Deep learning-based single point sound source localization in spherical microphone arrays
Time: 8:40 am
Author: Soo Young Lee
Abstract ID: 2094
In this contribution, we present a high-resolution and accurate sound source localization via a deep learning framework. While the spherical microphone arrays can be utilized to produce omnidirectional beams, it is widely known that the conventional spherical harmonics beamforming (SHB) has a limit in terms of its spatial resolution. To accomplish the sound source localization with high resolution and preciseness, we propose a convolutional neural network (CNN)-based source localization model as a way of a data-driven approach. We first present a novel way to define the source distribution map that can spatially represent the single point sources position and strength. By utilizing paired dataset with spherical harmonics beamforming maps and our proposed high-resolution maps, we develop a fully convolutional neural network based on the encoder-decoder structure for establishing the image-to-image transformation model. Both quantitative and qualitative results are demonstrated to evaluate the powerfulness of the proposed data-driven source localization model.
Optimizing the acoustic properties of a meta-material using machine learning techniques
Time: 8:00 am
Author: Alessandro Casaburo
Abstract ID: 2294
The scope of this work is to consolidate research dealing with vibroacoustics of periodic media. This investigation aims at developing and validating tools for the design of global vibroacoustic treatments based on foam cores with embedded periodic patterns, which allow passive control of acoustic paths in layered concepts. Firstly, a numerical test campaign is carried out by considering some solid (but still non-perfectly rigid) inclusions in a 3D-modeled porous structure; this causes the excitation of additional acoustic modes due to the periodic nature of the meta-core itself. Then, some design guidelines are provided in order to predict several possible sets of characteristic parameters (i.e. inclusion geometry, elastic and foam properties) that, constrained by the imposition of mass and thickness of the acoustic package, may satisfy the target functions (i.e. the frequency at which the first Transmission Loss peak appears, together with its amplitude). Results are obtained through the implementation of machine learning algorithms, which may constitute a good basis in order to perform preliminary design considerations that could be interesting for further generalizations.
Explainable machine learning: A case study on impedance tube measurements
Time: 8:20 am
Author: Merten Stender
Abstract ID: 2342
Machine learning (ML) techniques allow for finding hidden patterns and signatures in data. Currently, these methods are gaining increased interest in engineering in general and in vibroacoustics in particular. Although ML methods are successfully applied, it is hardly understood how these black box-type methods make their decisions. Explainable machine learning aims at overcoming this issue by deepening the understanding of the decision-making process through perturbation-based model diagnosis. This paper introduces machine learning methods and reviews recent techniques for explainability and interpretability. These methods are exemplified on sound absorption coefficient spectra of one sound absorbing foam material measured in an impedance tube. Variances of the absorption coefficient measurements as a function of the specimen thickness and the operator are modeled by univariate and multivariate machine learning models. In order to identify the driving patterns, i.e. how and in which frequency regime the measurements are affected by the setup specifications, Shapley additive explanations are derived for the ML models. It is demonstrated how explaining machine learning models can be used to discover and express complicated relations in experimental data, thereby paving the way to novel knowledge discovery strategies in evidence-based modeling.
Partial discharge monitoring using deep neural networks with acoustic emission
Time: 6:40 am
Author: Saichand Gourishetti
Abstract ID: 2373
The occurrence of partial discharge (PD) indicates failures in electrical equipment. Depending on the equipment and operating conditions, each type of PD has its own acoustic characteristics and a wide frequency spectrum. To detect PD, electrical equipment is often monitored using various sensors, such as microphones, ultrasonic, and transient-earth voltage, whose signals are then analyzed manually by experts using signal processing techniques. This process requires significant expertise and time, both of which are costly. Advancements in machine learning, aim to address this issue by automatically learning a representation of the signal, minimizing the need for expert analysis. To this end, we propose a deep learning-based solution for the automatic detection of PD using airborne sound emission in the audible to the ultrasonic range. As input to our proposed model, we evaluate common time-frequency representations of the acoustic signal, such as short-time Fourier, continuous wavelet transform and Mel spectrograms. The extracted spectrum from the PD signal pulses is used to train and evaluate the proposed deep neural network models for the detection of different types of PD. Compared to the manual process, the automatic solution is seen as beneficial for maintenance processes and measurement technology.
Investigating the influence of microphone mismatch for acoustic traffic monitoring
Time: 11:20 am
Author: Saichand Gourishetti
Abstract ID: 2375
The development of robust acoustic traffic monitoring (ATM) algorithms based on machine learning faces several challenges. The biggest challenge is to collect and annotate large high-quality datasets for algorithm training and evaluation. Such a dataset must reflect a broad variety of vehicle sounds since their emitted acoustic noise patterns depend on a variety of factors such as engine noises at different speeds and road conditions. Additionally, the characteristics of the employed microphones have a strong influence on the data. If microphones with different directionality and frequency responses are used during the model development and the final deployment phase, a data mismatch is caused, which can have a deteriorating effect on the performance of machine learning algorithms. In this paper, the influence of mismatched recording locations and microphone characteristics on the proposed ATM system is investigated. To evaluate these effects, we implement state-of-the-art convolutional neural networks to detect passing vehicles, classify their type, and estimate their speed and direction of movement. The evaluated models perform well on low- and high-quality recordings at different locations when using the same recording device for training and testing. However, the results indicate that microphone mismatch causes several issues, which need to be carefully addressed.
Experimental force reconstruction on plates of arbitrary shape using neural networks
Time: 7:40 pm
Author: Tyler Dare
Abstract ID: 2397
Measuring the forces that excite a structure into vibration is an important tool in modeling the system and investigating ways to reduce the vibration. However, determining the forces that have been applied to a vibrating structure can be a challenging inverse problem, even when the structure is instrumented with a large number of sensors. Previously, an artificial neural network was developed to identify the location of an impulsive force on a rectangular plate. In this research, the techniques were extended to plates of arbitrary shape. The principal challenge of arbitrary shapes is that some combinations of network outputs (x- and y-coordinates) are invalid. For example, for a plate with a hole in the middle, the network should not output that the force was applied in the center of the hole. Different methods of accommodating arbitrary shapes were investigated, including output space quantization and selecting the closest valid region.
Generative Adversarial Neural Network for Semi-supervised Planetary Gearbox Fault Diagnosis
Time: 8:00 pm
Author: Gyuwon Kim
Abstract ID: 2479
Detecting bearing faults in advance is critical for mechanical and electrical systems to prevent economic loss and safety hazards. As part of the recent interest in artificial intelligence, deep learning (DL)-based principles have gained much attention in intelligent fault diagnostics and have mainly been developed in a supervised manner. While these works have shown promising results, several technical setbacks are inherent in a supervised learning setting. Data imbalance is a critical problem as faulty data is scarce in many cases, data labeling is tedious, and unseen cases of faults cannot be detected in a supervised framework. Herein, a generative adversarial network (GAN) is proposed to achieve unsupervised bearing fault diagnostics by utilizing only the normal data. The proposed method first adopts the short-time Fourier transform (STFT) to convert the 1-D vibration signals into 2-D time-frequency representations to use as the input to our (DL) framework. Subsequently, a GAN-based latent mapping is constructed using only the normal data, and faulty signals are detected using an anomaly metric comprised of a discriminator error and an image reconstruction error. The performance of our method is verified using a classic rotating machinery dataset (Case Western Reserve bearing dataset), and the experimental results demonstrate that our method can not only detect the faults but can also cluster the faults in the latent space with high accuracy.