Defect Detection

Abstract

A method of detecting defects in a mechanical system, the method comprising the steps of: a. providing a mechanical system; b. subjecting the mechanical system to random, optionally broadband, vibration by a vibration device to cause the mechanical system to vibrate and output an output vibration spectrum; c. detecting the output vibration spectrum using a vibration detection device; d. using a processing system to carry out the substeps of: i. selecting a plurality of frequencies within the output vibration spectrum; ii. analysing the plurality of frequencies to extract phase information for the plurality of frequencies; iii. generating a continuous phase waveform representing modulation in phase over time for one or more frequencies of the plurality of frequencies; and iv. detecting peaks in the spectrum of the continuous phase waveform at multiples of the input vibration frequency to produce output data representing defects in the mechanical system.

Claims

1. A method of detecting defects in a mechanical system, the method comprising the steps of: a. providing a mechanical system; b. subjecting the mechanical system to random, optionally broadband, vibration by a vibration device to cause the mechanical system to vibrate and output an output vibration spectrum; c. detecting the output vibration spectrum using a vibration detection device; d. using a processing system to carry out the substeps of: i. selecting a plurality of frequencies within the output vibration spectrum; ii. analysing the plurality of frequencies to extract phase information for the plurality of frequencies; iii. generating a continuous phase waveform representing modulation in phase over time for one or more frequencies of the plurality of frequencies; and iv. detecting peaks in the spectrum of the continuous phase waveform at multiples of the input vibration frequency to produce output data representing defects in the mechanical system.

2. The method according to claim 1 wherein the output vibration is measured over a continuous measurement time period (T) and is expressed as the relationship between vibration amplitude and time.

3. The method according to claim 2 wherein the continuous measurement time period (T) is at least 8.5 seconds.

4. The method according to any one of claims 1 (3-wherein in substep (i) a plurality of groups of vibration measurements are selected, each group having a common time period, wherein successive groups partly overlap in time and are shifted in time by a time shift (dt.sub.2).

5. The method according to claim 1 wherein in substep (ii) each group of vibration measurements is subjected to a first Fourier Transform analysis to generate from each group of vibration measurements a respective vibration spectrum.

6. The method according to claim 5 wherein the sample duration in the first Fourier Transform analysis is shorter than wavelength of the given periodic phase modulation in substep (iii).

7. The method according to claim 1 wherein in substep (ii) phase information from each frequency of the respective vibration spectrums is extracted.

8. The method according to claim 7 the extracted phase information is wrapped within the range of −π to +π, and in substep (iii) the extracted phase information is unwrapped to provide a respective continuous phase waveform, representing changes in phase over time, for each group of frequencies in the vibration measurements.

9. The method according to claim 8 wherein each continuous phase waveform comprises only the positive or only the negative phase values.

10. The method according to claim 8 wherein in substep (iii) the continuous phase waveforms are combined, and then analysed to provide a change in phase with respect to time at a single frequency.

11. The method according to claim 10 wherein in substep (iii) the combined continuous phase waveforms are analysed at a sample rate (Fs.sub.2) of 1/dt.sub.2, where the time shift (dt.sub.2) determines the bandwidth of an output phase modulation spectrum.

12. The method according to claim 10 wherein in substep (iii) the continuous phase waveforms are combined to form a matrix M*(N/2) where M is the number of rows and (N/2) is the number of columns, each column corresponding to a respective phase modulation time series at a respective frequency.

13. The method according to claim 12 wherein in substep (iii) at least one of the columns is subjected to a second Fourier Transform analysis to determine a phase modulation spectrum at the respective frequency from a magnitude of the unwrapped phase values, wherein the phase modulation spectrum comprises the continuous phase waveform generated in substep (iii).

14. The method according to claim 13 wherein in substep (iii) the phase information in the column that is subjected to the second Fourier Transform analysis comprises only the positive or only the negative frequency values.

15. The method according to claim 1 wherein the mechanical system comprises an axle of a vehicle.

16. The method according to claim 15 wherein the mechanical system comprises a wheelset assembly of a railway vehicle, the wheelset assembly comprising an axle mounted between opposed wheels, each wheel being fitted to a respective opposite end of the axle.

17. The method according to claim 16 wherein the railway vehicle is a locomotive, a passenger carriage or a freight car or truck.

18. The method according to claim 15 wherein steps (b) to (d) are carried out while the vehicle is in motion.

19. The method according to claim 18 wherein steps (b) to (d) are carried out in real-time.

20. The method according to claim 15 wherein the input vibration frequency is a rotation frequency of the axle.

21. The method according to claim 15 wherein steps (c) and (d) are carried out using a wireless sensor node fitted to the mechanical system, the wireless sensor node comprising a vibration energy harvester for converting mechanical energy from vibration in the mechanical system into electrical energy, a sensor for measuring a parameter, wherein the sensor is an accelerometer mounted to an end of the axle, and a wireless transmitter for wirelessly transmitting the measured parameter or data associated therewith.

22. The method according to claim 1 wherein the method detects asymmetric defects in the mechanical system.

23. The method according to claim 22 wherein the asymmetric defects are cracks in the mechanical system.

24. An apparatus for detecting defects in a mechanical system, the apparatus comprising a vibration detection device for detecting an output vibration spectrum of the mechanical system which has, in use, been subjected to random, optionally broadband, vibration to cause the mechanical system to vibrate and output the output vibration spectrum; and a processing system comprising a frequency selection module for selecting a plurality of frequencies within the output vibration spectrum; an analysis module for analysing the plurality of frequencies to extract phase information for the plurality of frequencies; a generating module for generating a continuous phase waveform representing modulation in phase over time for one or more frequencies of the plurality of frequencies; and a detection module for detecting peaks in the spectrum of the continuous phase waveform at multiples of the input vibration frequency to produce output data representing defects in the mechanical system.

25. TheAs apparatus according to claim 24 wherein the vibration detection device is configured to measure the output vibration, expressed as the relationship between vibration amplitude and time, over a continuous measurement time period (T).

26. TheAn apparatus according to claim 24 wherein the frequency selection module is configured to select a plurality of groups of vibration measurements, each group having a common time period, wherein successive groups partly overlap in time and are shifted in time by a time shift (dt.sub.2).

27. Thes apparatus according to claim 24 wherein the analysis module is configured to subject each group of vibration measurements to a first Fourier Transform analysis to generate from each group of vibration measurements a respective vibration spectrum.

28. The apparatus according to claim 27 wherein the analysis module is configured to extract phase information from each frequency of the respective vibration spectrums.

29. The apparatus according to claim 28 wherein the generating module is configured to unwrap the extracted phase information to provide a respective continuous phase waveform, representing changes in phase over time.

30. The apparatus according to claim 29 wherein the generating module is configured to combine the continuous phase waveforms and then analyse the combined continuous phase waveforms to provide a change in phase with respect to time at a single frequency.

31. The apparatus according to claim 30 wherein the generating module is configured to analyse the combined continuous phase waveforms at a sample rate (Fs.sub.2) of 1/dt.sub.2.

32. The apparatus according to claim 30 wherein the generating module is configured to combine the continuous phase waveforms to form a matrix (T/S)2 where M is the number of rows and (T/dt.sub.2)/2 is the number of columns, each column corresponding to a respective phase modulation time series at a respective frequency.

33. The apparatus according to claim 32 wherein the generating module is configured to subject at least one of the columns to a second Fourier Transform analysis to determine a phase modulation spectrum at the respective frequency from a magnitude of the phase values, wherein the phase modulation spectrum comprises the continuous phase waveform generated by the generating module.

Description

[0027] Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0028] FIG. 1 is a schematic end view of an apparatus for detecting a crack in an axle of a wheelset assembly of a railway vehicle according to an embodiment of the present invention;

[0029] FIG. 2 is a schematic view of the processing system in the apparatus of FIG. 1;

[0030] FIG. 3 is a schematic process flowchart showing the steps taken in the method of detecting a defect in a mechanical system, for example using the apparatus illustrate in FIGS. 1 and 2;

[0031] FIGS. 4, 5 and 6 show the relationship, illustrated as a two-dimensional modulation spectrum, between the modulation depth (Y-axis) and the modulation frequency (X-axis) of, respectively, an uncracked axle, a heavily cracked axle and an axle with a small crack, measured in accordance with the preferred embodiment of the present invention; and

[0032] FIGS. 7 and 8 show the relationship, illustrated as a three-dimensional modulation spectrum, between the modulation depth (Z-axis), the frequency being modulated (Y-axis) and the modulation frequency (X-axis) of, respectively, an uncracked axle and a cracked axle, measured in accordance with the preferred embodiment of the present invention.

[0033] Referring to FIGS. 1 and 2, there is shown an apparatus 2 for detecting cracks in a component 4 of a wheelset assembly 6 of a railway vehicle 8. The railway vehicle 8 may be a locomotive, a passenger carriage or a freight car or truck. The wheelset assembly 6 comprises an axle 4, which is the component to be monitored, mounted between opposed wheels 10, 12, each wheel 10, 12 being fitted to a respective opposite end 14, 16 of the axle 4. In use, the wheels 10, 12 run on respective rails 50, 52 of a railway track 54.

[0034] Although the embodiment of the present invention is described with reference to an axle, the present invention may be used to detect defects such as cracks in any mechanical device using the principle of the method of the present invention.

[0035] A wireless sensor node 18 is fitted to the wheelset assembly 6. In the illustrated embodiment, the wireless sensor node 18 comprises a vibration energy harvester 20 for converting mechanical energy from vibration in the wheelset assembly 6 into electrical energy. A sensor 22 is provided for measuring a parameter, and in particular the sensor 22 is an accelerometer mounted to an end 14, 16 of the axle 4. A wireless transmitter 24 is provided for wirelessly transmitting the measured parameter or data associated therewith to a remote location for further processing and/or analysis; the remote location may be within the railway vehicle 8 which includes the tested wheelset assembly 6, or within a locomotive or other vehicle of a train which includes the wheelset assembly 6. Typically, each wheelset assembly 6 within a train is provided with a monitoring apparatus as described herein.

[0036] Preferably, as illustrated, the apparatus 2 comprises two of the wireless sensor nodes 18. Each wireless sensor node 18 is fitted to a respective opposite end 14, 16 of the axle 4, and each wireless sensor node 18 comprises a respective sensor 22 which is an accelerometer mounted to a respective end 14, 16 of the axle 4. The sensor 22 is a vibration detection device.

[0037] The apparatus 2 further comprises a processor 26 for processing the measured parameter to produce processed data. In the illustrated embodiment, the processor 26 is integral with the wireless sensor node 18, and the wireless transmitter 24 is arranged wirelessly to transmit the processed data. However, in alternative embodiments, the processor 26 is remote from the wireless sensor node 18, and the wireless transmitter 24 is arranged wirelessly to transmit the measured data which is then remotely processed by the processer 26 to produce the processed data.

[0038] The sensor 22 and processor 26 are arranged respectively to measure and process an axle vibration, in particular the axle vibration in the form of resonant vibration along the axle. The axle vibration is typically percussion driven vibration. The resonant vibration of the axle is typically within a frequency range of from 50 to 2000 Hz, more typically from 50 to 1750 Hz.

[0039] The sensor 22 measures the percussion vibration (at a sample rate S), and from that measurement a waveform is determined as described below. The determined waveform is dependent upon the axle condition and the axle load.

[0040] Accordingly, the apparatus of the illustrated embodiment comprises a vibration detection device, in the form of sensor 22, for detecting an output vibration spectrum of the mechanical system, in the illustrate embodiment axle 4, which has, in use, been subjected to random, optionally broadband, vibration by a vibration device, for example the interface between the wheelset and the track, to cause the mechanical system to vibrate and output the output vibration spectrum. The vibration detection device is configured to measure the output vibration spectrum, expressed as the relationship between vibration amplitude and time, over a continuous measurement time period (T), for example at least 10 seconds.

[0041] The processor 26 comprises a processing system which itself is comprised of a plurality of functional modules that may be programmed in hardware and/or in software. The processor 26 comprises a frequency selection module 28 for selecting a plurality of frequencies within the output vibration spectrum. The frequency selection module 28 is configured to select a plurality of groups of vibration measurements, each group having a common time period, wherein successive groups partly overlap in time and are shifted in time by a time shift (dt.sub.2). Each group may then have a common windowing function applied. Using the Fourier Transform computed for a number of discrete frequencies, the vibration detection device is configured to compute the output vibration spectrum at a sample rate (S) for each group. The successive groups selected by the frequency selection module 28 partly overlap to provide a series of overlapped output frequency spectra.

[0042] The processor 26 further comprises an analysis module 30 for analysing the plurality of frequencies to extract phase information for the plurality of frequencies. The analysis module is also configured to unwrap the extracted phase information to result in the continuous unbound phase as a function of frequency. The processor 26 further comprises a generating module 32 for generating a continuous phase waveform representing changes in the extracted phase over time for a given frequency of the plurality of frequencies, at a sample rate (Fs2) of 1/dt2. In the illustrated embodiment, the generating module 32 is configured to combine the continuous phase waveforms at different frequencies to form a matrix of dimensions M*(N/2) where M is the number of windowed time segments T/dt2 and N is the number of frequencies selected in the original output spectrum . The generating module 32 is furthermore configured to subject at least one of the (N/2) columns to a second Fourier Transform analysis to determine a phase modulation spectrum at the respective frequency from a magnitude of the phase values, wherein the phase modulation spectrum comprises the continuous phase waveform generated by the generating module 32.

[0043] The processor 26 further comprises a detection module 34 for detecting peaks in the continuous phase waveform at multiples of the input vibration frequency to produce output data representing defects in the mechanical system.

[0044] Typically, the wireless sensor node 18 may be adapted to be operated continuously over a monitoring period thereby continuously to measure the axle vibration and continuously to produce output data representing defects in the mechanical system, in particular cracks in the axle. The output data can be employed to provide an indication of axle condition in real-time and during service of the wheelset assembly 6.

[0045] In the preferred embodiments, the processor 26 further includes a load calculator module 36 which comprises a comparison module 38 arranged to compare a frequency of the axle percussion induced resonant vibration against a predetermined reference frequency value associated with an axle load and a calculation module 40 arranged to calculate a load on the axle based on the comparison.

[0046] These components permit an axle percussion induced resonant vibration frequency value to be continuously measured and compared against a calibrated reference value to provide an indication of axle load in real-time and during service of the wheelset assembly 6.

[0047] The apparatus 2 is used in a method of detecting defects in a mechanical system, in particular the axle 4, of the wheelset assembly 6, and the method may be for monitoring the mechanical system either periodically or continuously.

[0048] In the method, a wireless sensor node 18 as described above is fitted to the wheelset assembly 6, so that the sensor 22, in particular the accelerometer, is mounted to an end 14, 16 of the axle 4. As described above, in the preferred embodiment there are two wireless sensor nodes 18 each fitted to the wheelset assembly 6, so that each respective sensor 22 is mounted to a respective end 14, 16 of the axle 4.

[0049] While the railway vehicle is in motion, the vibration energy harvester 20 receives input vibration energy which is converted into electrical energy to power the wireless transmitter 24. When the processor 26 is integrated into the wireless sensor node 18 the vibration energy harvester 20 can provide the electrical energy to operate the processor 26. The vibration energy harvester 20 can provide the electrical energy to operate any other powered components of the wireless sensor node 18.

[0050] Also while the railway vehicle is in motion, during an in-service period, the axle vibration is measured using the sensor 22 and the measured axle vibration or data associated therewith is wirelessly transmitted using the wireless transmitter 24.

[0051] In the method of detecting defects in a mechanical system according to the present invention, a mechanical system is provided, such as the axle 4, and the mechanical system is subjected to random, optionally broadband, vibration by a vibration device such as by the wheel/track interface acting on the rotating axle 4, to cause the mechanical system to vibrate and output an output vibration spectrum. The input vibration frequency may be a rotation frequency of the axle 4.

[0052] The output vibration spectrum is detected by using the sensor 22 functioning as the vibration detection device. The output vibration spectrum is measured over a continuous measurement time period (T) and is expressed as the relationship between vibration amplitude and time. The continuous measurement time period (T) is at least 8.5 seconds. The output vibration spectrum is detected at a sample rate (S)

[0053] The processor is then used to carry out a series of substeps, as described below.

[0054] Referring in particular to FIG. 3, as described above, initially, it is assumed that a single-axis continuous measurement of acceleration is obtained for at least 8.5 seconds duration, using a sample rate (Fs) of at least 10 times that of the predicted modulation frequency. This is used as the input to the algorithm implemented by the substeps of the method of the invention.

[0055] In a first substep (a), the frequency selection module 28 selects a plurality of frequencies within the output vibration spectrum. In substep (a) a plurality of groups of vibration measurements are selected, each group having a common time period, wherein successive groups partly overlap in time and are shifted in time by a time shift (dt.sub.2). The successive groups partly overlap to provide a series of overlapped output frequency spectra.

[0056] Referring in particular to FIG. 3, in substep (a) the acceleration signal is split into a series of M overlapping groups of vibration samples, of length N and spaced apart by time dt.sub.2 seconds. This means each group has an overlap of N (Fs dt.sub.2) samples. The time dt.sub.2 determines the bandwidth of the output phase modulation spectrum.

[0057] In a second substep (b), the analysis module 30 analyses the plurality of frequencies to extract phase information for the plurality of frequencies. In substep (b) each group of vibration measurements is subjected to a Fourier Transform analysis to generate from each group of vibration measurements a respective vibration spectrum. In substep (b) phase information from each frequency of the respective vibration spectrums is extracted.

[0058] Referring to FIG. 3, in substep (b) the Fourier Transform is computed for each group individually to generate a vibration spectrum for each. Then, only the phase information from this is kept, and the magnitude information is discarded and not subsequently utilised.

[0059] The extracted phase information is wrapped within the range of −π to +π. In substep (b) the extracted phase information is unwrapped to provide a respective continuous phase waveform as a function of frequency, which is no longer wrapped around the range of −π to +π. In the preferred embodiment, only the first half of the phase information is kept (N/2 values), since the second half contains the negative frequency portion only.

[0060] In a third substep (c), the generating module 32 generates a continuous phase waveform representing modulation in phase over time for a given frequency of the plurality of frequencies. Preferably, the sample rate (S) is at least 10 times greater than the given frequency in substep (c). Preferably, the sample duration in the first Fourier Transform analysis is shorter than the given periodic phase modulation in substep (c). Preferably, the windowed segments are spaced apart such that the ratio (1/dt.sub.2) is at least a factor of 10 times less than the predicted modulation frequency.

[0061] Referring to FIG. 3, the arrays of unwrapped phase are then stacked on top of each other, which are illustrated as rows in FIG. 3, to form a matrix M*(N/2) in size, and the columns of this array are extracted which then represent the change in phase at a single frequency, over time. The new sample rate (Fs.sub.2) of this phase time series is 1/dt.sub.2. At this stage, the periodicity of any phase modulation may be discernible. The columns of the matrix represent the phase modulation time series at a single frequency.

[0062] For one or more of these frequencies, the time series is first detrended (e.g. the mean or slope is subtracted), and then the Fourier Transform is computed. Only the magnitude of this is then used, and only for the first half of the values, since the second half contain the negative frequency components. From this magnitude of the spectrum, strong modulation will be detectable from the presence of a narrow peak at the appropriate modulation frequency.

[0063] In a fourth substep (d), the detecting module 34 detects peaks in the continuous phase waveform at multiples of the input vibration frequency to produce output data representing defects in the mechanical system.

[0064] The final output of the continuous phase waveform, herein also referred to as the ‘modulation spectrum’ here, can be interpreted as highlighting the strength of any broadband frequency modulation present. By looking for the clear presence of a narrow peak in the modulation spectrum (as compared to neighbouring frequency bins) the presence of a potential crack in the axle can be deduced.

[0065] By computing the modulation spectra produced in substep (d) for a number of different frequencies, the strength of modulation present at each of these frequencies can be determined. This can be valuable in optimising the signal-to-noise ratio, since the modulation may be easier to discern in certain frequency regions compared to others.

[0066] FIGS. 4, 5 and 6 show the relationship, illustrated as a two-dimensional modulation spectrum, between the modulation depth (Y-axis) and the modulation frequency (X-axis) for an axle measured under a 4 kN load. The dashed vertical line indicated the axle rotation frequency. FIGS. 4, 5 and 6 show the output for an uncracked axle, a heavily cracked axle and a lightly cracked axle respectively. The heavily cracked axle shows additional harmonics of the axle rotation frequency.

[0067] The modulation spectra for all frequencies can also be displayed in the form of a 3D surface, as shown in FIGS. 7 and 8. In FIGS. 7 and 8, ere, the horizontal axis is the modulation frequency, the vertical axis (bottom to top) is the frequency being modulated, and the height (out of the page, corresponding to the grey scale) represents the depth of modulation present. A horizontal slice through this surface produces the modulation spectrum, as shown in FIGS. 4 to 6. FIGS. 7 and 8 show that the peaks in modulation due to the axle crack are more clearly distinguishable at lower frequencies (below 700 Hz)—above this frequency the noise floor increases, whilst the peak amplitude at the modulation frequencies stays relatively constant. Various modifications to the preferred embodiments of the present invention will be apparent to those skilled in the art.