Search

In this paper, we propose a learning-based approach to identify the type and position of sounds using a single microphone in a real-world building. We attempt to treat this problem as a joint classification problem in which we predict the exact positions of sounds while classifying the types that are assumed to be from pre-defined types of sounds. The most problematic issue is that while the types are readily classified under supervised learning frameworks with one-hot encoded labels, it is difficult to predict the exact positions of the sound from unseen positions during training. To address this potential discrepancy, we formulate the position identification problem as a zero-shot learning problem inspired by the human ability to perceive new concepts from previously learned concepts. We extract feature representations from audio data and vectorize the type and position of the sound source as ‘type/position-aware attributes’, instead of labeling each class with a simple one-hot vector. We then train a promising generative model to bridge the extracted features and the attributes by learning the class-invariant structure to transfer the knowledge from seen to unseen classes through their attributes; generative adversarial networks are conditioned on the class-embeddings. Our proposed methods are evaluated on an indoor noise dataset, SNU-B36-EX, a real-world dataset collected inside a building.

Ray-based blind deconvolution (RBD) is a method that estimates the source waveform and channel impulse response (CIR) using the ray arrival in an underwater environment. The RBD estimates the phase of the source waveform by using beamforming. However, low sampling, array shape deformation, and other factors can cause phase errors in the beamforming results. In this paper, phase correction is applied to the beamforming estimated source phase to improve RBD performance. The impulsiveness of the CIR was used as additional information to correct the initially estimated source phase. Kurtosis was used to measure impulsiveness, and the phase correction that maximized the kurtosis of the CIRs was calculated through optimization. The proposed approach is called ray-based blind deconvolution with maximum kurtosis phase correction (RBD-MKPC) and is based on a single-input multiple-output system. The RBD-MKPC was tested with several CIRs and source waveform combinations in the shallow-water acoustic variability experiment 2015 using broadband high-frequency pulses (11–31 kHz) as the source and a sparse vertical 16-element line array as receivers. The results indicate that the RBD-MKPC improves the estimation performance. In addition, from an optimization point of view and compared with other initialization methods, the proposed method showed superior convergence speed and estimation performance.

In many industrial systems involving pipe or duct structures, sound transmission modeling is critical to noise treatment and control. A typical noise treatment method applied to pipe structures is to use a silencer. In particular, a dissipative silencer with a sound absorbing material has a noise control ability to reduce noise through successive sound reflections by impedance mismatch while dissipating incident sound energy as heat. In this study, the acoustic performance of a dissipative silencer lined with multi-layered absorbing materials for fluid-filled pipe system is investigated. Herein, the absorbing materials are modeled applying poroelastic and viscoelastic theories. In the analysis, the eigenvalues and eigenfunctions in the dissipative silencer are determined from the rigid wall boundary condition and the interfaces between the absorbing materials, and then the acoustic performance is compared by obtaining the transmission loss using the mode matching method. Also, a parametric study is conducted to investigate the effect of the number and arrangement of layers on the acoustic performance of the silencer according to the absorbing material properties.

Ray-based blind deconvolution (RBD) has demonstrated remarkable performance on several in situ underwater acoustic data. RBD estimates the channel impulse response by compensating the source phase estimated by beamforming. However, when beamforming fails to achieve the desired signal-to-noise ratio, RBD fails to achieve the expected performance. To address this issue, we propose an improved RBD framework that uses the kurtosis of the channel impulse response to provide additional phase correction after beamforming. The proposed method is validated with the measurement data from the shallow-water acoustic variability experiment 2015 using a high-frequency (11–31 kHz) broadband source and a vertical line array of 16 sensors located approximately 3 km away. The results are compared with conventional RBD and discussed from an optimization point of view.

Heat straightening of welding-induced distortion, which has mostly relied on shipyard experience, was systemized for computation. The process of geometrically flattening a curved surface with fixed edges was formulated step by step based on thermal forming mechanisms. The proposed algorithm divides the measured surface into patches, connects them with virtual springs, and optimally arranges the patches on a plane in a direction that minimizes the potential energy within the weld lines. The overlaps between the geometrically calculated patches become the amount to shrink. The relationship between spot heating distance, curvature, and shrinkage was investigated through experiments and thermal elastic-plastic analysis to convert shrinkages into process parameters. It showed that the plastic shrinkage increases as the spot heating distance narrows, but the rate decreases. Also, the plastic shrinkage keeps constant regardless of the curvature, whereas the total shrinkage decreases as the curvature decreases. For validation of the proposed methodology, flame straightening experiments were carried out. It has been shown that the case using the plastic shrinkage produced fewer heating spots with higher efficiency than that using the total shrinkage.