VIBROACOUSTIC MEASURMENT FOR BEARING FAULT DETECTION ON HIGH SPEED TRAINS S. BELLAJ (1), A.POUZET (2), C.MELLET (3), R.VIONNET (4), D.CHAVANCE (5) (1) SNCF, Test Department, 21 Avenue du Président Salvador Allende, 94407 Vitry-sur-Seine Cedex, selim.bellaj@sncf.fr (2) SNCF, Test Department, 21 Avenue du Président Salvador Allende, 94407 Vitry-sur-Seine Cedex, arthur.pouzet@sncf.fr (3) SNCF, Test Department, 21 Avenue du Président Salvador Allende, 94407 Vitry-sur-Seine Cedex, cyril.mellet@sncf.fr (4) SNCF, Engineering Department, 6 Avenue François Mitterrand, 93574 La Plaine Saint Denis Cedex regis.vionnet@sncf.fr (5) SNCF, Test Department, 21 Avenue du Président Salvador Allende, 94407 Vitry-sur-Seine Cedex, daniel.chavance@sncf.fr Summary One major challenge in railway maintenance lies in the transfer from a preventive process to a predictive one. It represents a tremendous economical gain for railway operators. Roller bearing fault monitoring is part of that issue. Considering the number of trains, coaches, hence the amount of axles and roller bearings they imply, it clearly appears that any substantial gain in the maintenance costs ends up as a huge interest for an operator. Regarding the roller bearings, one optimization direction lies in their monitoring in order to detect which axles have to be taken care of with the highest priority. In this context, a research program was set up by the SNCF in order to define a new system for defective bearing detection. That is why SNCF organized a test campaign involving a typical high speed train which was mounted with faulty bearings. Vibroacoustic measurements were made on the trackside as well as onboard. Several configurations were tested to investigate the emission and propagation of a defect. For each test run, statistical indicators were computed in addition to various frequency analyses such as third octave band spectrum and envelope detection. The study demonstrated that onboard, the accelerometers placed on the bearing boxes made possible the identification of a characteristic behaviour for defective bearing, and trackside, the results obtained from the acoustic measurements allow a good detection rate from several pass by measurements. 1. Introduction The bearings, being an element of interface between the wheels and the bogies of rolling stock material, play a considerable role in the reliability of a train s performance. The monitoring of their state of deterioration is a key factor is the strategy of preventative maintenance. Given the increase of high speed lines in France, SNCF, in its role as railway operator, has experienced considerable growth in its fleet of high speed trains (TGV). The challenge of preventative maintenance, particularly in terms of running gear, on trains operating at up to 320 km/h, has become a priority of SNCF. With regards to preventative maintenance on bearings, acoustic and vibratory systems already exist and function mainly to be applied on classic rolling stock (i.e. freight). For high speed rolling stock, new problems are posed, for example the presence of motorized bogies which might damage the detection because of the presence of additional sounds. Therefore, field testing was conducted by SNCF during which a test train (TGV) was equipped with faulty bearings. Vibroacoustic measurements were made trackside as well as onboard. The paper presents the test itself, a description of the faulty bearings used, and of the different acoustic systems gathered for the trial, as well as the results of both trackside and onboard analyses.
2. Field test presentation 2.1 Test Train Description The test train was a TGV Sud-Est TVM 300 V270, equipped with 8 defective bearings, positioned according the diagram below. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Figure 1. Faulty bearing placement on the test train (TGV Sud-Est TVM 300 V270) In order to clarify the following results, the axles were numbered from 1 to 26. Following this numbering, the defective axels are 1, 2, 5, 6, 9, 11, 13 and 14. 2.2 Measurement Description Measurement points were situated at both trackside as well as onboard the test train. In terms of onboard measurements, three axles - numbers 5, 8 and 9 were equipped with accelerometers on their bearing boxes (axle 5 is on a motorized trailer bogie and axle 8 and 9 are on intermediate carrying bogies). In regards to trackside measurements, six microphones, spaced 1m apart, and placed at 2.235m from the external stretch of rails, and 50cm above the running surface, were installed. Speeds sensors were installed on the track in order to obtain information on speed, as well as the location of the axels on temporal signals. 2.3 Set-Up of The Test The pass bys of the test train, in front of the trackside measurement points, were conducted on track possession. This made it possible to perform some 50 pass bys in both directions, and at various speeds, ranging from 20 to 120 km/h. 3. Data Analysis 3.1 Rudiments of Faulty Detection Vibroacoustic analysis is one of the most effective methods used to survey the functioning of the rotating machine components (bearings, gears, etc). In the case of bearings, the presence of a defect in one of its components provokes the appearance of periodical shocks. In this way, the presence of a modulating frequency, called a bearing defect frequency, is observed. This frequency is a function of the kinematics and geometrical parameters of the bearing (diameters, number of rollers, rotational speed). In the context of the field test, only the BPFO (Ball Pass Frequency Outer Race) was represented. The formula which permits the calculating of this frequency is as follows: D i BPFO N 2Dm where : is revolution number per second, N is the number, Di is the inner race diameter, is the rolling element diameter. Dm Outer race bearing defect > 50 mm Outer race bearing defect < 50 mm Sound bearing In reference to the test train, the defect frequencies range from 20 to 130 Hz. The detection of these defect frequencies on the signature of a bearing can be carried out by the spectral envelope analysis technique. Moreover, to increase the signal to noise ratio, a signal analysis must be performed around a resonant frequency, which allows for the concentration of vibratory energy. In the case of faulty bearings, a resonance appears in the high frequency domain, due to the impact generated by the interaction between the rollers and the defect. After the demodulation process and a spectral analysis, the envelope spectrum obtained reveals the defect frequencies.
3.2 Field Test Data Analysis The time of acquisition plays an important part for the spectral analysis. Indeed, to be able to extract the defect frequencies, a pretty fine frequential resolution must be obtained. Regarding to onboard vibratory measurements, obtaining a sufficiently long acquisition time was easy. On the other hand, concerning the acoustic measurements, the acquisition duration of an axle passing by faced with a microphone was considerably reduced because of signal analysis on both sides of the sensor over a 1.5m length. The solution was thus to place several microphones along the way so as to lengthen the duration of the signal coming from an axle. Hence, a method was developed with the aim of concatenating the signals resulting from several successive sensors. The first data processing consists in correcting the Doppler Effect. Indeed according to the speed of the train, this effect can be strongly marked and modify the frequential contents, as well as the signal amplitude received on the microphone. Therefore, a correction on of the acoustic signals is applied in accordance with the following formula: R( t) S( t ) S ( ) c f t R( t) where S is the signal gross from the microphone, R(t) the distance at the moment t between the source and the microphone and C the sound celerity in the air. Thereafter, the process of concatenation is described by the following Figure 2: - for the period of passage of the source in zone 1, the signal from the first microphone is preserved; - at the time of the passage in zone 2, the signals of microphones 1 and 2 are compared (analyzes delay and level of correlation) and a zone where the signals are most similar is defined. In this zone, a linear interpolation is carried out in order to connect the signals; - in zone 3, only microphone 2 is used; - in zone 4, the same process that in zone 2 is applied; - finally in zone 5, the signal from the last microphone is preserved. In order to illustrate the process of concatenation, the following figure shows an example of comparisons of two signals resulting from close microphones. This example corresponds to phase 2 of Figure 3. Magnitude (Pa) Time (s) Figure 2. Processus description of signals concatenation Figure 3. Comparison of two microphone signals dedoplerised after retiming In this example, it clearly appears that after the process of retiming, the signals follow a similar evolution. The connection point between the two microphones is then defined as the moment when the two signals are similar. This process thus makes it possible to connect signals resulting from successive sensors without creating sudden break or fictitious periodicity. The useful signal duration is considerably increased and allows a finer frequency analysis. 4. Results 4.1 Onboard Vibratory Measurements Vertical and longitudinal accelerations were measured on two defective axle boxes and on one sound axle box. Figure 4 shows the third octave band spectrums of two vertical accelerations (defective bearing on axle 9 and sound bearing on axle 8) at different speeds.
Vertical acceleration on axle 8 Vertical acceleration on axle 9 Acceleration level (db / ref. 10 e -4 m.s -2 ) Acceleration level (db / ref. 10 e -4 m.s -2 ) Third octave band frequency (Hz) Third octave band frequency (Hz) Figure 4. Comparison between the third octave band spectrums of axle 8 (left) and axle 9 (right) at different speeds First, it can be noticed that levels of acceleration are very different between the defective and sound bearing. In addition, for both axles, there is a significant influence in terms of speed on every third octave band, but this influence is not linear: - For a frequency domain ranging from 20 Hz to 2 khz, the dispersion is in the same order of magnitude (about 20 db for each bearing). - For higher frequencies, the speed seems to have a greater impact on axle 9. Indeed, the dispersion can reach more than 30 db against 10dB for axle 8. This result is in accordance with the phenomenon of resonance in the high frequencies domain mentioned for faulty bearings. Concerning longitudinal accelerations, same results are obtained. In the same way, there is no influence in terms of the pass by direction. Note: The phenomenon observed at about 80 Hz (invariant with speed) seems to correspond to the first mode of bending axle. In order to more precisely identify the bearing defects, spectral envelop analyses have been made. Figure 5 illustrates the comparison of the envelop spectrums obtained from axles 5, 8 and 9 box accelerations for a 60 km/h pass by. Envelop spectrum of axle 5 bearing box acceleration Acceleration level (db / ref. 10 e -4 m.s -2 ) Envelop spectrum of axle 8 bearing box acceleration Envelop spectrum of axle 9 bearing box acceleration Figure 5. Envelop spectrums for axles 5, 8 and 9 box accelerations at 60 km/h
In this case the BPFO is about 60 Hz and is clearly identifiable with several harmonics frequencies on graphs of axle 5 and 9. This analysis was applied on all pass bys and their study demonstrates that the BPFO is present for the majority of them. However, in some cases the fundamental frequency is not clearly visible and these harmonic frequencies can be more representative of the presence of a defect. 4.1 Trackside Acoustic Measurements The following results were established in order to define characteristic indicators for axles having a defective bearing. For each tested indicator, the axles were placed in order according to their values. The reliability of an indicator is evaluated through a statistical study on all recorded pass bys: the eight axles showing the most important levels according to the indicator are compared with the eight defective axles. The indicator is regarded efficient if a majority of defective axles are included in the eight noted axles. The first study was carried out by taking measures, and consisted of analysing the temporal evolution of the frequential contents on one microphone. An example is provided in Figure 6 hereafter. BOGIE 1 BOGIE 5 BOGIE 7 Time (s) Figure 6. Spectrogram obtained from one microphone for a 100km/h pass by 30 db range These representations make it possible to distinguish each bogie of the TGV. It is also interesting to notice that some bogies have stronger levels than others in the high frequencies domain. It is in particular the case for the first and the fifth bogies (motorized) and the seventh bogie (carrying). These three bogies comprise at least one axle with a faulty bearing. Nevertheless it is not sufficient to conclude with these results, for there is no generalization which can be established for all the pass bys. The acoustic measurements were treated in order to locate, in the frequential domain, a typical emergence of the damaged axles. For each axle of the test train, the octave and third octave band spectra were calculated, then energetically averaged on the six microphones available. In each frequency band, the axles were placed in order according to the effective values of the acoustic pressure. A statistical analysis was carried out on 50 passages and consisted of studying the percentage at which a given axle is revealed as one of the eight noisiest. Generally, the results demonstrate that the damaged axles have significantly raised levels compared to the others in the 16 khz octave band.
Indeed, at relatively important speeds (higher than 60 km/h), the defective axles are easily identified, particularly at 100 km/h where defective axles 1, 2, 5, 9, 13 and 14, almost always appear among the eight noisiest axles in the 16 khz octave band. However, on axle 22, not faulty a priori, a defect was also always detected. Moreover, for low speeds, discrimination according to this indicator is not very satisfactory as several operational axles were observed as being defective. Thus the statistical analysis was pursued using spectral envelop analysis. The concatenation method explained in paragraph 3.2, made it possible to obtain a sufficiently thin frequential resolution to distinguish an emerging peak from the measurement noise. For each axle, an envelop spectrum is obtained and underlines some frequencies (BPFO and its harmonics) in a more or less marked way. An example is provided below (Figure 6) for the axle 2 (on motorized bogie) during a 30 km/h pass by. Harmonic 4 Fondamental Frequency Harmonic 1 Harmonic 2 Harmonic 3 Magnitude (db / réf. 2.10 e - Pa) Harmonic 5 Figure 6. Envelop spectrum of axle2 at 30 km/h An axle classification is carried out in terms of the levels integrated on a frequency band included around the considered frequency (BPFO) into more or less 10 %. The eight axles having the strongest levels were retained. In the same way as the preceding indicators, this one shows some limitations for low speeds since several operational axles are detected as defective ones. Nevertheless, at higher speeds, the results show a relatively good reliability. At 100 km/h for example, speed measured with eight recoveries during the field test, six faulty axles are detected in at least two passages out of three, whereas no operational axle is retained in a recurring way. In the same way, at 60 km/h, six axles with defects are detected at least for three cases out of four, compared to one operational axle statement for one case out of two. Therefore the reliability of the results is strongly increased using this indicator. When all the passages are considered, without taking into account the direction or the speed pass by, only six axles are detected and they are actually damaged. That shows the discriminating capacity of this last indicator. 4. Conclusion The methods of measurement and analysis, performed within the context of the field test, demonstrated their effectiveness in terms of tools of detection and diagnosis of the bearing defects. More particularly, the application of these methods to the case of high-speed trains, like the TGV,
removed the doubts in terms of capacity of detection because of appearance of additional noise sources due to the motorization of a great number of bogies. Indeed onboard measurements demonstrated an increase in vibratory levels in the high frequencies domain (beyond 2 khz) for the axles equipped with damaged bearings. Moreover, the root mean square level of acceleration seems to provide a threshold from which the detection of the faulty bearings is effective from a 60 km/h speed. Finally, the envelop analysis allows the defective bearing detection thanks to the defect frequencies modulation. Nevertheless, the defect identification is not always revealed by the fundamental frequency but also these harmonics. Concerning the track side acoustic measurements, it was noticed that the signal noise levels filtered in the 16 khz octave band allow a right discrimination of the sound and defective bearings. However detection, with a reasonable confidence index, is impossible withe only one measurement of a pass by. A statistical analysis on several pass bys allows nevertheless a detection rate of six axles out of eight particularly for speeds higher than 60 km/h. In the same way, the spectral envelop analysis makes possible to distinguish the defect frequencies modulation thanks to the development of the concatenation method of the signals. Thereby the results are perceptibly improved but do not allow the systematic detection of the defective bearings with only one pass by. The application of more advanced techniques of signal analysis, for example the wavelets transform analysis, could also considerably improve detection, particularly for low speeds pass bys.