Experimental tests of a multichannel active noise control system with single and multiple reference signal applied to the cabin of a tractor Francesco Mori 1 , Patrizio Fausti 2 , Francesco Pompoli 3 , Andrea Santoni 4 University of Ferrara, Engineering Department Via Saragat 1, 44122, Ferrara, FE, Italy Paolo Bonfiglio 5 Materiacustica s.r.l. Via C. Ravera 15/A, 44122, Ferrara, FE, Italy Pietro Nataletti 6 INAIL, National Institute for Insurance against Accidents at Work Via S. Gradi 55, 00413, Rome, RM, Italy

ABSTRACT A feedforward active noise control system represents a valid solution whenever it is necessary to attenuate a disturbing noise at low frequencies. One example of application is the case of tractors, in which the agricultural operators are exposed for a prolonged time to loud low-frequency noise, mainly generated by the engine. Generally, the feedforward systems make use of one reference sensor to measure the noise produced by the main source. However, this way of proceeding could be inadequate to describe the disturbing noise, especially when the noise consists of both structure- borne and air-borne components. In this article, a comparison between the use of single and multiple reference signals, processed in a multichannel active noise control system applied to the cabin of a motionless tractor, is proposed. In particular, an accelerometer positioned on the turbocharger is used to sense the vibration generated by the engine, while a microphone placed near the cooling fan is used to sense the air-borne noise. The active system implements the filtered-X least mean squares (FXLMS) algorithm to produce the anti-noise signals. The results show that the use of two simultaneously acquired reference signals generates only small benefits in the performance of the active system.

1 francesco.mori@unife.it 2 patrizio.fausti@unife.it 3 francesco.pompoli@unife.it 4 andrea.santoni@unife.it 5 paolo.bonfiglio@materiacustica.it 6 p.nataletti@inail.it

1. INTRODUCTION

The noise transmitted inside the cabin of the tractors generally has a predominance of low-frequency components associated with the engine orders, typically up to 1000 Hz. Passive treatments of the cabin for the attenuation of these components are not feasible due to the limited space inside the volume and due to the necessity of guaranteeing transparency of the windowed walls for visibility while maneuvering. Moreover, passive solutions are traditionally ineffective in the low-frequency range associated with the first orders of engine noise. On the other hand, active noise control (ANC) represents a valid solution to attenuate low-frequency noise levels. Such an approach can significantly reduce an undesired sound by exploiting the superposition with an opposite-phase signal, calculated starting from measured signals, which are processed in real-time. Due to the low sampling rate, necessary to process the signals in real-time, such a solution is suitable in the low-frequency range, where passive solutions cannot effectively reduce the disturbing noise. Furthermore, thanks to the use of an adaptive algorithm, the ANC system can dynamically track the variations of the disturbance source, instead of acting on a fixed frequency target as the traditional solutions. In the literature, only a few studies about ANC applications on tractors with a cabin can be found, implementing a feedback ANC system with two error microphones and two control sources, in which the reference signal is internally modeled [1, 2]. Feedforward ANC systems have been studied for applications on similar machines, endowed with a cabin, such as a fork-lift [3] or a harvesting machine [4], with one reference sensor, aimed to cancel the harmonic components of the engine noise. However, it is worth noticing that the engine is only one of the sources which affect the operator’s hearing inside the cabin and the noise propagation can be both air-borne or structure-borne.

In this article, a multichannel feedforward ANC system applied to a tractor with a cabin is studied, comparing the effect of sensing one or two reference signals. In particular, one accelerometer detects the vibration, induced by the engine and propagating on a structural path, while a microphone identifies the air-borne component. The experimental tests are carried out on a motionless tractor, varying the number of rounds per minute of the engine, with the aim of canceling the first engine orders in the frequency range up to 1000 Hz at the driver’s head position. The ANC system uses a filtered-X least mean squares (FXLMS) algorithm with two error microphones and two control loudspeakers. The characteristics of the algorithm are described in Section 2, while the experimental setup involved in the ANC system is illustrated in Section 3. The results of the tests and the comparison between the use of one or two reference sensors are reported in Section 4. Finally, Section 5 illustrates the conclusions and the next steps of the development. 2. DESCRIPTION OF THE ALGORITHM

The algorithm implemented in the feedforward ANC system is the classical FXLMS, for its simple scalability in the case of multiple sensors and actuators and for its robustness. Let’s consider a system composed by 𝑁 𝑟 reference sensors (in this case, microphones and/or accelerometers), 𝑁 𝑒 error microphones and 𝑁 𝑠 control sources. Moreover, we indicate with 𝐿 the length of the adaptive filters and with 𝑀 the length of the filters representing the secondary paths (i.e., the impulse responses from the 𝑗 -th control source to the 𝑖 -th microphone). Extending the algorithm operations illustrated in [5] to a more general case, the following real-time processing operations can be deduced. Firstly, the filtered reference signals 𝒙 𝒌,𝒋𝒊

′ (𝑛) = [𝑥 𝑘,𝑗𝑖

′ (𝑛) 𝑥 𝑘,𝑗𝑖

′ (𝑛−1) … 𝑥 𝑘,𝑗𝑖

′ (𝑛−𝐿+ 1)] are computed by convolving the estimations of the secondary paths 𝒔 𝑗𝑖

′ = [𝑠 𝑗𝑖,0

′ 𝑠 𝑗𝑖,1

′ … 𝑠 𝑗𝑖,𝑀−1

′ ] with the k -th reference signal 𝒙 𝒌 (𝑛) = [𝑥 𝑘 (𝑛) 𝑥 𝑘 (𝑛−1) … 𝑥 𝑘 (𝑛−𝐿+ 1)] , as in Equation 1:

𝑀−1

′ (𝑛) = ∑𝑠 𝑗𝑖,𝑚

′ 𝑥 𝑘 (𝑛−𝑚)

𝑥 𝑘,𝑗𝑖

. (1)

𝑚=0

The secondary paths are estimated with the offline procedure described in reference [5], by emitting a broadband noise from the control source and estimating the filter with a least mean squares approach. In a second moment, the coefficients of the adaptive filters 𝒘 𝑘𝑗 (𝑛) =

[𝑤 𝑘𝑗,0 (𝑛) 𝑤 𝑘𝑗,1 (𝑛) … 𝑤 𝑘𝑗,𝐿−1 (𝑛)] are updated with Equation 2:

𝑁 𝑒

𝑤 𝑘𝑗,𝑙 (𝑛+ 1) = 𝑤 𝑘𝑗,𝑙 (𝑛) −𝜇∑𝑒 𝑖 (𝑛)𝑥 𝑘,𝑗𝑖

𝑙= 0, 1, ⋯, 𝐿−1, (2)

′ (𝑛−𝑙)

𝑖=1

where 𝑒 𝑖 (𝑛) is the i -th measured error signal, while 𝜇 is the step-size parameter, which regulates the speed of convergence and the stability of the system. Finally, the output of the j -th control source is computed as:

𝑁 𝑟

𝐿−1

𝑦 𝑗 (𝑛) = ∑∑𝑤 𝑘𝑗,𝑙 (𝑛)𝑥 𝑘 (𝑛−𝑙)

(3)

𝑙=0

𝑘=1

These operations are repeated for any generic time instant n .

The sampling frequency chosen for the ANC system is 6400 Hz and, ideally, it allows the system to be effective up to 1600 Hz (a quarter of the sampling frequency). However, to limit the instability effects related to the high-frequency components and considering that the driver is mainly affected by the components up to 1000 Hz, a low-pass filter with a cut-off frequency of 1000 Hz is introduced on the reference signals before processing. 3. EXPERIMENTAL SETUP AND TESTS

The experimental setup adopted for this study is illustrated in Figure 1. The reference sensors, measuring the reference signals, are located outside the cabin, in the proximity of the engine, and consist of a microphone placed near the cooling fan, outside the case containing the engine, and an accelerometer on the outlet of the turbocharger. The zone of quiet, where the noise should be cancelled, is defined by two error microphones placed next to the position of the operator’s head, hanging from the cabin sunroof, as shown in Figure 1b). The acquired signals are processed in real- time by a National Instrument cRIO 9063 hardware, embedded with FPGA (Field Programmable Gate Array), which treats the measured data in a deterministic way, executing most of the operations on simultaneous parallel circuits. The computed outputs, given in Equation 3, drive the control sources shown in Figure 1b), embedded with an amplifier and placed on a shelf mounted behind the seat. The cut-off frequency of the control loudspeakers, above which they have a flat frequency response function, is about 100 Hz. This characteristic affects the effectiveness of the ANC system in the frequency range below 100 Hz, in which the system cannot provide sufficient power, as shown in the results section.

Figure 1: Experimental setup implementing the ANC system on a tractor: a) outdoor reference

transducers; b) indoor error transducers and control sources.

The target of the ANC system is to reduce the noise inside the tractor cabin at the driver’s head position. The tractor is equipped with an in-line six cylinders diesel engine, with a combustion every 120° of the crankshaft rotation. Therefore, the combustion orders are expected at multiples of three times the rotational frequency of the crankshaft. The tests are carried out for three different stationary values of rounds per minute of the crankshaft, with a minimum value of 763 rpm, a mean value of 1495 rpm and a maximum value of 2340 rpm. This choice ideally reproduces the operative conditions of the tractor, where the rounds per minute of the engine are usually set on a steady value. The tests are repeated using a single reference sensor (either the accelerometer on the turbocharger or the microphone near the cooling fan) and simultaneously the two reference sensors.

4. RESULTS

The results of the tests are shown in this section. As previously mentioned, the ANC system becomes very effective starting from the third-octave band of 100 Hz due to the limitation introduced by the frequency response function of the control loudspeakers. Figure 2 shows the results associated with the minimum value of rpm. The simultaneous use of the two reference signals allows for a slightly more effective noise reduction between the third-octave bands of 80 Hz and 160 Hz on the left-hand side, while no significant differences can be observed on the right-hand side between the setups with one or two reference signals. This value of rpm represents the minimum operativity condition of the engine. Thus, the other noise sources sensed by the microphone (e.g., the cooling fan) have a relevant weight and can effectively be attenuated.

Figure 3 shows the results associated with 1495 rpm. It is not possible to identify, in the case of a single reference sensor, whether using a microphone or an accelerometer would be more convenient; moreover, the simultaneous use of both reference sensors provides a limited improvement of the performance of the ANC system. The ANC efficiency is particularly evident on the right-hand side, in Figure 3b), between the third-octave band of 80 Hz and 400 Hz. Finally, Figure 4 illustrates the results at 2340 rpm. The engine is at its maximum operativity condition, masking the other noise sources sensed by the microphone, which could be attenuated in the previous cases. Thus, no improvement can be observed by using both reference sensors rather than a single one. From Figure 4, it can be evidenced that the system is very effective on the third-octave band of 125 Hz, where the third engine order is present. In particular, a reduction of more than 10 dB on the left-hand side and 3.5 dB on the right-hand side can be obtained for all the configurations.

Figure 2: Third-octave band sound pressure level spectra measured at 763 rpm: a) left error microphone; b) right error microphone.

Right Enor - 763 pm cor oe sto

Figure 3: Third-octave band sound pressure level spectra measured at 1495 rpm: a) left error microphone and b) right error microphone.

Left Eror = 1498 rpm Taxe onto SANC_On Tatton ‘a a too Right Eror- 1498 pm eeesensgssess es axe off Pree on mute SAxC on Fo ave. On Tatton y wy

Figure 4: Third-octave band sound pressure level spectra measured at 2340 rpm: a) left error microphone and b) right error microphone.

Tables 1 and 2 report the achieved attenuations, respectively in terms of the overall linear and A- weighted levels. At 763 rpm, the noise reduction is not relevant because the sound pressure level is dominated by the third combustion order in the third-octave band of 40 Hz, outside the range of frequency where the control sources can work efficiently. At 1495 rpm the attenuation of the A- weighted level is similar for all the studied configurations, while the use of two reference sensors improves the attenuation on the linear overall sound pressure level, in particular on the right-hand side, thanks to the action of the ANC system on the third-octave bands centred around 80 Hz, 125 Hz and 160 Hz. At the maximum operativity condition, the performance increases in terms of A-weighted levels, because the main spectral components concurring at the overall level are shifted towards higher frequencies.

Table 1: Attenuation of the overall non-weighted sound pressure level.

Turbocharger Fan Turbocharger+Fan rpm Left Right Left Right Left Right 763 1.4 -0.2 2.7 2.3 0.1 1.2 1495 1.0 1.3 -0.3 3.1 1.3 4.6 2340 3.6 2.2 3.8 1.8 3.4 2.0

‘Right Eroe- 2340 pm he EsftEor-23409m vo En yo SRSSSRSREZEEEEE seenegneesees Fe . Freq . saxc.or TA rato eon tn SANE on in TaN Ott Case one {Ane onto ry Dy

Table 2: Attenuation of the overall A-weighted sound pressure level.

Turbocharger Fan Turbocharger+Fan rpm Left Right Left Right Left Right 763 1.1 0.3 0.6 0.0 1.3 0.9 1495 0.4 1.1 0.0 1.1 0.9 1.8 2340 3.2 2.8 1.4 1.0 2.3 2.1 5. CONCLUSIONS

In this article, an application of a multichannel feedforward ANC system installed on a motionless tractor is studied, analyzing the effect of implementing one or multiple reference sensors on the reduction of the spectral components associated with the main noise sources affecting the driver inside the tractor cabin. The results show that an ANC system with two reference sensors, such as an accelerometer placed on the turbocharger and a microphone located near the cooling fan, provides a more effective noise reduction, especially when the operativity condition of the engine is at minimum rpm. In particular, the best improvement reached in the main third-octave bands is of 4.7 dB for the minimum operativity condition and 1.0 dB for the case at 2340 rpm, with respect to the best case with one reference sensor. The overall performance in terms of global attenuation, either linear or A- weighted, is limited, partially because low-frequency components, falling outside the range where the control loudspeakers work efficiently according to their frequency response, dominate the overall levels. Based on the evidence of this study, in the follow-up of this project, the feedforward ANC system will be further developed by implementing only one reference sensor. In fact, the noise perceived by the driver is mostly dominated by the engine and a second reference sensor provides only limited benefits while increasing the computational and economical cost. 6. ACKNOWLEDGEMENTS

This work was supported by the BRIC 2019 ID-14 project from INAIL (National Institute for Insurance against Accidents at Work). The authors wish to thank Pietro Nataletti (INAIL) and Franco Cotana (CIRIAF) for their support in carrying out the research. 7. REFERENCES

1. Nataletti, P., Del Duca, L., Ruggeri, D., Annesi, D., Vassalini, G., Cerini L. & Sanjust, F. Sviluppo

e sperimentazione di un sistema di controllo attivo del rumore per applicazioni nelle cabine dei trattori. Proceedings of the 36 th National Congress of Associazione Italiana di Acustica (AIA) . Torino, Italy, June 2009. 2. Fanigliulo R., Del Duca L., Fornaciari L., Grilli R., Tomasone R. & Pochi D. Efficiency of an

ANC system in the tractor cabin under controlled engine workload. Noise Control Engineering Journal, 68(5) , 339–357 (2020). 3. Andrén, L., Johansson, S., Winberg M. & Claesson, I. Active noise control experiments in a fork-

lift truck cabin. Proceedings of the 11 th International Congress on Sound and Vibration (ICSV) . St. Petersburg, Russia, July 2004. 4. Gulyas, K., Pinte, G., Augusztinovicz, F., Desmet, W. & Sas, P. Active noise control in

agricultural machines. Proceedings of ISMA 2002. Leuven, Belgium, September 2002. 5. Mori, F., Bonfiglio, P., Fausti, P., Pompoli, F., Santoni, A. & Nataletti, P. Preliminary

experimental tests in hemi-anechoic chamber for active noise control inside the cabin of a tractor. Proceedings of the 27 th International Congress on Sound and Vibration (ICSV) . July 2021.