Improving vehicle aeroacoustics using machine learning
Introduction
Noise inside vehicles has a negative impact on a passenger's general comfort. One of the most significant sources of noise inside a car at high speed (over 100 km/h) is the aerodynamics related wind noise that is generated by the air flow around the vehicle (George, 1990). In fact, at 140 km/h the wind noise can completely prevail over other sources of noise, such as the engine, tires, etc. Therefore, an improvement to the aerodynamic properties is important for increasing driving comfort, but this should not be at the expense of the vehicle's aesthetic and mechanical design.
Aeroacoustics engineers are often presented with real world design and economic constraints that allow only minor changes to be made to the vehicle. This means that the airflow is improved by a careful selection of the vehicle's minor external components, such as windshield wipers, door seals, antennas, etc. However, to determine the best set of components, extensive aeroacoustic testing is needed, and this is usually performed in a wind tunnel. Since running the wind tunnel is time consuming and expensive, there is a strong need in the automotive industry to automate and speed up the process of improving the aeroacoustic properties of a vehicle and thus reducing the cost.
In this paper we propose a way of automating the aeroacoustics improvement process by employing machine learning methods, with the aim to free aeroacoustics engineers from involvement in repetitive tasks. Moreover, software that supports the process can guide the improvement process and can also be very useful for training new engineers.
The rest of the paper is structured as follows. First, we look at the related work. We then follow this with an overview of the aeroacoustics improvement process. The following section then gives a detailed description of the tool that was developed to support the improvement process: the prediction of driver's subjective wind noise evaluation. The paper is concluded with a discussion.
Section snippets
Related work
First, we will discuss the related work regarding the prediction of a subjective wind noise evaluation. The problem of subjective sound (noise) evaluation is traditionally tackled with some form of jury evaluations (tests), where a number of people drive the vehicle or listen to the vehicle sound recordings and give their subjective evaluations (Otto et al., 2001, Otto, 1997, Society of Automotive Engineers, 2000). Since such evaluations are time consuming, some research has focused on
Overview of the improvement process
The process as presented in Fig. 1 was developed in collaboration with aeroacoustics engineers at the Fiat Group Automotive wind tunnel and researchers from the Fiat Research Center. They specified the steps and tasks that are usually carried out in their daily work and a functional requirements' analysis was performed. The result of this analysis is a process that uses computer software to steer and aid the improvement of the vehicle's aeroacoustics performance.
The aeroacoustics improvement
Subjective evaluation value prediction
A subjective evaluation value (SEV) for aerodynamic comfort is originally based on extensive road tests. It is performed by several drivers (non-engineers) testing a car on highways, who later fill in an extensive questionnaire to answer questions on different aspects of vehicle comfort. One section of this questionnaire deals with aeroacoustic comfort, and the result of this section is a single value in the range [1,10] that expresses the individually perceived comfort. The questionnaire
Discussion
In this paper we proposed a 6-step approach to reducing the wind noise in cars. The approach was implemented within the X-Media project3 and it was used as one of the main demonstrators of knowledge management in complex environments. We have shown that machine learning methods can successfully be applied to the domain of aeroacoustics. In addition to the sufficiently accurate predictions (confirmed by the domain expert) made by our tool, engineers can also gain
Acknowledgements
This work has been funded by the X-Media project (www.x-media-project.org) sponsored by the European Commission as part of the Information Society Technologies (IST) programme under EC grant No. IST-FP6-026978.
References (31)
Qualitative process theory
Artif. Intell.
(1984)- et al.
Index for vehicle acoustical comfort inside a passenger car
Appl. Acoust.
(2008) - et al.
Prediction of wind noise radiated from passenger cars and its evaluation based on auralization
J. Wind Eng. Ind. Aerodyn.
(1999) - et al.
Qualitatively faithful quantitative prediction
Artif. Intell.
(2004) - et al.
Computational aero-acoustic analysis of a passenger car with a rear spoiler
Appl. Math. Modell.
(2009) - et al.
Learning qualitative models from numerical data
Artif. Intell.
(2011) A comparison of some loudness measures for loudspeaker listening tests
J. Audio Eng. Soc.
(1992)- Ando, K., Takamura, A., Saito, I., 2010. Automotive Aerodynamic Design Exploration Employing New Optimization...
- et al.
Improving the effectiveness of paired comparison tests for automotive sound quality
- et al.
Sound quality assessment of internal automotive road noise using sensory science
Acta Acustica United Acustica
(2010)
When is “nearest neighbor” meaningful?
A reflective Newton method for minimizing a quadratic function subject to bounds on some of the variables
SIAM J. Optim.
Understanding how customers make their decisions on product sound quality
Cited by (9)
Improving aircraft performance using machine learning: A review
2023, Aerospace Science and TechnologyReview on China's Automotive Engineering Research Progress: 2023
2023, Zhongguo Gonglu Xuebao/China Journal of Highway and TransportUncertainty quantification of aeroelastic wings flutter using an optimized machine learning approach
2022, Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering