Autonomous design of noise-mitigating structures using deep reinforcement learning.

This paper explores the application of deep reinforcement learning for autonomously designing noise-mitigating structures. Specifically, deep Q- and double deep Q-networks are employed to find material distributions that result in broadband noise mitigation for reflection and transmission problems. Unlike conventional deep learning approaches which require prior knowledge for data labeling, the double deep Q-network algorithm learns configurations that result in broadband noise mitigations without prior knowledge by utilizing pixel-based inputs. By employing unified hyperparameters and network architectures for transmission and reflection problems, the capability of the algorithms to generalize over different environments is demonstrated. In addition, a comparison with a genetic algorithm highlights the potential for generalized design in complex environments, despite the algorithms tending to predict local maxima. Furthermore, we examine the impact of hyperparameters and environment types on agent performance. The autonomous design approach offers generalized learning while avoiding restrictions to specific shapes or prior knowledge of the task.

Relevance of phase information for object classification in automotive ultrasonic sensing using convolutional neural networks.

Automotive ultrasonic sensors come into play for close-range surround sensing in parking and maneuvering situations. In addition to ultrasonic ranging, classifying obstacles based on ultrasonic echoes to improve environmental perception for advanced driver-assistance systems is an ongoing research topic. Related studies consider only magnitude-based features for classification. However, the phase of an echo signal contains relevant information for target discrimination. This study discusses and evaluates the relevance of the target phase in echo signals for object classification in automotive ultrasonic sensing based on lab and field measurements. Several phase-aware features in the time domain and time-frequency features based on the continuous wavelet transform are proposed and processed using a convolutional neural network. Indeed, phase features are found to contain relevant information, producing only 4% less classification accuracy than magnitude features when the phase is appropriately processed. The investigation reveals high redundancy when magnitude and phase features are jointly fed into the neural network, especially when dealing with time-frequency features. However, incorporating the target phase information facilitates the identification quality in high clutter environments, increasing the model's robustness against signals with low signal-to-noise ratios. Ultimately, the presented work takes one further step toward enhanced object discrimination in advanced driver-assistance systems.

Exposure-response relation for vibration-induced white finger: inferences from a published meta-analysis of population groups.

PURPOSE: It is questioned whether the exposure-response relation for the onset of vibration-induced white finger (VWF) in ISO 5349-1:2001 needs to be revised based on the epidemiologic studies identified by Nilsson et al. (PLoS One https://doi.org/10.1371/journal.pone.0180795 , 2017), and whether the relation they derive improves the prediction of VWF in vibration-exposed populations. METHODS: A pooled analysis has been performed using epidemiologic studies that complied with selection rules and reported a VWF prevalence of 10% or more, and exposure constructed according to the provisions of ISO 5349-1:2001. The lifetime exposures at 10% prevalence were calculated for various data sets using linear interpolation. They were then compared to both the model from the standard and that developed by Nilsson et al. RESULTS: Regression analyses reveal excluding extrapolation to adjust group prevalences to 10% produce models with 95-percentile confidence intervals that include the ISO exposure-response relation but not that in Nilsson et al. (2017). Different curve fits are obtained for studies involving daily exposure to single or multiple power tools and machines. Studies with similar exposure magnitudes and lifetime exposure durations but markedly different prevalences are observed to cluster. CONCLUSIONS: A range of exposures and A(8)-values is predicted within which the onset of VWF is most likely to occur. The exposure-response relation in ISO 5349-1:2001, but not that proposed by Nilsson et al., falls within this range and provides a conservative estimate for the development of VWF. In addition, the analyses suggest that the method for evaluating vibration exposure contained in ISO 5349-1:2001 needs revision.

A multi-fidelity Gaussian process for efficient frequency sweeps in the acoustic design of a vehicle cabin.

Highly accurate predictions from large-scale numerical simulations are associated with increased computational resources and time expense. Consequently, the data generation process can only be performed for a small sample size, limiting a detailed investigation of the underlying system. The concept of multi-fidelity modeling allows the combination of data from different models of varying costs and complexities. This study introduces a multi-fidelity model for the acoustic design of a vehicle cabin. Therefore, two models with different fidelity levels are used to solve the Helmholtz equation at specified frequencies with the boundary element method. Gaussian processes (GPs) are trained on each fidelity level with the simulation results to predict the unknown system response. In this way, the multi-fidelity model enables an efficient approximation of the frequency sweep for acoustics in the frequency domain. Additionally, the proposed method inherently considers uncertainties due to the data generation process. To demonstrate the effectiveness of our framework, the multifrequency solution is validated with the high-fidelity (HF) solution at each frequency. The results show that the frequency sweep is efficiently approximated by using only a limited number of HF simulations. Thus, these findings indicate that multi-fidelity GPs can be adopted for fast and, simultaneously, accurate predictions.

Convolutional neural network with data augmentation for object classification in automotive ultrasonic sensing.

Today's low-cost automotive ultrasonic sensors perform distance measurements of obstacles within the close range of vehicles. For future parking assist systems and autonomous driving applications, the performance of the sensors should be further increased. This paper examines the processing of sensor data for the classification of different object classes and traversability of obstacles using a single ultrasonic sensor. The acquisition of raw time signals, transformation into time-frequency images, and classification using machine learning methods are described. Stationary and dynamic measurements at a velocity of 0.5 m/s of various objects have been carried out in a semi-anechoic chamber and on an asphalt parking space. We propose a scalogram-based signal processing chain and a convolutional neural network, which outperforms a LeNet-5-like baseline. Additionally, several methods for offline and online data augmentation are presented and evaluated. It is shown that carefully selected augmentation methods are useful to train more robust models. Accuracies of 90.1% are achieved for the classification of seven object classes in the laboratory and 66.4% in the outdoor environment. Traversability is correctly classified at an accuracy of 96.4% and 91.5%, respectively.

Non-negative aeroacoustic source contributions to radiated sound power.

A new approach that determines the contribution of aeroacoustic sources to sound power is presented. The method combines the Lighthill source distribution with an acoustic impedance matrix constructed from radiation kernels of the free-field Green's function. To demonstrate the technique, the flow noise produced by a pair of co-rotating vortices is examined. Results are initially compared with those obtained using Möhring's analogy of two-dimensional vortex sound radiation. The contribution to sound power for each component of the Lighthill tensor is presented for a range of wave numbers and vortex separation distances. For acoustically compact cases, the aeroacoustic source contributions for the diagonal components of the Lighthill tensor show a similar trend observed in sound maps for longitudinal quadruples. In contrast to the acoustically compact cases where the central focal area is mostly unchanged with variation in Mach number, significant variation in the focal areas occurs for non-acoustically compact cases. Using the aeroacoustic source contribution technique, the nature and location of dominant flow noise sources to sound power can be identified.

Sound attenuation enhancement of acoustic meta-atoms via coupling.

Arrangements of acoustic meta-atoms, better known as acoustic metamaterials, are commonly applied in acoustic cloaking, for the attenuation of acoustic fields or for acoustic focusing. A precise design of single meta-atoms is required for these purposes. Understanding the details of their interaction allows improvement of the collective performance of the meta-atoms as a system, for example, in sound attenuation. Destructive interference of their scattered fields, for example, can be mitigated by adjusting the coupling or tuning of individual meta-atoms. Comprehensive numerical studies of various configurations of a resonator pair show that the coupling can lead to degenerate modes at periodic distances between the resonators. We show how the resonators' separation and relative orientation influence the coupling and thereby tunes the sound attenuation. The simulation results are supported by experiments using a two-dimensional parallel-plate waveguide. It is shown that coupling parameters like distance, orientation, detuning, and radiation loss provide additional degrees of freedom for efficient acoustic meta-atom tuning to achieve unprecedented interactions with excellent sound attenuation properties.

Direct discrete complex image method for sound field evaluation above a non-locally reacting layer.

Thin layers of porous media are widely adopted in sound absorption and noise control applications due to their compact arrangement. Particularly, porous media with low flow resistivity exhibit complex, non-local reaction behavior. Therefore, sound field prediction above these media is computationally challenging. This is due to singularities and branch points in the expression for the reflection coefficient. This paper introduces a framework based on the direct discrete complex image method to analyze the sound field above a rigid-backed, non-locally reacting porous sample. In contrast to traditional complex image applications, the proposed framework avoids the extraction of the quasi-static term and the poles from the reflection coefficient. Instead, the reflection coefficient-including its singularities-is directly approximated in terms of a series of complex exponentials, whose coefficients are determined with the matrix pencil method. This study analyzes the sound field above two test samples made of melamine and rockwool. The sound field computation is efficient and accurate in the near- and in the far-field. Moreover, predicted specific impedances agree well with experimental in situ impedance measurements. The proposed framework serves for more applications including object detection in multilayered porous grounds or sound propagation prediction in layered atmosphere.

Theoretical analysis of a contactless transportation system for cylindrical objects based on ultrasonic levitation.

Contactless transportation systems based on near-field acoustic levitation have the benefit of compact design and easy control which are able to meet the cleanliness and precision demands required in precision manufacturing. However, the problems involved in contactless positioning and transporting cylindrical objects have not yet been addressed. This paper introduces a contactless transportation system for cylindrical objects based on grooved radiators. A groove on the concave surface of the radiator produces an asymmetrical pressure distribution which results in a thrusting force to drive the levitator horizontal movement. The pressure distribution between the levitator and the radiator is acquired by solving the Reynolds equation. The levitation and the thrusting forces are obtained by integrating the pressure and the pressure gradient over the concave surface, respectively. The predicted results of the levitation force agree well with experimental observations from the literature. Parameter studies show that the thrusting force increases and converges to a stable value as the groove depth increases. An optimal value for the groove arc length is found to maximize the thrusting force, and the thrusting force increases as the groove width, the radiator vibration amplitude, and the levitator weight increase.

Using rectified linear unit and swish based artificial neural networks to describe noise transfer in a full vehicle context.

Vehicle interior noise is a quality criterion of passenger cars. A considerable amount of resources is used to evaluate and design the acoustic environment with respect to given requirements. The customer's perception in the end-of-line vehicle is the main criterion. Therefore, full vehicle testing is a large part of today's sound comfort development. To increase efficiency, it is desirable to limit the hardware testing to a specific component. A later reassembly of the full vehicle is done virtually using transfer functions. These transfer functions of the substructures can be derived numerically or through measurements. However, full vehicle simulations are still challenging. Hence, transfer functions are typically measured but come with the burden of complex procedures. In this work, the authors propose a machine learning algorithm to reduce the effort for finding suitable transfer models in the automotive context. Artificial neural networks with rectified linear unit and swish activation functions are trained on full vehicle measurements. Multiple operation conditions are used for training. The networks compute spectral system responses and relative sensitivities for the input features. The performance is discussed with respect to the full vehicle validation data. The results indicate an effective procedure to reduce the costs of full-size vehicle measurements.

Generative adversarial networks for the design of acoustic metamaterials.

Metamaterials are attracting increasing interest in the field of acoustics due to their sound insulation effects. By periodically arranged structures, acoustic metamaterials can influence the way sound propagates in acoustic media. To date, the design of acoustic metamaterials relies primarily on the expertise of specialists since most effects are based on localized solutions and interference. This paper outlines a deep learning-based approach to extend current knowledge of metamaterial design in acoustics. We develop a design method by using conditional generative adversarial networks. The generative network proposes a cell candidate regarding a desired transmission behavior of the metamaterial. To validate our method, numerical simulations with the finite element method are performed. Our study reveals considerable insight into design strategies for sound insulation tasks. By providing design directives for acoustic metamaterials, cell candidates can be inspected and tailored to achieve desirable transmission characteristics.

Acoustics Apps: Interactive simulations for digital teaching and learning of acoustics.

This paper presents the Acoustics Apps, an e-learning platform that offers an interactive and playful environment for teaching and learning the principles of acoustics and vibration. The Acoustics Apps address the increasing demand for digitized teaching methods, which might be suitable for home schooling or as a complement to physical experiments by adding interactive simulation. The apps combine learning by experimenting, observing, and exploring using state-of-the-art scientific methods and numerical simulations. The ability to visualize and control acoustic phenomena facilitates understanding of the relevant physical principles. The apps are designed to be used intuitively and can be tailored to suit the existing knowledge of the user. As such, a wide range of users can benefit from this learning aid. It has been developed to allow barrier-free access to modern educational tools, requiring only a device with a browser and Internet access. The necessary computing power is provided by an external server using the COMSOL ServerTM technology. The Acoustics Apps are freely available for academic and teaching purposes at apps.vib.mw.tum.de.

Spatial reconstruction of the sound field in a room in the modal frequency range using Bayesian inference.

Spatial characterization of the sound field in a room is a challenging task, as it usually requires a large number of measurement points. This paper presents a probabilistic approach for sound field reconstruction in the modal frequency range for small and medium-sized rooms based on Bayesian inference. A plane wave expansion model is used to decompose the sound field in the examined domain. The posterior distribution for the amplitude of each plane wave is inferred based on a uniform prior distribution with limits based on the maximum sound pressure observed in the measurements. Two different application cases are studied, namely a numerically computed sound field in a non-rectangular two-dimensional (2D) domain and a measured sound field in a horizontal evaluation area of a lightly damped room. The proposed reconstruction method provides an accurate reconstruction for both examined cases. Further, the results of Bayesian inference are compared to the reconstruction with a deterministic compressive sensing framework. The most significant advantage of the Bayesian method over deterministic reconstruction approaches is that it provides a probability distribution of the sound pressure at every reconstruction point, and thus, allows quantifying the uncertainty of the recovered sound field.

Reduction of vibration-induced signal loss by matching mechanical vibrational states: Application in high b-value diffusion-weighted MRS.

PURPOSE: Diffusion encoding gradients are known to yield vibrations of the typical clinical MR scanner hardware with a frequency of 20 to 30 Hz, which may lead to signal loss in diffusion-weighted MR measurements. This work proposes to mitigate vibration-induced signal loss by introducing a vibration-matching gradient (VMG) to match vibrational states during the 2 diffusion gradient pulses. THEORY AND METHODS: A theoretical description of displacements induced by gradient switching was introduced and modeled by a 2-mass-spring-damper system. An additional preceding VMG mimicking timing and properties of the diffusion encoding gradients was added to a high b-value diffusion-weighted MR spectroscopy sequence. Laser interferometry was employed to measure 3D displacements of a phantom surface. Lipid ADC was assessed in water-fat phantoms and in vivo in the tibial bone marrow of 3 volunteers. RESULTS: The modeling and the laser interferometer measurements revealed that the displacement curves are more similar during the 2 diffusion gradients with the VMG compared to the standard sequence, resulting in less signal loss of the diffusion-weighted signal. Phantom results showed lipid ADC overestimation up to 119% with the standard sequence and an error of 5.5% with the VMG. An 18% to 35% lower coefficient of variation was obtained for in vivo lipid ADC measurement when employing the VMG. CONCLUSION: The application of the VMG reduces the signal loss introduced by hardware vibrations in a high b-value diffusion-weighted MRS sequence in phantoms and in vivo. Reference measurements based on laser interferometry and mechanical modelling confirmed the findings.

Investigation of radiation damping in sandwich structures using finite and boundary element methods and a nonlinear eigensolver.

The fully coupled vibroacoustic interaction of sandwich panels is studied using the finite and the boundary element methods. The extent of radiation damping is quantified for various configurations based on both harmonic response analyses and modal analyses. The underlying nonlinear eigenvalue problem is solved using a projection method based on contour integration yielding the shifted (wet) eigenfrequencies, modal radiation loss factors, and air-loaded structural modes. The numerical results clearly illustrate the relevance of air-loading when studying the vibration of sandwich structures. Further, the numerically obtained estimates for radiation damping are compared to both theoretical expressions and experimental results found in the literature. Although good agreement is observed in general, the comparison indicates the limited applicability of commonly used theoretical expressions when coincidence occurs in a frequency range where the modes are still well separated. Moreover, possible sources of error when experimentally determining radiation damping are discussed in detail. The results presented in this paper provide deep insights into the phenomenon of acoustic radiation damping and help to estimate its relevance in future research.

Stabilizing near-field acoustic levitation: Investigation of non-linear restoring force generated by asymmetric gas squeeze film.

The instability of a floating object is the main factor preventing near-field acoustic levitation (NAFL) from being widely used in the manufacture of micro-electro-mechanical systems. Therefore, investigating the restoring force due to the generation mechanisms of NAFL is necessary to ensure the stable levitation of the floating object. This study presents a theoretical analysis to evaluate the restoring force based on the gas-film-lubrication theory. The gas-film pressure between the reflector and the radiator is expressed in the form of the dimensionless Reynolds equation in a cylindrical coordinate system, which is solved by an eight-point discrete grid method due to the discontinuous gas-film distribution. An experimental rig is constructed to measure the restoring force at various eccentricities, which can be used to support the developed numerical model. The theoretical results show that the restoring force increases with an increment in eccentricity, which agrees with experimental results. Furthermore, theoretical prediction results indicate that the restoring force increases when the amplitude of the radiator and weight of the levitator increases, which indicates higher system stability. The results of the radiator vibration mode on the restoring force show that the restoring force is the largest in the first-order mode.

Acoustic metamaterial capsule for reduction of stage machinery noise.

Noise mitigation of stage machinery can be quite demanding and requires innovative solutions. In this work, an acoustic metamaterial capsule is proposed to reduce the noise emission of several stage machinery drive trains, while still allowing the ventilation required for cooling. The metamaterial capsule consists of c-shape meta-atoms, which have a simple structure that facilitates manufacturing. Two different metamaterial capsules are designed, simulated, manufactured, and experimentally validated that utilize an ultra-sparse and air-permeable reflective meta-grating. Both designs demonstrate transmission loss peaks that effectively suppress gear mesh noise or other narrow band noise sources. The ventilation by natural convection was numerically verified, and was shown to give adequate cooling, whereas a conventional sound capsule would lead to overheating. The noise spectra of three common stage machinery drive trains are numerically modelled, enabling one to design meta-gratings and determine their noise suppression performance. The results fulfill the stringent stage machinery noise limits, highlighting the benefit of using metamaterial capsules of simple c-shape structure.

Non-negative intensity and back-calculated non-negative intensity for analysis of directional structure-borne sound.

Non-negative intensity (NNI) is an approach to identify the surface areas of a structure that contribute to sound power. NNI is evaluated in terms of the acoustic impedance matrix obtained directly at the structural surface and as such can only identify surface contributions to sound power at a far-field receiver surface that fully circumscribes the structure. In contrast, back-calculated NNI is evaluated in terms of the acoustic impedance matrix obtained at a far-field receiver surface, and hence can identify surface contributions to sound power at a far-field receiver surface that does not fully circumscribe the structure. In this work, NNI and acoustic intensity obtained numerically using the boundary element method and experimentally from near-field acoustic holography measurements are compared for different modes. Back-calculated NNI evaluated for full and partial receiver surfaces is also compared with acoustic intensity for the different modes. Results for back-calculated NNI show that different regions on the plate surface contribute sound to different receiver locations.

Non-negative intensity for coupled fluid-structure interaction problems using the fast multipole method.

The non-negative intensity (NNI) method is applied to large-scale coupled fluid-structure interaction (FSI) problems using the fast multipole boundary element method (FMBEM). The NNI provides a field on the radiating structure surface that consists of positive-only contributions to the radiated sound power, thus avoiding the near-field cancellation effects that otherwise occur with the sound intensity field. Thus far the NNI has been implemented with the boundary element method (BEM) for relatively small problem sizes to allow for the full BEM coefficient and inverse matrices to be explicitly constructed and stored. In this work, the FMBEM is adapted to the NNI by calculating the eigenvalue solution of the symmetric acoustic impedance matrix using the FMBEM via a two-stage solution method. The FMBEM implementation of the NNI is demonstrated for a large-scale model of a submerged cylindrical shell. The coupled FSI problem is first solved using a finite element-FMBEM model and the resulting surface fields are then used in the FMBEM calculation of the NNI. An equivalent reactive NNI field representing the evanescent near-field radiation is demonstrated and the effect of the chosen number eigenvectors on the NNI field is investigated.

Supersonic intensity and non-negative intensity for prediction of radiated sound.

Two numerical methods to identify the surface areas of a vibrating structure that radiate sound are presented. The supersonic intensity identifies only the supersonic wave components of the sound field contributing to far-field radiated sound. The supersonic intensity is calculated using a two-dimensional convolution between a spatial radiation filter and the sound field. To compute the spatial radiation filter, the shortest surface distance between two points on the structure is calculated using the geodesic distance method. The non-negative intensity is based on acoustic radiation modes and identifies the radiated sound power from a vibrating structure. Numerical models of a baffled plate, a cylinder and an engine crankcase are presented. The supersonic intensity is shown to be difficult to implement at low frequencies due to the size of the spatial radiation filter and accuracy of the surface distances. A cut-off coefficient associated with the acoustic wavenumber of the spatial radiation filter is used to reduce the aperture error. A comparison of the two intensity-based techniques both in terms of a sound power ratio and the modal assurance criterion is introduced to identify the optimal values of the cut-off coefficients that result in better convergence between the intensity techniques.