ANALYSIS OF WHEEL-RAIL NOISE
FIELD OF THE INVENTION
This invention relates to the analysis of noise generated by train wheels on rail tracks, in particular but not only for assistance with maintenance of the tracks.
BACKGROUND TO THE INVENTION
A high proportion of the cost associated with operation of a railway relates to the replacement of wheels and rails as a result of wear induced at the wheel-rail interface.
Wear can occur in various ways particularly around curves in the track, and is influenced by many factors including curve radius, wheel and rail profiles, suspension stiffness, lubrication, wheel speed and load. Lubrication of the track is one of the more effective methods to reduce wear of the wheel and rail faces, and thereby to reduce maintenance costs.
A number of track-side systems for measurement of wear are known, based on force, friction, vibration, heat, roughness and acoustic emissions, for example. None of these are on-board systems. Another known system uses a laser to make direct measurements of the profile of a rail or the wheel. Such systems are not able to measure wear as it is occurring.
Few other than the laser system are specific to wear of the wheel flange and rail gauge face. A number of on-board wheel-rail noise systems are known, such as the NOISEMON system provided by AEA Technology. However, none attempt to identify specific noise types, and none attempt to relate noise to the degree of wear.
Several distinct noise types are known to exist, including rolling noise, squeal noise, flanging noise and impact noise. Different noise types may indicate differing needs for repair or lubrication. However, none of the existing on-board systems attempt to analyse specific noise types for assistance in determining regions of the track that are likely to
cause wear. Most noise types are also irritating to the human ear and must be reduced to acceptable levels in populated areas.
SUMMARY OF THE INVENTION
It is an object of the invention to provide an improved system for analysis of wheel-rail noise, or at least to provide an alternative to existing systems. Analysis of the noise can assist maintenance of both the wheels and the track by indicating portions of a track at which lubrication, repair or speed control may be required.
In one aspect the invention may therefore be said to reside in a method of detecting types of wheel-rail noise generated by the motion of a train, including: locating a microphone on the train to record noise from one or more wheels, carrying out a frequency analysis of noise events that are detected by the microphone as the wheels move along a track, and determining whether the noise events include one or more of squeal noise, flanging noise and impact noise. Preferably the noise types are differentiated by one or more of the methods indicated below.
In one embodiment the noise levels are determined by recording the noise through a microphone located in the vicinity of the wheels, and recording position and speed data in conjunction with the presence of squeal noise. Noise data is recorded substantially continuously for a plurality of successive time periods during a train journey.
In another aspect the invention resides in apparatus for detecting types of wheel-rail noise on a train, including: a microphone mounted to detect noise from one or more wheels on the train, a positioning system mounted on the train in relation to the microphone, a data analysis system that processes noise data from the microphone to identify the presence of one or more of squeal noise, flanging noise and impact noise, and a data storage system that records the presence of noise types determined by the analysis system in conjunction with data from the positioning system.
In another aspect the invention may be said to consist in a method of detecting flanging noise from one or more wheels on a train, including: recording noise from the wheel for a period of time to create a data frame, calculating a frequency spectrum for data within the frame, determining noise power levels for a group of relatively high frequency bands within the spectrum, determining whether the power level in a trough band having the lowest power level of the group exceeds a first predetermined threshold, if the threshold is exceeded, calculating a ratio of total noise power within the group to total noise power within the spectrum, and recording flanging noise as being present within the frame if the ratio exceeds a second predetermined threshold.
In another aspect the invention resides in a method of detecting squeal noise from one or more wheels on a train carriage, including: recording noise from the wheel for a period of time to create a data frame, calculating a frequency spectrum for data within the frame, determining noise power levels for consecutive frequency bands within the spectrum,comparing noise power in a peak band having the highest power level in the spectrum with noise power in each band adjacent to the peak band, and recording squeal noise as being present within the frame if the power level in the peak band exceeds the power level in each adjacent band by more than a predetermined threshold.
In still another aspect the invention may be said to consist in a method of detecting impact noise from one or more wheels on a train, including: recording noise from the wheel for a period of time to create a data frame, dividing the frame into multiple sub frames, determining the noise power level in each sub frame, recording impact noise as being present within the frame if the difference in power level between adjacent sub frames exceeds a predetermined threshold.
Preferably the method further includes carrying out the method above to detect the presence of flanging noise within the data frame, and recording impact noise as being present within the frame only if flanging noise is not also recorded as being present.
Lubrication such as water or oil may be applied from the train to the track following detection of noise types. Other noise reducing steps may also be taken, such as slowing the train to reduce squeal noise in populated areas, or determining a need to carry out repair of the track.
LIST OF FIGURES
Preferred embodiments of the invention will be described with respect to the accompanying drawings, of which: Figure 1 indicates the origin of three principal types of wheel-rail noise,
Figure 2 schematically shows a system for detection and analysis of wheel-rail noise,
Figure 3 outlines functionality of the system in Figure 2, Figure 4 compares frequency spectra for different noise types, Figures 5a, 5b indicate how squeal noise may be analysed,
Figure 6 gives an analysis of the spectra in Figure 4, Figure 7 is a flowchart outlining an algorithm for analysis of squeal noise, Figures 8a, 8b, 8c indicate how flanging noise might be analysed, Figure 9 is a flowchart outlining an algorithm for analysis of flanging noise, Figure 10 is a data frame indicating an impact noise event,
Figure 11 indicates how impact noise may be analysed,
Figure 12 is a flowchart outlining an algorithm for analysis of impact noise, and Figure 13 indicates how lubrication may be applied immediately after detection of noise.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
Referring to the drawings it will be appreciated that the invention may be implemented in various ways for a range of different purposes related to the performance of railways. The embodiments described here are given by way of example only. Known methods of
acoustic analysis will be appreciated by a skilled person and need not be described in detail.
Figure l(a), (b), (c) are cross sections indicating three of the principal types of noise generated by the motion of train wheels on a rail track. Rolling noise originates from structural vibrations that are generally caused by combined roughness at the wheel-rail interface, as opposed to impact noise which is caused by larger but intermittent irregularities such as joints. Squeal noise is generally caused by lateral creep of wheel tread across the rail head, and lies within the higher frequency part (1-10 KHz) of the range of human hearing (20 Hz -20 KHz). A stick-slip model of squeal noise has been proposed. Flanging noise is induced by rubbing of the wheel flange against the rail gauge corner or gauge face, and the power spectrum may have a predominant peak among a series of peaks in the higher frequency range. Squeal noise and flanging noise are usually most evident from the inside and outside leading wheels of each bogie respectively when the train rounds a sharp curve in the track.
Figure 2 is a cross section through a train carriage 20 indicating an on-board system for detection of noise types arising from the wheels 21 and rails 22 while the train is in motion. The system is typically based around a laptop 23 or other portable computer containing software for data acquisition and analysis. A microphone 24 is preferably placed under the carriage but may in practice be placed anywhere on the carriage where a suitable signal can be detected from the wheels, such as outside a window. A satellite positioning device 25 is preferably placed on the roof of the carriage but again may be placed any that an acceptable satellite signal can be detected. The device may include a GPS antenna and receiver for example, although non-satellite systems might also be used. The microphone and the position device each have a wired or wireless connection to the computer 23 for transmission of data. When used for track maintenance the system is coupled to a lubrication device further back on the train, or data may be flagged to indicate a need for track repair. When used for speed control the system is coupled to a drive system that operates the engine of the train.
Figure 3 indicates the computer system in Figure 2 in an abstract form. A user interacts manually with a system controller 30 through a graphical interface 35. Signals from the on-board noise and position sensors 31 are received by a data acquisition program 32 and processed by a signal processing program 33. Sound data is typically collected and digitised in frames of about one second duration, using a sample frequency of 22 KHz or 44 KHz, then A-weighted and converted to a 1/3 or 1/6 octave spectrum. The existence of particular noise types is determined from the spectrum by algorithms as described below. Each frame is processed in real time by the program 33 and the result recorded along with train position and speed information in a data log 34. Each result for a frame typically includes an indication of whether or not particular types of noise were present and the average noise level for the frame.
Wheel-rail noise is detected as the sound pressure variation above or below the static value of atmospheric pressure. Acoustic noise level is measured in decibels (dBA) to accommodate the wide range of human hearing. A weighting scheme is usually adopted because the human ear does not respond in a uniform manner to different frequencies. Most commonly used in acoustic analysis is A-weighting which provides a reasonable correlation with the human response. The noise energy level distribution as a function of frequency is determined by a spectrum analyser, preferably a 1/3 or 1/6 octave band analyser. Further details of acoustic measurements can be found in a number of text books such as "Handbook of Acoustical Measurements and Control", McGraw-Hill 1998.
Figure 4 is a typical 1/3 octave spectrum illustrating rolling, squeal, flanging and impact noise measured on the same train at the same speed. There are differences between the spectra of each noise type that can be used by the processor in Figure 3 to analyse the signal from the microphone and determine the existence of a particular type of noise at a particular region on the track. Rolling noise energy tends to be concentrated in the range 100 to 2000 Hz and decreases rapidly over about 1 KHz. Impact noise appears similar to rolling noise but has an overall increase towards 2000 Hz and is very short lived in time. Squeal noise is usually tonal, with a peak in the range 1 - 10 KHz, at around 6.3 KHz in
this example, but otherwise appears similar to rolling noise. Flanging noise has a relatively high energy content across the higher frequencies.
Figure 5 (a) indicates how squeal noise may be indicated by a frequency peak in a band having a noise level that exceeds the levels in neighbouring bands by a predetermined threshold. Alternatively Figure 5(b) indicates how the peak may be present in two adjacent bands having similar noise levels but again exceeding the levels in their respective neighbouring bands by a predetermined threshold.
Figure 6 gives a comparison of the peak noise levels for each type of noise in Figure 4. It can be seen in the bottom row of data that the noise level differences of 21 dBA and 24.2 dBA are substantially greater for squeal noise than for the other noise types. A threshold of around 10 dBA for the difference between the peak noise level and neighbouring noise levels is thought to be sufficient for reliable detection of squeal noise.
Figure 7 outlines an algorithm that may be used by the signal processor to determine the existence of squeal noise in a data frame. The initial steps in the algorithm are generally self explanatory in relation to data acquisition and signal processing. Once the frequency band containing the peak power level P(k) at frequency f(k) has been determined, the least difference Dl over the power levels P(k-1) and P(k+1) in the adjacent bands is determined. If Dl is greater than the threshold then squeal has been detected within the frame according to Figure 5(a). Otherwise analysis continues for the possible case where the peak is spread over two bands and the differences D2 and D3 are calculated according to Figure 5(b). The processor continues to test for squeal noise until stopped by the operator.
Figure 8(a) indicates how flanging noise might be determined by setting a minimum threshold for all noise levels in the higher frequency range. In this example the threshold is set at 80 dBA in the spectrum of Figure 4, and only the flanging noise bands exceed the threshold in every case. The magnitude of the threshold will depend on the distance between the microphone and the wheel at which flanging takes place, ideally 1-3 m. In some cases however, other kinds of noise may also meet a simple test of this kind, such as
when the train travels in a tunnel or at high speed. Figure 8(b) indicates a spectrum in which impact noise crosses the threshold in every high frequency band. A more sophisticated algorithm is generally required.
Figure 9 outlines an algorithm that may be used by the signal processor to determine the existence of flanging noise in a data frame. The initial steps in the algorithm are generally self explanatory in relation to data acquisition and signal processing. A minimum band noise level L(k) is determined within the high frequency range 2 — 10 KHz and then compared with a threshold as mentioned above in relation to Figure 8(a). If the minimum noise level is above the threshold then a band power ratio BPR is calculated as shown (although the index k relates to the bottom of the frequency range rather than the band with minimum level). This ratio is then compared With a further threshold of about 40%, although the value of either ratio may be varied to suit conditions. The highest BPR found for impact noise in experiments so far is about 32%, so a threshold above 35% is thought to be satisfactory. Figure 8(c) gives a comparison between this test applied to Figures 8(a) and Figure 8(b) with BPRs of around 58% and 91% respectively for flanging noise. If the second threshold is exceeded then the signal processor records the existence of flanging noise in the frame. The processor continues to test for flanging noise until stopped by the operator.
Figure 10 shows the time waveform for a typical event of impact noise, in the absence of other noise types. The duration of these events is typically less than 1/10 second. In practice it is difficult or impossible to separate impact noise from rolling noise based on spectrum information alone, so the time waveform must also normally be used.
Figure 11 indicates how impact noise may be detected in a data frame by determination of a distinct peak in the time waveform. The frame is divided into sub frames each about 0.2 seconds long, creating five sub frames in a 1 second frame for example. Any sub frame having a peak in noise level is compared with neighbouring sub frames. If the difference between the peak from and each adjacent sub frame exceeds a threshold then an impact
noise event may be present, in the absence of other noise types. A threshold of about 3 dBA has been found satisfactory.
Figure 12 outlines an algorithm that may be used by the signal processor to determine the existence of impact noise in a data frame. The initial steps in the algorithm are generally self explanatory in relation to data acquisition and signal processing. The overall noise level in the data frame is determined and compared with the noise level of the previous frame. If the difference exceeds a threshold then impact noise may be present. The frame is then analysed further by division into sub frames as shown in Figure 11. If the threshold difference between sub frames is not exceeded then impact noise is probably not present in the frame. The same threshold is typically used in each of these steps. Otherwise a frequency spectrum and BPR are calculated for the frame generally as indicated in Figure 9. If the PBR is less than the threshold for flanging noise the signal processor records the existence of impact noise.
Noise data collected using on-board equipment of the present kind may be used in various ways, either in real time or after the journey has ended. Figure 13 shows how lubrication such as water or oil may be applied directly from the train to the track before the train has passed the point where noise was detected. The lubrication may be applied in different ways depending on the type of noise, by varying the type of lubricant and the portion of the track for example. Lubricant to reduce flanging or squeal noise is applied to the flange or the top face of the track respectively. Other noise reducing steps may also be taken, such as slowing the train to reduce squeal noise in populated areas. The level of noise may also be analysed to indicate the degree of wear and therefore the volume of lubricant required and whether or not a portion of track needs manual repair.
It will be appreciated that systems for analysing wheel rail noise according to present invention may be implemented in a portable computer system without need of extensive calibration for a particular train. The data may be used for track maintenance by recording locations along the track at which repair or lubrication may be required, or for other systems such as a speed controller on the train.