient conditions.

[0040] According to particular features, the notch filtering and the low-pass filtering are implemented as IIR filters with fixed predetermined coefficients.

[0041] Therefore, the implementation of the notch filter and the low pass filter has very low complexity. IIR filtering requires minimal storage of past motor signal samples (typically only 4 samples at most, much less than the SFFT method which requires several thousand samples).

[0042] Because the motor signal is resampled with equal phase increments and the filter bandwidth and cutoff frequency are proportional to the speed level, the IIR filter implementation can be achieved with constant internal coefficients even if the fundamental frequency evolves, for example under transient conditions (such as speed ramps).

[0043] The coefficients of the notch filter and the low pass filter do not need to adapt to evolving speed conditions. Sudden changes in the IIR coefficients may cause the output produced by the filter to become unstable, which conflicts with the goal of being able to track fault conditions with good accuracy under rapidly changing operating conditions. In contrast, the method of implementing an IIR with fixed coefficients is optimized to track the level of motor fault characteristics during speed transients. According to a particular feature, the method further includes the steps of:

[0044] - performing envelope detection on the sensed motor signal,

[0045] - determining a proportionality factor between the level of the fault harmonics and the level of the fault in the motor based on the envelope and the fundamental frequency,

[0046] - performing angular resampling of said scaling factor,

[0047] - notch filtering of the resampled scale factor,

[0048] - Low pass filtering the notch filtered resampled scale factors.

[0049] Therefore, the fault level is determined from the demodulated level of the fault signature signal. The evolving fault level can indicate to the application user how to set the time to trigger preventive maintenance in order to repair the fault level. As an example, an eccentricity fault can gradually grow, reducing the minimum distance between the rotor and the stator. A technician can be dispatched to the site to repair the eccentricity problem before the eccentricity is too high and brings the potential risk of rotor-stator collision, which may have catastrophic consequences to the electromechanical system chain.

[0050] Since the proportionality factor passes through exactly the same signal processing chain (notch filter + LPF filter), it will experience the same delay from the processing chain and the final result is more sensitive to changes in the signal envelope (torque) or fundamental frequency. The estimation of the fault level is robust to fast transient conditions.

[0051] The proportionality factor compensates for the fault harmonics as a function of the torque level and speed level.

[0052] The characteristics of the invention will emerge more clearly on reading the following description of an exemplary embodiment, which is produced with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] [ Figure 1 ]

[0054] Figure 1 A first example of a device for detecting a fault in an electric machine during transient operation according to the invention is shown.

[0055] [ Figure 2 ]

[0056] Figure 2 An example of a block diagram showing a heterodyne demodulation module used in the present invention.

[0057] [ Figure 3 ]

[0058] Figure 3 An example of a block diagram showing an infinite impulse response filter used in accordance with the present invention.

[0059] [ Figure 4 ]

[0060] Figure 4 An example of a block diagram showing a phase module of an apparatus for detecting motor faults in transient operation according to the present invention.

[0061] [ Figure 5 ]

[0062] Figure 5 A second example of a device for detecting a motor fault during transient operation according to the invention is shown.

[0063] [ Figure 6 ]

[0064] Figure 6 An example of an algorithm for detecting motor faults during transient operation according to the present invention is shown.

[0065] [ Figure 7a ]

[0066] Figure 7a A first example of spectral processing of a motor signal in the presence of a motor fault according to the invention is shown.

[0067] [ Figure 7b ]

[0068] Figure 7b A second example of spectral processing of a motor signal in the presence of a motor fault according to the invention is shown. DETAILED DESCRIPTION

[0069] Figure 1 A first example of a device for detecting a fault in an electric machine during transient operation according to the invention is shown.

[0070] The device for detecting motor faults during transient operation comprises: means 103 for identifying a fundamental frequency fs of a motor signal, the fundamental frequency being the frequency of a voltage waveform driving a stator of the motor; means for performing heterodyne incoherent demodulation of the motor signal over a frequency band, the centre frequency of the frequency band and the bandwidth of the frequency band being proportional to the identified fundamental frequency; and means for determining the level of a fault in the motor by comparing the demodulated signal with a signal derived from the motor signal.

[0071] The motor signal is, for example, a current waveform flowing in one stator winding of the motor. In another example, the device includes an inverter and a controller, and the motor signal is a reference voltage determined by the controller and used to control the inverter that feeds power to the motor. In yet another example, the device includes a vibration sensor, and the motor signal is a sensed vibration signal sensed by the vibration sensor.

[0072] The apparatus for detecting motor faults during transient operation may also include angle resampling of the motor signal and an infinite impulse response filter. The apparatus for detecting motor faults during transient operation may also include a notch filter.

[0073] The apparatus for detecting motor faults during transient operation may also include a scoring function that scales the level of fault harmonics detected by the incoherent heterodyne demodulation to a fault level of the motor using a ratio appropriate to the transient speed and current conditions experienced by the motor.

[0074] The scoring function includes an envelope detection module 101 that determines the norm ‖x‖(kT) of the motor signal x(kT). The norm ‖x‖(kT) is provided to a fault harmonic model module 102 that determines a proportionality coefficient R(kT) between the fault harmonic level and the fault level in the motor.

[0075] The scoring function may also include angular resampling 108 of the motor signal, a notch filter 110 and an infinite impulse response filter 113 .

[0076] The basic detection module 103 determines the angular frequency f of the motor signal x(kT) s (kT).

[0077] Angular frequency f s (kT) is provided to the fault harmonic model module 102 and the phase determination module 104 .

[0078] refer to Figure 4 The phase determination module 104 is disclosed in greater detail.

[0079] In the first embodiment of the present invention, the phase determination module 104 provides the down-sampling ratio N to the angle resampling modules 105 , 106 and 108 , and provides the phase signal θ(kT) to the angle resampling module 106 .

[0080] In the second embodiment of the present invention, the phase determination module 104 provides a phase signal θ(kT) to the angle resampling modules 105 , 106 and 108 .

[0081] The concept of angle resampling is to produce a set of samples with equal phase distance from samples with equal time distance. Together with the fixed ratios between base frequency, filter bandwidth, filter cutoff frequency, angle resampling enables notch and LPF filters to be implemented as IIR filters with constant coefficients even at varying base frequencies, for example in the presence of speed ramps.

[0082] In a first embodiment of the invention, the angular resampling comprises downsampling the input signal using a ratio N which is inversely proportional to the fundamental frequency.

[0083]

[0084] In a second embodiment of the invention, the angular resampling consists in selecting the samples of the input signal corresponding to the times at which the determined phase angle crosses horizontal steps spaced apart by Δθ.

[0085] As an example, Δθ is chosen to represent a value between 1° and 20°.

[0086] The angle resampling module 105 provides the angle samples x(nΔθ) to the notch filter 109 .

[0087] The filtered angle samples x′(nΔθ) are provided to the heterodyne incoherent demodulation module 100 .

[0088] The angle resampling module 106 provides the angle samples nΔθ to the heterodyne incoherent demodulation module 100 .

[0089] refer to Figure 2 The heterodyne non-coherent demodulation module 100 is described in more detail.

[0090] The angle resampling module 108 provides the angle samples R(nθΔ) to the notch filter 110 .

[0091] The output of the notch filter 110 is provided to a low pass filtering module 113 .

[0092] The apparatus for detecting motor faults during transient operation of the motor comprises a normalized notch ratio determination module 107 which provides the ratio x between the bandwidth and the fundamental frequency of the notch filters to the notch filters 109 and 110. n .

[0093] The apparatus for detecting motor faults during transient operation comprises a low pass filter frequency determination module 112 which provides a ratio x between the cut-off frequency of the low pass filter and the fundamental frequency to a low pass filter 113 and a heterodyne non-coherent demodulation module 100. c .

[0094] The apparatus for detecting motor faults during transient operation comprises a demodulation frequency determination module 111 which provides a ratio x between the demodulation frequency and the base frequency to the heterodyne non-coherent demodulation module 100. m .

[0095] The ratio triple (x n ,x c ,x m ) is chosen so that the expected fault harmonic is located within the search window f fault / f fundamental [x m -x c / 2;x m +x c / 2]. As an example, when the fault is an eccentric fault and x m =0.5,x c = 0.1, since the related lower sideband (fs-fr) fault frequency is located in the window f fault / f fundamental[0.4 0.6], thus making it possible to detect an eccentricity fault as long as the frequency slip is below 10%. Since the harmonics (fs-fr) are always located in the selected demodulation frequency band, the demodulation is effective for any frequency slip. Therefore, the method is robust to changes in the rotational speed fr caused by changes in the torque applied to the motor by the load.

[0096] As another example, when x m =1.5,x c = 0.1, since the relevant upper sideband (fs+fr) fault frequency is located in the window f fault / f fundamental [1.4 1.6], therefore, as long as the frequency slip rate is less than 10%, the eccentricity fault can be detected.

[0097] As another example, when x m =3.5,x c =0.1, enabling detection of upper sideband fault harmonics around the third fundamental harmonic. More generally, a triplets of ratios can be predetermined for any frequency signature associated with each type of motor fault, and the method is equally applicable to detection of any fault.

[0098] The apparatus for detecting motor faults during transient operation comprises a score determination module 114 which determines an estimated level of motor fault by comparing a demodulated signal from a heterodyne incoherent demodulation module 110 and a signal from the output of a low pass filter 113 derived from a signal provided by the motor.

[0099] The level of fault harmonics varies with speed and current. The model can convert the fault harmonics H fault_est The level of is associated with the level of speed fs and current I. As an example, the estimated level of motor fault is given by the formula H fault_est =A*Fault*fs^B*I^C, where A, B and C are constant values.

[0100] As another example, H fault_est =A*Fault(1+B*fs), where A and B are constant values.

[0101] In the presence of a fast speed ramp, the demodulated signal H is fault There is some delay compared to the transient speed and current.

[0102] According to the present invention, the model ratio K=H fault_est / Fault passes through a signal processing chain that simulates the delay caused by the signal processing chain. This includes angle resampling 108, notch filter 110 and LPF filter 113. The resulting ratio is then used to compare the output H of the heterodyne demodulation. fault Perform division operation to generate fault estimation Fault_est = H fault / K.

[0103] Figure 2 An example of a block diagram showing a heterodyne demodulation module used in the present invention.

[0104] The non-coherent heterodyne demodulation module 100 includes three multiplication modules 200 , 201 , 204 , a cosine determination module 202 , a sine termination module 204 , two low-pass filters 205 and 206 , and an I&Q combiner 207 .

[0105] The multiplication module 200 converts the modulation frequency ratio x m The result of the multiplication module 200 is provided to the cosine determination module 202, the output of which is provided to the multiplication module 201. The result of the multiplication module 200 is further provided to the sine determination module 204, the output of which is provided to the multiplication module 203.

[0106] The multiplication module 201 is an in-phase mixer that multiplies the filtered angle samples x′(nθΔ) by the output of the cosine determination module 202 .

[0107] The output of the multiplication module 201 is provided to a low pass filter 205 .

[0108] The multiplication module 203 is a quadrature mixer that multiplies the filtered angle samples x′(nθΔ) by the output of the sine determination module 204 .

[0109] The output of the multiplication module 203 is provided to a low pass filter 206 .

[0110] The low-pass filters 205 and 206 preferably have a cut-off frequency x c Infinite impulse response filter.

[0111] The outputs of the low-pass filters 205 and 206 represent the frequency band (x m + / -x c / 2)*fs. These outputs are provided to the I&Q combiner 207. The I&Q combiner 207 outputs the square root value of the sum of the square values ​​of the in-phase channel and the quadrature channel.

[0112] The output of the I&Q combiner 207 is provided to the scoring module 114 .

[0113] The ratio x between the notch filter bandwidth fn, the mixer frequency fm, the LPF frequency fc, and the stator frequency fs s ,x m ,x c is constant and, as an example, for a machine with two pole pairs is given by:

[0114] x m =0.5

[0115] x c =1 / 12 and x n =1 / 12.

[0116] Other ratios are possible and can be selected to detect frequency components of various failure modes such as static / dynamic eccentricity faults, broken rotor bars, etc. The characteristic signal is located at the frequency component of [fm-fc, fm+fc] = fs*[x m -x c ,x m +x c ] defined in the frequency band.

[0117] The level of the fault feature is usually small compared to the fundamental signal. The notch filter first reduces all components in the frequency band fn around the stator frequency. The mixer shifts all frequency components to the left so that the corresponding frequency component fm is at DC. The LPF reduces all signals outside the demodulation bandwidth fc, so that the remaining fundamental signal and higher-order harmonics disappear. The I&Q combiner sums the energy of the in-phase channel and the orthogonal channel to eliminate the phase uncertainty of the unknown signal located in the desired frequency band. The I&Q combiner estimates the level of single-tone harmonic signals whose frequencies are contained in the frequency band [fm-fc, fm+fc].

[0118] Figure 3 An example of a block diagram showing an infinite impulse response filter used in accordance with the present invention.

[0119] The IIR filter is implemented using a set of delay lines 300, 301, 302, 316, 317, 318, amplifiers 303, 304, 305, 306, 313, 314, 315 and summers (307, 308, 309, 310, 311, 312). Its complexity is very small, so it can be easily implemented in software or hardware. It can filter the input signal in real time with minimal storage of past signal events. The filter coefficient b k applied to the input, and the coefficient a k Applied to output.

[0120] The filter coefficients are determined to achieve a desired spectral response. As an example, an IIR filter may be determined from an analog filter transfer function by bilinear approximation or impulse invariant transform.

[0121] For a first-order notch filter, the desired response function is

[0122] and the coefficients are given by a bilinear transformation where

[0123]

[0124] b1=a1

[0125] Where K = 2F s / ω s ,α=ω n / ω s .

[0126] F s / ω s is constant due to resampling, and due to ω n / ω s is chosen to be constant, so that if the speed changes the filter coefficients do not need to be recalculated.

[0127] For a 3rd order Butterworth LPF filter, the desired response function is and the coefficients are given by the bilinear transformation as,

[0128]

[0129] Where K = 2F s / ω c ,F s is the sampling frequency.

[0130] F s / ω s is constant due to resampling, and due to x c =ω c / ω s is chosen to be constant, so that if the speed changes the filter coefficients do not need to be recalculated.

[0131] In a preferred embodiment, F s and ω c are all selected to be normalized relative to the stator speed, that is, F s =1 / Δθ,ω c =f c / f s In the case of an IIR filter where the speed changes between consecutive samples are very fast, this will result in even smaller changes.

[0132] Filters 109, 110, 113, 205 and 206 are infinite impulse response filters.

[0133] Figure 3 The IIR filter shown is an mth order filter.

[0134] like Figure 3 As shown, the input IN of the IIR filter is provided to the multiplication module 303 which multiplies the input samples by the coefficient b0 and to the inverse Z-transformation module 300 .

[0135] The output of the inverse Z transform module 300 is provided to the multiplication module 304 which multiplies the sample by the coefficient b1 and to the inverse Z transform module 301 .

[0136] The output of the inverse Z transform module 301 is provided to the multiplication module 305 which multiplies the sample by the coefficient b2 and the mth inverse Z transform module 302 .

[0137] The output of the inverse Z transform module 302 is provided to multiply the samples by the coefficient b M The multiplication module 306.

[0138] The output of the mth multiplication module 306 is added to the output of the multiplication module 305 by the summation module 309 .

[0139] The output of summing module 309 is added to the output of multiplying module 304 by summing module 308 .

[0140] The output of summing module 308 is added to the output of multiplying module 303 by summing module 307 .

[0141] The output of summing module 307 is provided to summing module 310 which provides filtered output samples OUT.

[0142] The output of the summation module 307 is provided to the inverse Z-transform module 316 .

[0143] The output of the inverse Z transform module 316 is provided to the multiplication module 313 which multiplies the sample by a coefficient of −a1 and to the inverse Z transform module 317 .

[0144] The output of the inverse Z transform module 317 is provided to the multiplication module 314 and the inverse Z transform module 318 which multiplies the sample by a coefficient of -a2.

[0145] The output of the inverse Z transform module 318 is provided to multiply the samples by a factor -a m The multiplication module 315.

[0146] The output of the mth multiplication module 315 is added to the output of the multiplication module 314 by the summation module 312 .

[0147] The output of the summing module 312 is added to the output of the multiplying module 313 by the summing module 311 .

[0148] The output of summing module 311 is added to the output of summing module 307 by summing module 310 .

[0149] Figure 4 An example of a block diagram showing a phase module of an apparatus for detecting motor faults in transient operation according to the present invention.

[0150] The phase module 104 includes an amplifier 400, an integrator 401 that derives the angle from the angular frequency ω(kT), and an integrator 402 that derives the angle from the input frequency f s Determine the downsampling ratio N as The sampling ratio determination module 402.

[0151] Figure 5 A second example of a device for detecting a motor fault during transient operation according to the invention is shown.

[0152] The device 50 for detecting a motor fault during transient operation of the motor has, for example, components connected by a bus 501 and by means of a Figure 6 The architecture of the program-controlled processor 500 disclosed in FIG.

[0153] The bus 501 connects the processor 500 to a read-only memory ROM 502 , a random access memory RAM 503 , and an input / output I / O IF interface 505 .

[0154] The input-output I / O IF interface 505 enables the device 50 to sense motor signals that may contain spectral signatures of motor faults.

[0155] Memory 503 contains a memory device for receiving and Figure 6 The disclosed algorithms are related to the program variables and instruction registers.

[0156] Read-only memory or possibly flash memory 502 contains Figure 6 The instructions of the program related to the algorithm disclosed in the ROM memory 502 are loaded into the random access memory 503 when the device 50 is powered on. Alternatively, the program can also be executed directly from the ROM memory 502.

[0157] The computations performed by device 50 may be implemented in software by execution of a set of instructions or a program by a programmable computing machine, such as a PC (personal computer), a DSP (digital signal processor), or a microcontroller; or in hardware by a machine or dedicated components, such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0158] In other words, the device 50 includes circuits or devices including circuits that enable the device 50 to perform the same Figure 6 Programs related to the algorithms disclosed in .

[0159] Figure 6 An example of an algorithm for detecting motor faults during transient operation according to the present invention is shown.

[0160] The present algorithm is disclosed in one example where it is executed by a processor 500 of a device 50 .

[0161] In a first step S600, the processor 500 identifies the ratios xn, xm, xc between the bandwidth fn of the notch filter, the demodulation frequency fm, the demodulated bandwidth fc and the fundamental frequency of the motor signal. For a given fault type, the ratios are constant. As an example, monitoring the upper sideband harmonics around the first harmonic of an eccentricity fault corresponds to (xn, xm, xc) = (1 / 12, 3 / 2, 1 / 12). Then, the processor 500 moves to step S601.

[0162] At step S601 , the processor 500 senses a motor signal from the I / O interface 505 and moves to step S602 .

[0163] In step S602, the processor 500 identifies the fundamental frequency fs of the motor signal sensed in step S601. The fundamental frequency is the frequency of the voltage waveform driving the stator of the motor.

[0164] The motor signal is, for example, a current waveform flowing in one stator winding of the motor. In another example, the motor signal is a reference voltage determined by a controller and used to control an inverter that feeds power to the motor. The motor signal may also be a sensed vibration signal sensed by a vibration sensor.

[0165] At next step S603, the processor 500 determines the phase by integrating the fundamental frequency determined in step S602.

[0166] In a first embodiment of the invention, the phase determination will provide a downsampling ratio N and a phase signal θ(kT),

[0167]

[0168] In a second embodiment of the invention, the phase determination provides a phase signal θ(kT).

[0169] At step S604 , the processor 500 performs angular resampling on the motor signal x(kT) to provide signal samples x(nΔθ) using regular angular sampling.

[0170] At step S605 , the processor 500 performs angular resampling of the phase signal, which provides phase samples nΔθ using regular angular sampling.

[0171] In step S606, the processor 500 performs filtering of the resampled signal x(nΔθ) provided by the angle resampling step S604. The filtering is as described in reference Figure 1 Notch filtering disclosed in .

[0172] In step S607, the processor 500 uses the phase samples nΔθ, the ratio x m and x c Heterodyne non-coherent demodulation is performed on the filtered resampled signal x'(nΔθ) provided by the filtering step S606. The processor 500 multiplies the phase sample nΔθ by the ratio xm and calculates the sine and cosine functions of the multiplied phase. The processor 500 mixes the filtered resampled motor signal with the results of the sine and cosine functions. The processor 500 applies an IIR low-pass filtering step to the obtained in-phase signal and quadrature signal. The processor 500 generates the demodulation result as the square root of the sum of the squares of the IIR filter output.

[0173] In step S608, the processor 500 performs envelope detection to determine the norm ‖x‖(kT) of the motor signal x(kT). In a preferred variant, the motor signal is a current flowing in one phase of the motor, the current flowing in another phase of the motor is also sensed in step S601, and the norm is determined as the square root of the sum of the squares of the currents flowing in all phases.

[0174] In step S609, the processor 500 determines a fault harmonic model based on the norm ‖x‖(kT) and the fundamental frequency fs(kT). The processor 500 determines a proportionality coefficient R(kT) between the fault harmonic level and the fault level in the motor. In a first embodiment of the present invention, R(kT)=A*fs(kT)^B*‖x‖(kT)^C, where A, B and C are constant values. In a second embodiment of the present invention, R(kT)=A(1+B*fs(kT)), where A and B are constant values.

[0175] At step S610, the processor 500 performs angular resampling on the scaling coefficient R(kT) to obtain scaled samples R(nθΔ) using regular angular sampling.

[0176] At step S611, the processor 500 performs filtering of the sample R(nθΔ). The filtering is as described in reference Figure 1 The filtering is the same as the filtering performed in step S606.

[0177] At step S612, the processor 500 performs a filtering on the samples provided by the filtering step S610. The filtering is a low pass filtering identical to the filtering performed at step S607.

[0178] In step S612, as shown in reference Figure 1 As disclosed in , the processor 500 determines an estimated level of motor fault by comparing the demodulated signal obtained in step S607 and a signal derived from the signal provided by the motor, the signal from the output of filtering step S612.

[0179] Then, the processor 500 moves back to step S601.

[0180] In a variant, the level of motor fault estimated in step S612 is averaged over consecutive signal steps.

[0181] In a variant, steps S608 to S613 are only performed when the fundamental frequency identified in step S602 exceeds a predetermined level. As an example, the predetermined level is equal to the maximum rotation speed of the motor divided by two.

[0182] Figure 7a A first example of spectral processing of a motor signal in the presence of a motor fault according to the invention is shown.

[0183] The abscissa axis (f / fs) represents the signal frequency normalized to the fundamental frequency, while the ordinate axis (level) represents the level of the signal frequency component. The motor signal contains a strong fundamental frequency component at the abscissa 1 and a weaker fault characteristic signal at the abscissa 1-fr / fs.

[0184] The figure also shows the frequency response of the notch filter, which is centered around the strong fundamental frequency component and has a normalized suppression band xn=fn / fs. The notch filter is designed to suppress all frequency components that are within the suppression band. The filter band determines the reaction time of the notch filter. When away from the notch band, the notch filter has a unit frequency response and leaves other frequency components of the signal unaffected. The coefficient xn is selected so that the fault characteristic frequency remains in the unaffected area.

[0185] The figure also shows the position of the normalized center frequency xm and the frequency band in which incoherent heterodyne demodulation is applied according to the invention. The frequency band includes frequencies between (xm-xc)*fs and (xm+xc)*fs.

[0186] According to the present invention, as long as the frequency of the fault characteristic belongs to the frequency band, the incoherent heterodyne demodulation outputs the level of the fault characteristic frequency component.

[0187] By implementing a notch filter, despite the presence of a low-pass filter with a normalized cutoff frequency xc, interference caused by strong adjacent fundamental frequency components present near the frequency band that would occur if there were no notch filter can be suppressed. In this first example, the fault signature represents an eccentricity fault, and the coefficients (xn, xm, xc) are selected to detect the lower sideband harmonics generated by the eccentricity fault. The motor has 2 pole pairs, a positive torque is applied by the load, and the fault signature signal has a frequency 1-fs / fr located to the right of the center frequency xm=l / 2.

[0188] Figure 7b A second example of spectral processing of a motor signal in the presence of a motor fault according to the invention is shown.

[0189] In this second example, the fault signature indicates an eccentricity fault. The coefficients (xn, xm, xc) are selected to detect the upper sideband harmonics generated by the eccentricity fault. The motor has 2 pole pairs, a negative torque is applied by the load, and the fault signature signal has a frequency 1+fs / fr located to the right of the center frequency xm=3 / 2.

Claims

1. A method for detecting a motor fault during transient operation of the motor, characterized in that The method comprises the following steps: - sensing at least one motor signal, - identifying the fundamental frequency of the motor signal, which is the frequency of the voltage waveform driving the stator of the motor, - performing heterodyne incoherent demodulation of the motor signal over a frequency band whose center frequency and whose bandwidth have a fixed proportionality factor to the identified fundamental frequency, - determining the level of a fault in the motor by comparing the demodulated signal with a signal derived from the motor signal.

2. The method according to claim 1, characterized in that: The non-coherent demodulation comprises the following steps: - mixing the motor signal with a sine waveform and a cosine waveform oscillating at the center frequency, - low pass filtering the mixed signal using a cut-off frequency equal to the bandwidth of said frequency band, - Combine the low pass filtered signals.

3. The method according to claim 2, characterized in that The method further comprises the following steps performed before the non-coherent demodulation: -Performing a notch filter on the motor signal, the center frequency of the notch filter is equal to the identified base frequency, the width of the notch has a fixed proportional coefficient to the base frequency, and the center frequency of the frequency band is outside the frequency band of the notch filter.

4. The method according to any one of claims 1 to 3, characterized in that For a given fault type, the center frequency of the frequency band, the bandwidth of the frequency band, and the proportionality coefficient of the center frequency of the notch filter to the base frequency are predetermined.

5. The method according to any one of claims 1 to 3, characterized in that The method further comprises the following steps performed before the heterodyne incoherent demodulation and / or notch filtering: - angularly resampling the motor signal, the resampled signal having equal phase increments of the signal rotating at the identified fundamental frequency.

6. The method according to claim 5, characterized in that The notch filtering and low-pass filtering are implemented as IIR filters with fixed predetermined coefficients.

7. The method according to claim 5 or 6, characterized in that: The method further comprises the following steps: - performing envelope detection on the sensed motor signal, - determining a proportionality factor between the level of the fault harmonics and the level of the fault in the motor based on the envelope and the fundamental frequency, - performing angular resampling of said scaling factor, - notch filtering of the resampled scale factor, - Low pass filtering the notch filtered resampled scale factors.

8. A device for detecting motor faults during transient operation of a motor, characterized in that The device comprises: - means for sensing at least one motor signal, - means for identifying the fundamental frequency of the motor signal, said fundamental frequency being the frequency of the voltage waveform driving the stator of said motor, - means for performing heterodyne incoherent demodulation of said electromotor signal over a frequency band, the centre frequency of said frequency band and the bandwidth of said frequency band having a fixed proportionality factor with the identified fundamental frequency, - means for determining the level of a fault in said motor by comparing the demodulated signal with a signal derived from said motor signal.

Images