Quantifying the annoyance caused by flats on the wheels of railway vehicles Manfred Liepert 1 Moehler + Partner Ingenieure AG Prinzstraße 49, 86153 Augsburg, Germany Christine Huth 2 Moehler + Partner Ingenieure AG Prinzstraße 49, 86153 Augsburg, Germany ABSTRACT As part of a research project by the Federal Environment Agency, measurements of the pass-by noise of railway vehicles with and without flats on the wheels were conducted. The aim of the project was to define an acoustic maintenance criterion for the wheels. To do so, listening tests were carried out to quantify the annoyance of the noise caused by a flat. In order to determine the level adjustment equivalent to the increased annoyance, the following method of adjustment was used. The test sub- jects were each presented a pair of sounds consisting of a vehicle passing with and a vehicle passing without a flat. By raising or lowering the level of the signal without a flat, the test subjects were able to adjust the two signals individually to “equal annoyance”. Furthermore, different noise levels and psychoacoustic indices were tested in order to describe the impact of the flat on the caused annoy- ance. 1. INTRODUCTION

As part of a research project by the German Federal Environment Agency (Umweltbundesamt) on the subject of "measuring wheel flats and defining an acoustic maintenance criterion", listening tests were carried out to quantify the annoyance of passing trains with a so-called flat spot (wheel out-of- roundness). The aim of the tests was to quantify the annoyance of audible flat spots and to propose a method for evaluating the annoyance based on a level adjustment. 2. AIM OF THE STUDY

With the conversion of the freight wagons running on the German rail network to low-noise braking systems, a significant contribution was made to the noise protection target 2020 [1] of the Federal Ministry of Transport and Digital Infrastructure, the reduction of railway noise by half. The "loud" block brakes with cast iron brake blocks that had been the standard up to that point were replaced by so-called LL blocks (for retrofitting) or K blocks (for the commissioning of new freight cars). The reduced roughening of the wheel treads by these new brake blocks results in a reduction in rolling

1 Manfred.liepert@mopa.de

2 Christine.huth@mopa.de

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

noise of around 5 dB(A) at 100 km/h on an average track. Depending on the condition of the track, significantly higher reductions in level can be determined by measurement. In the course of the conversion, freight wagons with LL blocks or K blocks and cast iron blocks can be coupled in a train set. Due to the different braking curves (dependence of the speed of a braking train on the braking distance), individual wheel sets can be blocked. When the locked wheel slides on the rail, the softer wheel wears down, because the wheels are made of a softer steel than the rail. This creates what is known as a flat spot. In addition to this mechanism of the formation of wheel flats, there are other mechanisms of formation of different wheel shape errors that lead to a noticeable increase of rolling noise. With the successful reduction of the undisturbed rolling noise, the flat spot noises are now subjec- tively more prominent and are perceived as annoying by residents. In this study, the annoyance caused by these noises are evaluated. The two main questions are: • How can the annoyance of the flat spots noise be quantified in terms of a level adjustment? • Which acoustic measure is most suitable for predicting the annoyance of flat spots? 3. DATA

For the subjective evaluation of the noise caused by wheel flats, listening tests were carried out in the laboratory as part of the research project. A survey among residents on the salience of flat surface noises did not appear to be expedient due to the unpredictable occurrence of flats. For the listening tests, recordings of passing trains with simultaneous detection of flat spots in the train were therefore required. The recordings were made over a period of one month at the railway line between Munich and Rosenheim. The measurement campaign was carried out in cooperation with Müller-BBM Rail Technologies GmbH in the immediate vicinity of the monitoring station of the Federal Railway Authority (Eisenbahn-Bundesamt) [2]. The wheel monitoring system (WMS) [3] was used to identify flat spots. This detection was validated and enhanced by signals from accel- eration sensors on the rail foot and microphone signals. The microphone signals were recorded at a distance of 7.5 m and 25 m from the measured track.

WMS WMS-Database

left

right

7,5 m Microphone signal

25 m

left

Acceleration signal

right

right rail left rail

Figure 1: Measuring setup for recording trains passing by with simultaneous flat spot detection

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

During the measurement period of one month, a total of around 3,000 train passes were recorded, 1,130 of which were freight trains and 1,825 passenger trains. 3. LISTENING TESTS

A selection was made from the recordings of trains passing by, which were recognized by the WMS as passings with a wheel flat and whose audio recording contained a perceptible (i.e. noticeable) noise from the wheel flat.

3.1. Experimental design The listening tests for the subjective evaluation of the flat spot noises were carried out in the listening booth of Möhler + Partner Ingenieure AG. The listening booth is characterized by a quiet and visually neutral environment. The sounds were presented using electrostatic headphones from the Lamda se- ries by Stax. A total of 20 subjects aged between 25 and 51 took part in the listening tests.

Figure 2: Listening booth of Möhler + Partner Ingenieure AG; The sounds are presented using Stax electrostatic headphones

3.2. Experimental layout for the level adjustment A tracking method was used to assess the annoyance of a flat spot noise in the form of a level adjust- ment. For this purpose, a train pass that was as inconspicuous and uniform as possible was selected as the reference stimulus without a flat spot. A total of 14 noise sections with flat spot noises were extracted from the data pool of train passes with recognized flat spots and then mixed into the reference stimulus. Care was taken to ensure that the reference drive-by and flat spot sample were recorded at a comparable drive-by speed. This resulted in sound pairs of the same train passing with and without a flat spot. Initially both signals of the sound pair are presented with the same level control, so that they only differ in the mixed-in flat spot.

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

90

S0

dB(A)

80

70

60

0 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 s

90

S7

dB(A)

80

70

60

0 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 s

Figure 3: Reference pass in the original without a flat spot (above) and with a flat spot sample mixed in (below) In the listening test, the sound pairs were presented in direct succession as a sequence to the test subjects. The task was to then assess whether the second signal without a flat spot (the reference train passing) was perceived as less annoying, equally annoying or more annoying than the first signal with a flat spot.

Signal 1 Signal 2 Beurteilung

Assessment

Train passing with flat spot Train passing without flat spot

Is signal2 in comparison to signal 1:

∆ L

less annoying / equalliy annoying/ more annoying?

sequence

Figure 4: Schematic representation of a sequence for the listening test using a tracking method. De- pending on the answer, the level of signal 2 is adjusted in the next sequence.

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

Depending on the test subject's assessment, the level of the second signal is raised or lowered for the presentation of the sequence in the next repetition step. The amount of level increase or decrease is reduced with each repetition, so that the level of the second signal adapts to the same perceived an- noyance as the signal with the flat spot. Table 1: Exemplary response curve of the test subject with associated level adjustment SEQUENCE RESPONSE OF THE SUBJECT ADJUSTMENT OF SIGNAL 2

7 8 signal no. 10 11 12 13 14 15

Sequence 1 Signal 2 is less annoy- ing than Signal 1 + 2 dB

Sequence 2 Signal 2 is more annoy- ing than Signal 1 - 1 dB

Sequence 3 Signal 2 is less annoy- ing than Signal 1 + 0,5 dB

Sequence 4 Signal 2 is more annoy- ing than Signal 1 - 0,5 dB

Sequence 5 Signal 2 is equally an- noying than Signal 1 Result: ∆ L = 1 dB

The level differences ∆ L resulting from the tracking method indicate the required increase (+) or decrease (-) of the reference train passing compared to the passing with a flat spot in order to achieve the same annoyance. For the 16 sound pairs of the test series, the medians per test person and the median for all test persons are shown in the following figure:

-je—eee—see goo cree

Figure 5: Results for the annoyance adjustment; Green: results of the listening tests on average (median and interquartile), dark blue: results of the individual test subjects This resulted in level differences of up to 2.5 dB on average, depending on the respective flat spot. In order to check whether the test persons produce reliable results, the sequence was also presented to the subjects with both signals consisting of the reference stimulus. This comparison corresponds

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

to "Signal 0", for which a level difference of 0 dB was correctly adjusted by the test subjects. If the adjusted level difference is compared to the difference between the drive.by without and with a flat spot in the A-weighted maximum level, the result is the relationship shown below.

AL max [4B(A)]

Figure 6: Necessary level adjustment for a drive-by without flat spot to produce the same annoy- ance as a drive-by with a flat spot, plotted against the difference in the A-weighted maximum level L Amax of drive-by with and without the flat spot

In order to compensate for a level difference in the maximum level of around 10 dB caused by the flat spot in the perceived annoyance, the overall signal without a flat spot must be raised by 2.5 dB. An annoyance allowance in dB could therefore be based on the increase in the maximum level caused by the flat spot and estimated with the relation:

      0,25 ∙∆
  (1)

3.3. Experimental layout for the identification of suitable acoustic measures In order to identify the most suitable acoustic measure for predicting the annoyance of a flat spot, the random access method [2] was used as psychometric method for the listening test. For this purpose, 15 signals were presented as sound symbols to the test subjects in an anonymized and randomized order. Subjects could listen to them as often as required, then the symbols had to be sorted by “drag and drop” in order of increasing annoyance. The layout of the test is shown in Figure 7.

35 25 15 fapliv

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

Least annoying Most annoying

Figure 7: Layout of the test according to the random access method

The result of the listening tests is a ranking of the annoyance of the 15 signals for each test person. Figure 8 shows the value (median) for each of the signals averaged over all subjects.

Figure 8: Rank in annoyance (median and interquartile) averaged over all subjects for each of the 15 evaluated signals

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

In the following, these results are shown graphically over the maximum values of the acoustic and psychoacoustic measures in order to determine the correlation between the annoyance and the respective measure. In accordance with the literature [5], the highest correlation results for psychoacoustic loudness.

16 Rank in Annoyance Ratings 35 04 os 06 =o? Maximal Fluctuation Strength in vacil ” 5 Maximale Sharpnesin acum ® 09 0,663 65 aa

130 120 110 190 90, ‘Maximal Loudness in Sone 80 70 60 Souekouuy ul yuey sSuney aoueAouuy ul yt

“ s8uney douedouuy ur yuey 86 88 90 92 94 L_AFmax in dB 84 82 ™ sauney

Figure 9: From top to bottom, rank in annoyance over: A-weighted maximum sound pressure level L AFmax , maximum loudness, maximum fluctuation strength, and maximum sharpness. A linear re- gression line and the associated coefficient of determination are also shown.

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

An analysis of variance using multiple regression was carried out in order to check whether the inclusion of further measures could improve the result. The regression resulted in an explained variance between the rank of annoyance and loudness of 76%. The additional influence of sharpness and fluctuation strength on the explained variance between the rank in annoyance and loudness was calculated by means of multiple regressions. The sharpness has an additional influence of 4%, which corresponds to an explained variance of 80% when sharpness and loudness are considered together. The fluctuation strength has an additional influence of 13%, which corresponds to an explained variance of 89% when considering the combination of loudness and fluctuation strength. When considering all 3 perception parameters (loudness, sharpness, fluctuation strength) there is a variance explanation of 91%. The resulting values are shown in Table 1.

Table 2: Multiple regression analysis

Loudness + Roughness

Loudness Loudness + Sharpness

Loudness + Fluctuation Strength

Loudness + Sharpness + Fluctutation Strength

Explained Variance of the Rank in Annoyance

76% 80% 89% 77% 91%

Additional Explained Vairance

+ 4% + 13% + 1% + 15%

4. CONCLUSION

In this study, the annoyance of passing trains with flat spots was examined by means of listening tests. In summary, it can be stated that a train passing with a flat spot is comparable to a train passing without a flat spot, the level of which is increased by 25% of the difference in maximum level caused by the flat spot. Furthermore, the maximum values of different acoustic and psychoacoustic measures of a train pass- ing with flat spot were used as the basis for a regression analysis for the annoyance. There was a high correlation between the annoyance of a flat spot assessed in the listening test and the calculated max- imum loudness of the flat spot, but also with the A-weighted maximum level of the flat spot. If both the loudness and the degree of fluctuation are used to describe the annoyance of a flat spot, a variance explanation of 89% can be obtained.

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW

5. REFERENCES

1. BMVI, 2020. Die Schiene wird leiser! Ziel des BMVI bis 2030: Um die Hälfte weniger Lärmbe- troffene! [online]. 14. Dezember 2020 [last call on: 10. August 2021]. Not anymore available: https://www.bmvi.de/SharedDocs/DE/Artikel/K/virtuelle-pk-leise-schiene.html 2. EISENBAHN-BUNDESAMT, 2022. Lärm-Monitoring . Schallmessungen im Schienenverkehr [online]. [Zugriff am: 31. März 2022]. Verfügbar unter: https://www.laerm-monitoring.de/ 3. Müller-BBM Rail Technologies GmbH: Wheel Monitoring System (WMS), 2021. https://www.muellerbbm-rail.de/produkte/wheel-monitoring-system/ 4. Fastl, H.: Psychoacoustics and Sound Quality. In: Fortschritte der Akustik. DAGA, Dt. Gesell- schaft für Akustik e.V. (2002), S. 765–766 5. Widmann, U.: Ein Modell der Psychoakustischen Lästigkeit von Schallen. Dissertation. TU Mün- chen 1992 6. Huth, Ch., Forstreuter M., Liepert, M., Arlt, R.: Messung von Flachstellen und Ermittlung eines akustischen Instandhaltungskriteriums. Abschlussbericht des Forschungsvorhabens des Umwelt- bundesamts FKZ 3718 54 102-0, Veröffentlichung 2022.

inter.noise. 21-24 AUGUST SCOTTISH EVENT CAMPUS ? O? ? GLASGOW