CRC for Rail Innovation

Transcription

1 - CRC for Rail Innovation Established and supported under the Australian Government s Cooperative Research Centres Programme A review of railway noise source identification, mitigation methods and priorities within the Australian Context

2 DOCUMENT CONTROL SHEET CRC for Rail Innovation Floor 23, HSBC Building Brisbane Qld 4000 GPO Box 1422 Brisbane Qld 4001 Tel: Fax: Document: Title: A review of railway noise source identification, mitigation methods and priorities within the Australian context Project Leader: Richard Dwight Authors: Wenxu Li, Jiandong Jiang, Richard Dwight Project No.: R1.105 Project Name: CRC rail innovation project Synopsis: Controlling railway noise at source may be more cost-effective than using noise barriers. The main railway noise sources of relevance to Australian conditions include rolling noise, curving noise and traction noise. Current identification methods do not practically identify all of these noise sources. Under this context, a portable prototype system is proposed, which aims to discriminate noise sources and their contributing components. A review of current measures to reduce these noises at source indicates that wheel design, damping and screening; rail grinding, damping and vibration absorption; friction modifiers; and operational controls should all be considered. REVISION/CHECKING HISTORY REVISION DATE ACADEMIC REVIEW INDUSTRY REVIEW APPROVAL NUMBER (PROGRAM LEADER) (PROJECT CHAIR) (RESEARCH DIRECTOR) 0 [insert date] DISTRIBUTION REVISION DESTINATION Industry Participant for Review X Established and supported under the Australian Government s cooperative Research Centres Programme Copyright 2011 CRC for Rail Innovation This work is copyright. Apart from any use permitted under the Copyright Act 1968, no part may be reproduced by any process, nor may any other exclusive right be exercised, without the permission of CRC for Rail Innovation. CRC for Rail Innovation 30/06/2011 Page i

3 Contents 1 INTRODUCTION BACKGROUND RESEARCH PURPOSE STATEMENT STRUCTURE OF REPORT MITIGATION PRIORITIES ROLLING NOISE IDENTIFICATION AND SEPARATION METHODS REVIEW OF METHODS FOR WHEEL AND RAIL ROUGHNESS MEASUREMENT REVIEW OF METHODS FOR ROLLING NOISE SEPARATION DISCUSSION AND IMPLICATIONS FOR CRC RAIL PROJECT R CURVE SQUEAL AND TRACTION NOISE IDENTIFICATION METHODS REVIEW OF CURVE SQUEAL NOISE IDENTIFICATION METHODS REVIEW OF TRACTION NOISE IDENTIFICATION METHODS REVIEW OF RAILWAY NOISE IDENTIFICATION SYSTEMS SINGLE MICROPHONE SYSTEMS MICROPHONE ARRAYS COMBINED SENSORS COMPARISON OF EXISTING KNOWLEDGE AND IMPLICATIONS PROPOSAL OF A PROTOTYPE NOISE MONITORING SYSTEM MITIGATION MEASURES FOR ROLLING NOISE IMPLICATIONS FROM TWINS MODEL CONTROLLING OF ROLLING SURFACE ROUGHNESS MINIMISING WHEEL RADIATION MINIMISING TRACK RADIATION REDUCTION OF SOUND PROPAGATION CLOSE TO THE SOURCE MITIGATION METHODS FOR CURVING NOISE AND TRACTION NOISE REVIEW OF MITIGATION METHODS FOR CURVING NOISE REVIEW OF MITIGATION METHODS FOR TRACTION NOISE CONCLUSION AND FUTURE WORK REFERENCES CRC for Rail Innovation 30/06/2011 Page ii

4 Executive Summary The cost effective management of railway noise requires the identification of relevant noise sources. The first step towards this is a practical identification method and associated device to separate the various components of railway noise. These components may be defined according to either their generating mechanism or the elements radiating the noise and their relative contributions. Based on a complaint survey on NSW railway lines from 2001 to 2005, mitigation priorities for different noise sources have been identified. The significant noise sources, viz., rolling noise, curving noise and traction noise, are significant to the Australian context. In order to discriminate these noise sources, current identification and assessment techniques using microphone arrays and using individual microphones combined with other sensors, such as accelerometers, have been reviewed. The results of the review include recommended approaches to noise identification that can be incorporated into a prototype noise identification and monitoring system. The system comprises a microphone, a wheel sensor and a minimum of 2 accelerometers. This system will overcome the lack of resolution of the single microphone system and the high expense and lack of precision delivered by microphone arrays. The proposed system is portable, thereby making it suitable for deployment at any site along the railway track. This system incorporates algorithms that both identify the occurrences of and magnitude of curve squeal noise, rolling noise and locomotive noise. The methodology adapted to discriminate wheel squeal and flanging noise is the time-frequency characteristics. Locomotive noise can be identified by its physical location on the train. Rolling noise separation can be achieved through the use of established noise separation methods that include: the vibro-acoustic track noise method (VTN); the multi input single output (MISO) method; and the pass-by analysis (PBA) method. For traction noise sources, such as drive noise and fan noise, other sensors, including sensors on-board the train may be required. A review of current mitigation measures indicates that wheel design, wheel damping and screening, rail grinding and damping, vibration absorption, friction modifiers, and operational optimization may provide benefits. CRC for Rail Innovation 30/06/2011 Page iii

5 List of Figures and Tables Figure 1 Survey on Complaints of New South Wales railway lines between 2001 and Figure 2 Survey by Complaint Category on New South Wales railway lines between 2001 and Figure 3 Schematic diagram of wheel/rail rolling noise generation mechanism (Thompson, 2009)... 5 Figure 4 Rolling noise model in IMAGINE (Jones et al., 2007)... 5 Figure 5 Measurement set-ups for VTN method... 9 Figure 6 Comparison between VTN and TWINS results. Left: freight train. Right: passenger train (Verheijen and Paviotti, 2003) Figure 7 Spiral array developed at DB-AG (Nordborg et al., 2000) Figure 8 Example of a measurement result on ICE3 with the spiral array. Max Level 86 db(a), dynamic 8 db, v = 330 km/h (Van Beek et al., 2002) Figure 9 Instrumentation for noise prototypes system Figure 10 Main results of the STAIRRS project (Oertli and Hübner, 2007) Figure 11 Schematic view of the model for rolling noise, showing the main potential means of reducing rolling noise (Thompson, 2009) Figure 12 Schematic view of the model for curve squeal noise derived from Rudd s theory (1976), showing the main potential means of reducing curve squeal noise Table 1 Characteristics of different noise sources and priorities for mitigation in Australia... 4 Table 2 List of measurement signals. Optional signals are required for maximum accuracy (Verheijen and Paviotti, 2003) Table 3 Review of source separation methods developed in Europe (Verheijen, 2004) Table 4 Comparison of current noise identification systems Table 5 Instruments and installations for noise prototype system Table 6 Solutions for rolling noise control and acoustical gain in db (A) (Vincent, 2000) CRC for Rail Innovation 30/06/2011 Page iv

6 Abbreviations and Acronyms CRC Cooperative Research Centre IMAGINE Improved Methods for the Assessment of the Generic Impact of Noise in the Environment MISO Multiple Input Single Output PBA Pass-by Analysis STAIRRS Strategies and Tools to Assess and Implement noise Reducing measures for Railway Systems TWINS Track Wheel Interaction Noise Software VTN Vibro-acoustic track noise CRC for Rail Innovation 30/06/2011 Page v

7 Chapter 1 Introduction 1 Introduction 1.1 Background This document reports on a review of existing rail noise identification and mitigation methods. This will form the basis for either the application or the further development of these techniques to allow Australian rail organisations to manage railway-generated noise more effectively. Intervention in the process that results in railway noise reaching receivers can occur at various parts of the process. It is likely that controlling the generation of vibration energy, its propagation, radiation, transmission and reception will become progressively more expensive. The volume to be controlled will logically drive the cost of mitigation. The further away from the source of the energy, the larger will be the volume to be treated. Thompson (2000), for example, draws attention to the cost of noise barriers and their lack of viability in many cases. Despite this, it is apparent that a lack of knowledge or lack of capability to control noise before transmission (noise sources) drives rail organisations to use noise barriers or other receiver isolation techniques to mitigate the effects of noise. In order to understand the current limitations on cost effective noise management, the current knowledge of those factors that control the occurrence or magnitude of the noise radiated from a particular site is required. Effective treatments of noise can be identified or developed using this knowledge. The relevance of knowledge of the noise sources contributing at a specific site depends on the situation, including the extent of noise generation from other sources at the site. The classifications of noise that are significance are: 1) Rolling noise: the vertical excitation of the rail and wheel generated by variations or roughness of the wheel and rail running surfaces. Its amplitude is speed-dependent. It results in broadband noise in the frequency range from 50 to 5,000 Hz. It typically dominates over other noise for train speeds between 50 and 300 km/h, particularly on tangent track. 2) Impact noise: resulting from discontinuities in the running surfaces of the rail and wheel. Features creating such discontinuities include design features such as rail joints, points and crossings and defects, such as wheel flats, dipped welds and squats. In each case, the excitation mechanism is similar to rolling noise: a vertical discontinuity in the running surface causes vibration of the wheel and rail. The most apparent difference from rolling noise is the discrete nature of the event and therefore the short time duration of the noise. 3) Traction noise: generated by power units of any kind, including diesel or electrical power sources. It covers a range of possible mechanisms associated with the function of converting the supply energy to mechanical work. Relevant situations include diesel locomotives at idle; and electric or diesel powered systems under traction; and high-speed trains at low speed (below ~60 km/h) or at idling. 4) Friction braking noise: generated by the interaction between the friction material and the rotating element. Some classification systems regard this as a sub-set of traction noise. 5) Curving noise: caused by friction induced self-excitation of the wheel and rail in the lateral direction on low radius curves. 6) Aerodynamic noise: caused by the disturbance of the air flow over the train, which becomes a significant noise source with train speeds greater than 300km/h. 7) Other noise sources: including wagon bunching, coupler noise, bridge noise, warning signals, communications system noise, stabling and yard noise, maintenance noise; internal noise such as airconditioning noise, gangway noise. CRC for Rail Innovation 30/06/2011 Page 1

8 Chapter 1 Introduction 1.2 Research Purpose Statement This report is the result of initial work for project R1.105, Improved Noise Management, funded by the CRC for Rail Innovation. The project aims relate to 2 specific areas: 1. Curving noise, and 2. Noise management and mitigation at source: 1. Curve noise: To develop validated strategies for monitoring and control of curving noise for existing networks. The research problem is to establish reliable and repeatable prediction of the conditions under which curving noise is generated at a particular location by a particular wheel. The research is expected to involve the analysis of extensive field data collected by rail agencies to assist in removing the apparent random nature of its generation. 2. Noise management and mitigation at source: To develop the tools and methods for the rail industry to evaluate, trial and implement internationally available techniques for noise mitigation. This initial work will deliver a draft method and tools for determining the contribution of components to the radiation of rolling noise, together with guidelines for its use. 1.3 Structure of report 1. The practicality of existing noise source identification and assessment techniques are reviewed, with the aim of identifying suitable techniques for application to the Australian context. The purpose of this work is to identify suitable techniques for discriminating sources of noise generated at a specific site. Given that a feasible technique is identified, it will be trialled to confirm its suitability. 2. Existing noise mitigation techniques are reviewed in order to establish the status of knowledge about their applicability to Australian noise problems. CRC for Rail Innovation 30/06/2011 Page 2

9 Chapter 2 Mitigation priorities 2 Mitigation priorities Results of a survey of complaints (AAc/ /R01LDK, 2006) from local residents in New South Wales between 2001 and 2005 are presented in Figure 1 and Figure 2. Figure 1 lists the number in terms of separate noise sources, while Figure 2 displays complaints in terms of various operations. Complaints on NSW railway lines from 2001 to 2005 Number of Complaints % 20.00% 15.00% 10.00% 5.00% 0.00% Percentage of Complaints Noise Sources Figure 1 Survey on Complaints of New South Wales railway lines between 2001 and 2005 Number of Complaints Complaints on NSW railway lines from 2001 to % 40.00% 35.00% 30.00% 25.00% 20.00% 15.00% 10.00% 5.00% 0.00% Percentage of Complaints Noise Sources Figure 2 Survey by Complaint Category on New South Wales railway lines between 2001 and 2005 CRC for Rail Innovation 30/06/2011 Page 3

10 Chapter 2 Mitigation priorities Noise reported as general train noise is likely to be rolling noise predominantly and is the highest generator of complaints. Curve noise also features highly, particularly when considering that the number of residents located near curved track is likely to be significantly less than those near tangent track. Freight trains appear more frequently than other train types. This data tends to suggest priorities in terms of noise mitigation ignoring any benefit to cost ratio. These priorities are presented in Table 1. Table 1 Characteristics of different noise sources and priorities for mitigation in Australia Railway noise sources Locations Speed Conditions Priorities for mitigation in NSW General rolling noise tangent track 50~300Km/h High Curve noise sharp curves -- Medium Impact noise rail joints, welds or dipped welds or points and crossings -- Low Traction noise Engine noise near stations, shunting yards, along the line (acceleration or climbing a gradient) <60km/h or acceleration for a wide range of speeds Low Idling noise near stations, shunting yards Idling Low Train horn noise Medium Aerodynamic noise -- >300 Km/h Nil Mitigation measures on rolling noise radiation can also have important effects on impact noise though reducing the usage of joints and welds will be more effective. The results indicated in Table 1 support the concentration of mitigation activities on rolling noise, curving noise and traction noise. While intentional noise sources, including train horn activation and public address system announcements, are a significant source of complaint, their mitigation must rely on non-engineering solutions. Such noise sources are easily identifiable, and their mitigation will mainly be through control of operation. As a result, intentional noise sources, although they are worthwhile to study, are regarded as outside the scope of this project. It is worth noting that locomotive noise, as one form of traction noise, has become subject to NSW legislation in Environmental Protection Licence (Licence 3142, 1997). Noise limits have been set for different operating conditions. It is therefore important to take locomotive noise mitigation as a priority within the general category of traction noise. What remains unclear to some extent from this study is the accuracy of the noise source identification implicit in the data. The noise classification system is in terms of disparate categories such as generating mechanism: e.g., traction noise, curving noise; radiating body or general classifications: e.g., train noise; and, actions leading to the noise: e.g., increased noise following track work. A number of the categories overlap or may be ambiguous relative to the generating mechanism: e.g., curving noise or train noise. The economics of noise mitigation methods against noise sources and the appropriate means of either improving the impact of noise mitigation spending or reduce the spending required for a given required noise result needs to be established in order to prioritise the application of these methods properly. CRC for Rail Innovation 30/06/2011 Page 4

11 Chapter 3 Rolling noise identification and separation methods 3 Rolling noise identification and separation methods Rolling noise is accepted generally as being induced by the undulations in the contacting surfaces of the wheel and rail, commonly referred to in the rail industry as roughness. Thompson (1990, 1993) has developed an overall description of the rolling noise mechanism, replicated in Figure 3. The various relationships represented by Figure 3 have been coded into a suite of software called TWINS, which has been validated in terms of noise and vibration (Thompson, Fodiman and Mahé, 1996; Jones and Thompson, 2003). A series of European-based projects have also applied TWINS to design noise reduction measures (Thompson and Gautier, 2006), which further confirmed the validity of the TWINS model and also the roughness induced noise generation mechanism. An alternative approach to the TWINS model was developed under the European IMAGINE project and is depicted in Figure 4 (Jones et al., 2007). The initial process is still to determine the wheel and rail roughness. The transfer functions from roughness-to-radiation for both vehicle and track components are determined by field experiment. As the ambient influence is taken into account automatically during field measurements, this method is reliable when quantifying the wheel and rail contribution and evaluating the effects of noise control measures, providing that such influence can be assumed as constant. Figure 3 Schematic diagram of wheel/rail rolling noise generation mechanism (Thompson, 2009) Figure 4 Rolling noise model in IMAGINE (Jones et al., 2007) 3.1 Review of methods for wheel and rail roughness measurement Figure 3 and Figure 4 show that in order to define the noise contribution of the wheel and rail, roughness of wheel and rail should be determined. Different roughness measurement methods are reviewed here. CRC for Rail Innovation 30/06/2011 Page 5

12 Chapter 3 Rolling noise identification and separation methods Direct method The direct method is of course to scan the wheel and rail surface directly and separately. The combined roughness as seen by the wheel and rail in the contact area (so-called total effective roughness) can then be expressed as the energy sum of the wheel and rail roughness together. The roughness with wavelength shorter than the size of the contact patch (generally between 10 to 20 mm) will be attenuated. Through this method, the individual wheel and rail roughness can achieve very high accuracy, but not for the total effective roughness. This may be due to the uncertainties of the contact patch, e.g., the lateral position of the contact patch on the wheel and rail surfaces. It was shown by Thompson, Fodiman and Mahé (1996) that the variations in the effective roughness of up to 5 db occur because of the position of the contact patch on a corrugated wheel. If this method is to be used, an instrument that has enough measurement range should be chosen. In Australia, typical train speeds are between 30km/h and 160 km/h. To measure the typical rolling noise in audible range between 150 and 5000 Hz, roughness with wavelength between ~20mm and 300mm should be measured (less than 20mm will be attenuated by contact patch). Currently, various static frame-based roughness measurement instruments utilising displacement sensors or accelerometers have been commercially available, such as RM1200E (by Müller-BBM). These devices are typically capable of measuring profile samples of 1.2m length, which implies that only the roughness with short wavelengths less than 100 mm can be measured accurately. (Verheijen et al., 2003). The inability to measure longer wavelength roughness confines the applicability of these devices to measure the rail roughness that is only valid for low speed trains. For example, rolling noise at 150 Hz requires roughness excitation with 300 mm wavelength at a train speed of 160 km/h; however, measurement of the roughness at this wavelength is already beyond the capacity of these devices. By adding a few sections near the centre of the reference section of the ISO 3095 measurement protocol, wavelengths up to 315 mm can be achieved (Verheijen et al, 2003). Other measurement systems such as trolley-based system, e.g., CAT (rail roughness trolley) can measure roughness up to wavelength of 315mm without problems, but lack resolution in shorter wavelength bands from several centimetres down to the contact patch dimension (Verheijen et al., 2003). Attention should also be paid to the roughness data processing; in particular, spikes and pits should be removed (Cordier and Fodiman, 2000). Spikes refer to the roughness that are short (much shorter than the wheel and rail contact patch size), sharp and projecting out of the general surface. Such features will be crushed or strongly deformed in the contact patch. They will generally not lead to significant displacements between the wheel and rail. Pits are indentations in the surface that are insignificant relative to the size of the contact patch. Indirect method An alternate method for measuring the wheel and rail roughness is by inferring roughness through its effect in creating vibration or noise. This method may be useful in monitoring a track network as opposed to a specific small section of track. The total effective roughness gives rise to the measured signal and therefore cannot directly yield individual wheel and rail roughness spectra. However, if one roughness component is known by estimation or direct measurement, the other can be inferred via subtraction. The measurement error will be compounded in this approach. An indirect measurement method was developed in the European METARAIL project (Dittrich and Janssens, 2000; Janssens et al., 2006), which uses the rail vibration excited from train pass-bys to derive the total wheel and rail combined roughness. The rail roughness can be monitored by simply equating to the minimum total roughness from passages of many trains. Typical indirect roughness measurement equipments are axle-box accelerometers. High Speed Rail Corrugation Analyser (HSRCA) is a device used for routine measurement of corrugation and welds on the Australian National railway system (Grassie, 1996). Under project LECAV (Bongin, et al., 2010), HSRCA was used in France to measure longitudinal irregularities and acoustics. In the wavelength range 10 to 1000mm, experimental results show that it can give repeatable measurements of rail corrugation and acoustic roughness. Providing measurements are made at a similar speed, the equipment is calibrated and the track dynamic properties do not vary too much from CRC for Rail Innovation 30/06/2011 Page 6

13 Chapter 3 Rolling noise identification and separation methods where the system is calibrated, satisfactory accuracy (less than 5 db re 1μμμμ with direct measurement) can be achieved (Bongin, et al., 2010). Another approach trialled under project LECAV is the ARRoW system, which measures the roughness indirectly with rolling noise measurement by translating the noise level to roughness (Kuijpers et al., 2010). Direct roughness measurement was used to calibrate the results of the indirect measurements. The absolute roughness level for the whole track can be determined by combining the two results. The system delivered good agreements between the indirect and direct measurements with the average difference in each wavelength less than 5 db (refer to 1μμμμ). A good repeatability of the results was also found when measuring at the same speed, with the difference within 3 db (refer to 1μμμμ). Indirect roughness measurement has also been used to separate the vehicle and track noise contribution. All of the rolling noise separation methods introduced in the following section 3.2 are based on roughness induced vibration (or sound) measurement. 3.2 Review of methods for rolling noise separation The wheel and rail noise contribution can be estimated by utilising the TWINS model algorithms with given roughness inputs and appropriate model selection for wheel and rail (Thompson, 2009). Experimental methods such as the Equivalent Forces method (Dittrich and Janssens, 2000), Vibro-acoustic Track Noise method (VTN) (Verheijen and Paviotti, 2003), MISO method (Létourneaux.et al, 2002) and the PBA software tool (Dittrich et al., 2003) have also been developed under a number of European funded research projects. The first objective of these methods is to enable the separation of the noise contribution from vehicle and track. This allows a clearer test and quantification of the effect of noise control measures. It also allows for the separation of responsibilities of the track managers and of the vehicle operators. A second objective of such techniques is to provide measured data to enable the rolling noise emission of any vehicle on any track to be determined. After comparison in terms of robustness and cost-effectiveness, the most promising method may be specified. It is somewhat curious that there have been no published works reporting the further development or application of VTN and MISO since the initial publications associated with the STAIRRS project. It may be that these have been used internally. Equivalent forces method The equivalent forces measurement technique was developed under the METARAIL project (Dittrich and Janssens, 2000) to obtain the noise contribution from the track. An artificial excitation force is applied instead of actual excitation system (train passing-by) to generate a structural response. The relationship between the rail response and its corresponding sound radiation can be utilised to predict the sound generation from actual train pass-bys. A general description of the required steps required to apply this approach are: 1. Arbitrarily select a number (n) of positions at which the equivalent force is to be applied. The forces are stored in vector {F}. 2. Select a number (m) of response positions to monitor the structural response of the system; the responses are stored in vector {a}. 3. Select a number (p) of positions to determine the airborne sound response. The airborne responses are stored in vector {p}. 4. Determine transfer functions between excitation {F} and responses {a}. These functions form matrix [A]. 5. Determine transfer functions between excitation {F} and responses {p}. These form matrix [H]. 6. Determine the operational response {a} operational when a train passes by. 7. Derive equivalent forces from matrix equation [A] {F eq } = {a} operational. 8. Derive sound pressure estimates using {p} estimated= [H] {F eq }. The advantage of this approach is that the track noise can totally and explicitly be separated from the wheel/vehicle noise. No contribution of the wheel noise will be present in the final result. The disadvantage of CRC for Rail Innovation 30/06/2011 Page 7

14 Chapter 3 Rolling noise identification and separation methods this method is that the determination of the relationship between the excitation force and its radiated sound is performed without a train present on the track, which could affect the track dynamics. One effect is the change of rail pad stiffness, which directly affects the transmission distance of vibration along the track by changing track decay rate. Reference vehicle method This method was again devised under the METARAIL project. This method assumes that vertical railhead vibration is a fairly good indicator for the excitation in the wheel/rail contact patch, and that the vibro-acoustic behaviour of the track remains constant with time. With specially prepared vehicles that radiate substantially less noise than the track over the frequency of interest, irrespective of the wheel and rail roughness levels, the radiated sound can be assumed to be from the track. Research (e.g., Dittrich and Janssens, 2000) shows that as long as the 1/3-octave SPL spectrum of the reference vehicle is at least 10 db (A) below that of the track, the accuracy of the track noise level can be within +/-0.5 db (A). The track transfer function can be derived by subtracting the vibration spectrum (in logarithmic scale) from the radiated sound level spectrum (in logarithmic scale) during passages of several reference vehicles. Then, under different combinations of vehicles and this part of track, track contribution can be determined by a given the total vibration input. The vehicle contribution can be obtained by subtraction from the measured total noise radiation. The advantages of this method are that there are no requirements for roughness measurements, and there is no restriction on wheel and rail roughness. The transfer function is characteristic of the vibro-acoustic track behaviour. The disadvantage of this method is that a reference vehicle is required that radiates at least 10 db (A) below the track if 0.5 db (A) accuracy is needed. Measures such as damped wheels, small wheel diameters or the use of full wheel enclosures should be taken, which may prove impractical. Reference track method The reference track method is the counterpart of the reference vehicle method and allows the testing and characterisation of vehicles as opposed to track. The noise level of a vehicle is assumed to equal the total noise level at a quiet track (low track response function). Suggested solutions for conventional tracks (non-slab track) include adding damping to the rail or using optimised rail, such as lower height and narrower rail foot or low rail barriers or combination of those measures; however, the effects of these measures needs to be investigated carefully. VTN method The Vibro-acoustic Track Noise method (VTN) (Verheijen and Paviotti, 2003) was developed within the European project STAIRRS to address particular needs in the measurement and characterisation of railway noise. The method uses accelerometers attached to the rails to measure the lateral vibration of rail head and the vertical vibration under rail foot as shown in Figure 5. Vertical acceleration of sleepers may also be measured to increase accuracy. Vibration energy can be calculated from the acceleration and is equivalent to the sound energy. By using radiation and propagation models, noise contributed by the track at a certain distance from the track centre line can be estimated. Subsequently, vehicle contribution is able to be evaluated by subtraction. Measurement set-up of the VTN method (also suitable for MISO, PBA) is shown in Figure 5. It uses two measurement microphones, 5 (single axis) accelerometers and a trigger pulse generator. CRC for Rail Innovation 30/06/2011 Page 8

15 Chapter 3 Rolling noise identification and separation methods T1 L=lateral V=vertical d=sleeper spacing L1 T1 L2 S1 M2 1.2m M1 2d S1 M1 A A L1 L2 1.75m M2 M1 L1 V1 V2 7.5m V1 Figure 5 Measurement set-ups for VTN method Table 2 gives an overview of the signals required for VTN (Verheijen and Paviotti, 2003). The minimum number of channels required for usage of VTN is three (marked as essential in Table 3), but maximum (validated) accuracy is achieved using all 8 channels listed in Table 2. Table 2 List of measurement signals. Optional signals are required for maximum accuracy (Verheijen and Paviotti, 2003). Sig. Signal type Position Usage for VTN M1 Total noise 7.5m from track centre, 1.2m above rail head essential M2 Nearby noise 1.75m from track centre, 0m above rail head Optional V1 Vertical rail acceleration At mid-span under rail foot 1 essential V2 Vertical rail acceleration At mid-span under rail foot 2 Optional L1 Lateral rail acceleration At mid-span, side of rail head 1 Essential L2 Lateral rail acceleration At mid-span, side of rail head 2 Optional S1 Vertical sleeper vibration On top of the sleeper, near the fastener Optional T1 Wheel trigger pulse Preferably 2 sleeper bays away from the cross section For display only The method requires: 1. Vibration energy from the track and ideally the sleepers to be measured through the use of accelerometers. 2. The rail and sleeper sound radiation power is calculated utilising W = ρ 0 c 0 S < v 2 > σ (Fahy and Gardonio, 2006), where ρ 0 c 0 is the air impedance, S (m 2 ) is the surface area of the vibrating structure and< v 2 > is the squared velocity (m/s) normal to the surface, which is averaged both over time ( ) (s) and over the surface area (<> ), σσ is the radiation ratio. 3. The propagation of track noise to a position (typically 7.5 m from the tack centreline) is evaluated, taking account of absorptive properties of the ground and assuming the rail lateral and vertical vibration as line sources and sleeper vibration as point sources. 4. If the total noise at that distance is known, the vehicle noise contribution can be estimated by subtraction. The VTN method and a software package based on it were validated during the STAIRRS project through a dedicated measurement campaign in According to Verheijen and Paviotti (2003), accuracy up to 1.5dB (A) can be achieved for the track contribution calculation. The accuracy for the vehicle contribution depends on relativity to the track contribution but no event where accuracy better than ±2dB (A) was observed. Figure 6 shows a comparison between theoretical contributions obtained using the TWINS software and the results of the VTN method as applied in the STAIRRS project for a freight and a passenger train. It is evident that the calculated track contributions are close, with the difference being within 3dB in the frequency range from ~250Hz to ~3 KHz for both the freight train and the passenger train. Larger differences can be seen for the vehicle contribution in CRC for Rail Innovation 30/06/2011 Page 9

16 Chapter 3 Rolling noise identification and separation methods the frequencies below 2 KHz. During this frequency range, track radiation is generally over the wheel, a minor calculation shift of track contribution may result in a big variation of vehicle radiation. The oversimplification of the models for the rail and sleeper by line source and point source may result in a less reliable estimation of the track radiation, and hence the estimation of vehicle contribution. For instance, from the modelling of the radiation directivity of the rail, Thompson (2009) found that the at lower frequency, 500Hz, for example, the radiation area of the rail just limits to the vicinity of the forcing point which is contradictory to the assumption in VTN that the rail can be seen as a line source radiating uniformly along infinite distance. Therefore, the model for the track radiation is an area that still needs further development. Figure 6 Comparison between VTN and TWINS results. Left: freight train. Right: passenger train (Verheijen and Paviotti, 2003). MISO method The so-called MISO (Multi-Input Single Output) method was developed by French National Railways (SNCF) under the STAIRRS project to assess separately the contribution of the infrastructure and the rolling stock. The basic idea of the separation method is also to access the track radiation from the noise measured by standard microphones supplemented by additional sensors: accelerometers to characterise the vibration of the track, a near-field microphone to measure the acoustic radiation, and devices to detect the wheel passing. At least two accelerometers are necessary: one mounted on the web and one on the foot of one rail at the midspan of two adjacent sleepers (refer to the positions of L1 and L2 or L2 and V2 in Figure 5). The deployment of the near-field microphone is 1.75 m from the track centre line and 0 m above the rail head (refer to the position of M2 in Figure 5). Another microphone to measure the total railway is installed in the same position as M1 in Figure 5. Four fundamental steps for the MISO method used by Létourneaux et al. (2002) are given below: 1. Assess the track transfer function between track vibrations and near field noise emitted midway between the bogies, assuming that the track is, at that time, the dominating source. 2. Derive the overall track contribution for a whole pass-by that is equal to the track transfer function times the track vibrations. 3. Infer the vehicle contribution from the difference between the total noise measured and the contribution of the track. 4. Apply the same weight (ratio vehicle/track contribution) to the standard microphone. MISO method is based on the use of cross-spectra and partial coherences between an acoustic pressure and accelerations. These quantities, which are computed using signal phase relationships, are very sensitive to background noise. Therefore, optimisation of all the parameters that could improve the phase relationship is critical. Compensating and essential measures may include using an improved dynamic range of data acquisition system (e.g., 24 bit D/A converter) and deploying more accelerometers (e.g., 5 vertical accelerometer and 5 lateral accelerometers) to improve multiple coherence. CRC for Rail Innovation 30/06/2011 Page 10

17 Chapter 3 Rolling noise identification and separation methods PBA Analysis Tool Pass-by Analysis (PBA) software was developed by TNO as a part of the STAIRRS project. The PBA method uses inputs from a single trackside microphone and an accelerometer underneath the railhead or foot. The measured noise and subsequent transfer function that connects the roughness (derived from vibration) is only representative of one particular train and track combination. No separation of track and vehicle contribution can be achieved unless one of them is substantially less than the other. However, in combination of silent track or vehicle method, vehicle and track contribution can be differentiated from each other. 3.3 Discussion and Implications for CRC Rail Project R1-105 In addition to the noise source separation methods introduced in Section 3.2, referring to default values from literature is also a means to obtain vehicle or track transfer functions. A comparison of 5 selected options in terms of accuracy, robustness, economy, and ease of use is given by Verheijen (2004) and is shown in Table 3. This shows that VTN, PBA with silent vehicle method and MISO are all promising with respect to trials in Australia, though each has its own particular limitations. Further development of these methods to meet the Australian context is necessary. CRC for Rail Innovation 30/06/2011 Page 11

18 Chapter 3 Rolling noise identification and separation methods Table 3 Review of source separation methods developed in Europe (Verheijen, 2004). Option no. Accuracy of Transfer Function Robustness Economy Ease of use Possible Limitations Vehicle Track Total effective roughness + VTN fair, but dependent on track noise Good reasonable, tested in STAIRRS WP2 validation campaign mainly measuring and analysis costs, trains can be measured in normal operation, no test runs measuring reliable track acceleration signals needs some care Oversimplified track radiation model may result in over-estimation of the track radiation Total effective roughness + MISO fair, but dependent on track noise good, but dependent on vehicle noise reasonable, tested in STAIRRS WP2 validation campaign similar as above option success depends on degree of coherence between track acceleration and nearby noise Reported to be sensitive to background noise (Létourneaux et al., 2002) PBA + silent vehicle method fair, but dependent on track noise good reasonable, tested in STAIRRS WP2 validation campaign only once, a silent vehicle measurement campaign needs to be arranged; thereafter costs similar as for the first option depends on organisational effort for silent vehicle; measuring reliable acceleration Difficult to obtain a silent vehicle signals needs practising TWINS calculation Good good excellent a new TWINS model requires thorough measurement effort, unless a validated reference model and library is available obtaining reliable TWINS results requires engineering knowledge and careful checking of No commercial software currently available each model Using default values from literature varying, dependent on difference between actual and default tracks and vehicles unknown no direct costs, but indirect costs may arise from inaccurate noise calculations fairly easy No database available for Australian context CRC for Rail Innovation 30/06/2011 Page 12

19 Chapter 4 Curve squeal and traction noise identification methods 4 Curve squeal and traction noise identification methods 4.1 Review of curve squeal noise identification methods Curve squeal identification includes detection of squeal events and squealing wheels. The intense mono-tonal noise can be detected easily by listening to the recorded sound or listening to noise emission from train pass-bys in the field. By standing in a position near the track, the squealing wheels can also be detected. Another kind of curving noise the broad band flanging noise can also be detected by hearing the sound. Dwight and Jiang (2009) automated the identification process and developed a system, known as the Trackside-Noise system or TN system. This is a single microphone-based system that is able to identify and differentiate the occurrences of curve squeal and flanging noise based on their spectrum characteristics: curve squeal has a dominant frequency component while flanging noise is broad band and more flat when expressed in the spectral domain. The TN system also attempts to associate squealing events with a specific axle by assuming that the maximum noise level occurs when the squealing wheel is near the microphone. Constant sound strength is therefore required if this assumption is tenable. However, this may not true when wheels on several adjacent axles squeal together. Failure of TN system under these circumstances has been reported by them. Jiang and Dwight (2010) have also developed a method to identify the squealing wheels by comparing the vibration level of inner rail and outer rail. The rail on the squealing wheel side is supposed to have much higher vibration than the other rail without wheel squealing and this assumption has been verified in their study. They also tried to use forces measured from both rails to identify the squealing wheels. Glocker et al. (2009) detected the squealing wheels through a similar method by comparing the acceleration measured from both rails. He used two microphones at each side of the track to record curving noise. Previous experience of curving noise identification therefore shows that a single microphone based system is able to identify the squealing events and squealing wheels. Additional information from rail vibration measurement and the detection of passing-by wheels is also required to improve accuracy and to establish which wheel on one axle is the generation source of the squeal. This can be achieved by installing accelerometers on both rails to measure rail vibrations and a wheel sensor on one rail to record the wheel trigger signals, as used by Jiang and Dwight (2010) and Glocker et al. (2009). The identification process can also be automated. 4.2 Review of Traction noise identification methods Traction noise consists of various components radiating from the traction, and auxiliary power systems of powered vehicles. Under different operating conditions, the actual source strength, timing and duration of each source may vary. Dittrich and Zhang (2006) have identified the main influencing parameters including speed and power of driveshaft and the time duration each source lasting. They developed models to predict traction noise based on these key parameters. Braking noise may also be classified under the general heading of traction noise. Brake noise prediction is also included in the models of Dittrich and Zhang (2006). In the Australian context, locomotive noise has been identified as a significant traction noise component requiring control. The physical position of locomotive on the train can be utilised to identify locomotive noise. CRC for Rail Innovation 30/06/2011 Page 13

20 Chapter 5 Review of railway noise identification systems 5 Review of railway noise identification systems Commonly used railway noise identification systems include: a single microphone system, microphone arrays, and microphones combined with sensors. 5.1 Single microphone systems A single microphone system is composed of one standard sound level meter and a microphone windshield recommended by the manufacturer of the sound level meter. It is able to measure the acoustic radiation superposed by all the noise sources in the vicinity. Measurement protocols with a single microphone system have already been standardised (e.g., AS ). A single microphone system cannot identify the locations of the source on its own and therefore may not provide sufficient information for noise source identification purpose. However, by the analysis of the measured sound profile in both the temporal and spectral domain, some noise sources can be identified, such as curve squeal and curve flanging noise. An example of this application is the Trackside-Noise system (TN) developed by Dwight and Jiang (2009). 5.2 Microphone arrays Microphone arrays offer the possibility of identifying both location and amplitude of a noise source as long as an appropriate shape arrangement is selected, depending on the locations of the radiation sources to be identified. This technique has been the main means to locate aerodynamic noise sources from high speed trains, such as ICE (Brug Hl and Schmitz, 1993), TGV (Van Der Toorn et al., 1996), and the maglev train (Degen et al., 2001). The use of a microphone array for rolling noise source localisation and quantification on a passing freight train was also performed, for example the T-array used in a research project sponsored by EU METARAIL (Dittrich and Janssens, 2000). Different shape arrangements can be used to locate noise sources in different directions. For example, a single horizontal line array was used by Barsikow (1996) to locate sources along the train, and a vertical line to locate the noise source heights. Two dimensional microphone arrays in T or X shapes, fuller star shapes or a spiral array can be used to locate the noise sources both horizontally and vertically, for example the spiral array used by Degen et al. (2001) to locate wheel-rail noise and also superstructure noise(see Figure 7). However, the spatial resolution of a microphone array is limited by the microphone spacing and array length at a given frequency. It is also frequency dependent: the higher the frequency, the higher the resolution (Barsikow, 1996). To cover a wide frequency range, several spacing is required. The analysis of the array signal in different frequency ranges enables classification of different types of source as illustrated in Figure 8. From Figure 8, it is apparent that in this case both pantograph and wheel/rail are important radiators with an indicated maximum noise level approximately 86 db. Microphone array based systems have also been used to study rolling noise, for example by Nordborg et al. (2001) and by Kiagawa et al. (2001). However, microphone arrays tend to imply that the wheel is the dominant source of rolling noise. From TWINS model, Thompson (1993 and 2003) found that the rail can be the dominant source in much of the frequency range (from ~0.5 to ~1.5kHz). The comparison of the microphone array measurements and TWINS model was studied by Kitagawa and Thompson (2006). The conclusions can be summarised as follows: the use of a beam-forming microphone array usually assumes that the sources to be identified consist of a distribution of uncorrelated point sources located in a plane at some known distance from the array. However, the sources on the rail are correlated during high frequencies where free-wave propagation occurs. Microphone arrays are only able to detect the radiation from the region near the forcing point. This can make the radiation power measured from microphone arrays 10 db lower than the actual rail radiation. This may explain why measurements using microphone arrays tend to emphasise the wheel as the dominant source. The CRC for Rail Innovation 30/06/2011 Page 14

21 Chapter 5 Review of railway noise identification systems directions for possible improvements in this method include adjusting the delays in the microphone array processing to direct it at other angles as well as normal to the track, and the modification of the data processing methods to take the coherent nature of sources into account. Another application of microphone array-based systems is to identify the curve noise associated with particular axles passing a site, RailSQAD, for example is a commercial system developed by Vipac Engineers and Scientists Ltd. This system is able to detect curve noise sources and classify them into curve squeal or flanging noise. In conclusion, some weaknesses of microphone arrays are: 1) The resolution of a microphone array is determined by the spacing distance and the ability of a microphone array to separate neighbouring sound sources is frequency dependent. 2) The expense of microphone arrays is relatively high and requires a high-capacity data acquisition and analysis system proportional to the number of microphone inputs. Figure 7 Spiral array developed at DB-AG (Nordborg et al., 2000) Figure 8 Example of a measurement result on ICE3 with the spiral array. Max Level 86 db(a), dynamic 8 db, v = 330 km/h (Van Beek et al., 2002) 5.3 Combined sensors Though microphone arrays are accurate enough to identify the physical distributions and magnitudes of various noise sources, like the aerodynamic noise and locomotive noise, it remains difficult to discriminate the track and wheel contribution (Figure 8). Alternative techniques involving a combination of different sensors have been shown to be more effective. The applications include Vibro-acoustic Track Noise method (VTN) (Verheijen and Paviotti, 2003), MISO method (Létourneaux et al., 2002), and the PBA software tool (Dottrich et al., 2003) to identify rolling noise and verify effectiveness of noise mitigation measures, the details of these methods are shown in Section 3.2. These techniques have been investigated and developed as a protocol in Harmonoise (Van der Stap et al., 2004). Other applications by combined sensors include assessing noise in sharp curves (Vincent et al., 2006; Glocker et al., 2009), as discussed in Section 4.1. CRC for Rail Innovation 30/06/2011 Page 15

22 Chapter 5 Review of railway noise identification systems 5.4 Comparison of existing knowledge and implications A comparison of single microphone systems, microphone arrays and combined sensors is given Table 4. Four aspects: Efficiency, Economy, Application range, Ease of use and localisation are compared. Table 4 Comparison of current noise identification systems Method Capability Identifiable Noise Sources Cost Ease of use Noise source Localisation 1.single microphone system Total sound level measurement, identify curve squeal General locomotive noise, curve squeal, curve flanging Low High No 2.microphone arrays Identify precise locations and magnitudes of various noise sources Aerodynamic noise, traction noise and rolling noise High Low Yes, limited 3.combined sensors (like accelerometers) + a single microphone Wheel/rail contribution separation, identify important parameters possibly influencing curve squeal noise, like Angle of Attack (AoA), and wheel position Rolling noise, curve squeal noise, locomotive noise Medium Medium No Given that a range of sources are required to be identified and that rolling noise separation is important, a single microphone is not suitable on its own. The application of microphone arrays is constrained by high expense. Combined sensors cannot localise various noise sources directly, but with specifically designed algorithms, they are able to identify the occurrence and magnitudes of various noise sources at a relatively low cost. For this approach to work, some knowledge of the underlying mechanism and the use of logic related to the timing of the event relative to others are required. The most promising system for trial is therefore method 3 in Table 4, which can identify rolling noise, curving noise and locomotive noise with a medium expense. CRC for Rail Innovation 30/06/2011 Page 16

23 Chapter 6 Proposal of a Prototype Noise Monitoring System 6 Proposal of a Prototype Noise Monitoring System Objectives: A specific objective of this project is to develop a noise monitoring system that is able to: 1. Separate wheel and rail contribution to rolling noise and estimate wheel and rail combined roughness; 2. Identify curve squeal; and, 3. Identify locomotive noise. The system is also required to be practical to set up temporarily at a site to allow noise surveys to be conducted by rail organisations. Methodologies for noise identification/separation: 1. Rolling noise: rolling noise separation into wheel and rail contribution can be achieved by Vibro-acoustic Track Noise method (VTN) (Verheijen and Paviotti, 2003), Multi-Input Single-Output method (MISO) (Létourneaux et al., 2002) and Pass By Analysis method. By further attributing to the wheel and rail roughness, rolling noise can be investigated at a level of excitation sources. A possible method available to achieve this purpose includes the indirect method developed under METARAIL project, Europe (Dittrich and Janssens, 2000; Janssens et al., 2006). The method only requires inputs of rail vibration and wheel trigger signals, both of which can be easily accessed by the prototype system proposed here. However, supplementary rail or wheel roughness from direct measurement is still required in order to derive the other component. Direct rail roughness is recommended owing to its ease of access. 2. Curve squeal: Curve squeal identification is through detection of dominant frequency component of the recorded sound. By using signals from the axle counters and by comparing the noise intensity and acceleration levels of the measurements taken at the inside and outside of the curve, squealing wheels can be identified. 3. Locomotive noise will be identified by detecting the physical positions of locomotives on the train body. The algorithms for these measurement methodologies will need to be packaged together into a software kit with a user-friendly interface that can be easily accessed. System requirements: It is proposed to base the prototype system on noise measurement from a single microphone, with additional information from rail vibration and wheel trigger signal measurements. This conclusion is based upon comparison of several current noise identification systems: microphone arrays can identify both the locations and magnitudes of various noise sources, but are expensive to implement and the resolution in separating wheel/rail contribution is limited. A single microphone system is able to measure the total sound level near the microphone. With additional time-frequency analysis, it is possible that a single microphone could also be used to identify locomotive and squeal noise. Combining sensors with microphones can separate rolling noise components in addition to identifying noise sources, while the measurement set-up is still relatively straightforward. Instrumentation and installation: Figure 9 and Table 5 illustrate the instruments required and their installation details for the noise monitoring system. CRC for Rail Innovation 30/06/2011 Page 17

24 Chapter 6 Proposal of a Prototype Noise Monitoring System L2, V2 Travel S 1 T 1 L1, V1 V L M T: Wheel V & L: M:Microphone, 2m/1.2m Table 5 Instruments and installations for noise prototype system Figure 9 Instrumentation for noise prototypes system Signals Sensor type Location 1 Single direction Accelerometer Rail web in the mid-span of two adjacent sleepers 2 Single direction Accelerometer Rail foot in the mid-span of two adjacent sleepers 3 Single direction Accelerometer Rail web in the mid-span of two adjacent sleepers (another rail) 4 Single direction Accelerometer Centre of the underside of the rail foot mid-span between adjacent sleepers (another rail) 5 Single direction Accelerometer On the top surface of a sleeper, near the fastener, as per the diagram 6 Omnidirectional Microphone 2m* from the track centre line and 1.2m above the top of rail 7 Radar speed Sensor Actual location is not critical. Install as convenience. 8 Wheel sensor Top surface of rail foot in the mid-span of two adjacent sleepers 9 Weather Station Adjacent to the test site R: Speed radar All sensor signals are recorded using a common time base *Note: no specification for the distance which may be determined at the site to both facilitate the installation and ensure a better measurement result. Minimum inputs for rolling noise separation into wheel and rail radiation include signals 1, 2, 6, 9; however, if more accurate results are required, signals 3, 4 and 5 are required. Minimum required inputs for wheel and rail combined roughness determination include signals 1, 2, 7, 8. If roughness from another pair of wheel and rail combination is to be known, inputs of signals 3, 4 are also required. CRC for Rail Innovation 30/06/2011 Page 18

25 Chapter 6 Proposal of a Prototype Noise Monitoring System Minimum required inputs for curving noise identification includes signals 1, 3 or (2, 4), 6, 8, 9. Minimum required inputs for locomotive noise identification include inputs from signals 6, 8. CRC for Rail Innovation 30/06/2011 Page 19

26 Chapter 7 Mitigation measures for rolling noise 7 Mitigation measures for rolling noise Several European studies (e.g., Thompson and Gautier, 2006) on railway noise reduction confirmed that mitigation measures aimed at reduction at source can be more cost-effective than constructing noise barriers. One representation of cost-effectiveness of various noise control measures is shown in Figure 10. It indicates that track measures in combination with rolling stock measures are relatively high benefits/cost ratio, and noise barriers up to 4 m high have poor efficiency. The best effectiveness for the European context was determined to be with combined measures, wheel roughness reduction through composite brake blocks, optimised wheels, tuned rail absorbers, grinding, and noise barriers not higher than 2 m. The state of the art of noise mitigation and elimination at the source is reviewed in the following sections. 7.1 Implications from the TWINS model Figure 10 Main results of the STAIRRS project (Oertli and Hübner, 2007) The structure of the rolling noise generation mechanism is typically explained according to Figure 11, from which various parameters influencing rolling noise can be identified. Implications for noise reduction from these parameters can be drawn from this diagram. First, there is a potential to reduce the excitation due to roughness. This will influence all of the following noise generation process. Different measures can be used to reduce wheel/rail roughness such as rail grinding and wheel turning. Next, consideration may be given to various measures that can reduce the wheel/rail radiation responding to a given roughness excitation. These measures include adding damping to wheel and rail. It is important to note that, given that vibration energy introduced into the rail will eventually be extracted as either noise or transmitted to contacting components (particularly the rail pads and then to the sleepers and ballasts), the decay rate of the rail vibration as a function of the distance from the source (usually expressed in db/m) is relevant (Jones et al, 2006). Thompson (2006) found that a doubling of the decay rate in a particular frequency band can reduce the rail radiation noise in that band by 3 db. Rail pad stiffness and rail damping are widely accepted and will obviously be relevant to increasing this decay rate. Soft pads or resilient baseplates may reduce the loads applied to sleepers and ballasts but lead to relatively low decay rate, while stiff pads will increase the decay rate but also the dynamic coupling between the sleepers and the track which can increase the sleeper contribution instead. To achieve best performance, a compromise between pad stiffness and sleeper CRC for Rail Innovation 30/06/2011 Page 20

27 Chapter 7 Mitigation measures for rolling noise coupling was adapted by Vincent et al. (1996). Medium stiff pads were used to balance the noise contributions from the rail and sleepers. Figure 11 Schematic view of the model for rolling noise, showing the main potential means of reducing rolling noise (Thompson, 2009). A further and usually relatively costly mitigation measure is local or general shielding of the radiated noise from the receiver, i.e., interference in the propagation path of the noise. These include skirts and barriers close to the rails, and noise barriers or embankments adjacent to the rail corridor. However, they are usually expensive to implement and maintain and may cause over-shadowing and negative aesthetic impacts. This report concentrates on noise reduction measures at source. 7.2 Controlling of rolling surface roughness Wheel/rail contacting surface roughness directly affects the generation of rolling noise (Figure 11). According to Thompson (1996), under normal operating conditions, the rolling noise level is proportional to the combination of the wheel and rail interface roughness amplitude. This indicates that reducing the wheel/rail roughness can be a very effective way to control the rolling noise. The most notable approach to roughness control is on the trains where cast iron tread brakes were in service, which is particularly common on European freight wagons. The mitigation measure in this instance is the introduction of more suitable friction materials. This can lead to reductions in noise levels on good tracks of up to 10 db (Hemsworth, 1979). Cast-iron blocks have been shown to cause a rapid in-service increase in wheel tread roughness (and hence rolling noise), whereas composite blocks do not have this effect. It should be noted that cast-iron tread brake blocks are not used in Australia, so this approach does not offer similar benefits in the Australian context. For rail roughness control, the commonly used mitigation technique is rail grinding. According to Vincent (2000), noise reduction of up to 10~15dB (A) can be achieved with appropriate rail grinding at corrugated sites with good wheel conditions. Reduction by 3-6 db (A) is more typical, on regularly maintained networks. By use of traditional grinding method, i.e., the rotating grinding stones, the rail in poor condition can be grounded to a considerably smoother level; however, a noticeable tonal peak is left. The tonal peak is due to the resonance of the grinding stones and their drive system, its wavelength depends on the speed of the grinding and typically occurs at ~20 to CRC for Rail Innovation 30/06/2011 Page 21