Applied Acoustics最新文献

It is a huge challenge for underwater acoustic receivers to correctly identify modulation methods due to the complex underwater channel environment and severe noise interference. Combined with the lightweight network (SqueezeNet) and attention mechanism (SENet), a multi-attribute and multi-scale feature fusion model based on a hybrid neural network is proposed, which achieves efficient and accurate recognition for modulation modes. First, the wavelet time-frequency (WTF) spectrum, square power spectrum, and contour maps of cyclic spectrum are extracted as multi-attribute inputs for the network to reduce the impact of inherent defects in single attribute feature. Second, shallow and deep features based on the SqueezeNet model are obtained as multi-scale features, of which the key feature expression ability is enhanced by the SENet model to provide sufficient feature information for modulation recognition. The simulation experiments and sea trial data confirm that the suggested method demonstrates strong generalization capabilities and effectiveness when applied to underwater acoustic channels and environmental noise. In contrast to algorithms in existence, the method verifies superior recognition abilities.

In order to reduce the low frequency noise of water filled pipelines, a micro-perforated muffler with composite structure flexible back cavity is proposed. The sound pressure and the displacement are expanded into three or two-dimensional Chebyshev series forms respectively. A numerical model, based on Hamilton’s principle of minimum potential energy and Rayley-Ritz method, is proposed for the accurate prediction of sound pressure and transmission loss. The results show that the noise attenuation bandwidth can be widened by selecting the proper length of internal intubation. In a certain range, the peak frequency of transmission loss moves to low frequency by increasing the perforation diameter and the thickness of the MPP or reducing the perforation rate. By reducing the laying angle or the number of layers of the flexible wall, the lower peak frequency and lowest muffling frequency of transmission loss can be obtained. The transmission loss has a sudden change at the axial mode frequency, which results in peaks and dips. Finally, the experimental research is carried out to verify the theory of this paper.

The informative frequency band (IFB) plays a vital role in detecting defects in complex machinery through visible informative features. In the present work, a denoising filter has been designed to enhance the small non-stationarities present in the signal. Initially, the system impulse is computed to estimate the filter coefficients which are further optimized by the mountain gazelle optimization (MGO) based on the maximum value fitness function. The novel sparsity index based on kurtosis and negentropy (NE) is put forward as the fitness function. Then, optimized coefficients are convolved with the system impulse to design the denoising filter. The efficacy of the designed filter is verified through vibration and acoustic signals from the defective components of the belt conveyor system. The designed filter is better able to extract the impulsiveness from the signal, give improved values of kurtosis and signal-to-noise ratio (SNR), and reduce interferences from other machinery components and the environment simultaneously.

Fault diagnosis is crucial for mechanical systems, with early diagnosis of bearings playing a key role in ensuring the overall safety and smooth operation of the mechanical system. However, in real industrial environments, traditional diagnostic methods limit the extraction of fault signals from rotating machinery. This study aims to improve the fault diagnosis method for critical mechanical components and proposes a novel deep learning model, the Attention Improved CNN (AT-ICNN) fault diagnosis method. The method combines Convolutional Neural Network (CNN) and attention mechanism to extract key fault feature information from signals, enhancing the model’s ability to highlight fault features and capture global information. This improves the accuracy of fault type identification. The AT-ICNN model enhances traditional CNN models by introducing Improved Convolutional (IMConv) and integrating a hybrid attention mechanism to effectively extract relevant fault information. Experimental results demonstrate superior diagnostic performance of AT-ICNN on the CWRU bearing dataset and laboratory bearing dataset, with accuracy rates of 98.12% and 98.72%, respectively. This represents about 9% improvement over baseline models and other advanced methods. In-depth analysis of experimental results validates the significant advantages of AT-ICNN in the field of fault diagnosis for critical mechanical components.

This study investigates the perception of impulsivity in audio signals specifically caused by thermal expansion (cracking noise) in domestic refrigerators. It employs a multifaceted approach encompassing an analysis of sensitivity to impulsive events, an assessment of the prominence of impulse signals, and a correlation analysis of these data. The investigation revealed different responses to various sound samples, highlighting subjective variations. Using linear regression, correlation analysis demonstrated a robust and positive relationship (Pearson correlation coefficient of 0.851) between perceived impulsivity and the Impulsivity Prediction Model (IPM). This alignment underscores the reliability of the developed IPM in capturing and predicting subjective perceptions of low-amplitude transient signals. Comparisons between groups of participants, conducted using both Analysis of Variance (ANOVA) and t-tests, explored potential disparities related to gender, age, and acoustic knowledge. The results indicated no statistically significant differences in the perception of impulsivity concerning gender, age groups, or acoustic knowledge. In conclusion, this study provides insights into the perceptual aspects of impulsivity in audio signals from home refrigerators, specifically addressing thermal expansion noises, and establishes the reliability of the Impulsivity Prediction Model (IPM) as a tool for objective assessment. The congruence between subjective judgments and objective metrics enhances the applicability of IPM in diverse fields, from acoustic engineering to psychoacoustic research.

Compressive spherical beamforming (CSB) with spherical microphone arrays not only inherits the high spatial resolution and strong sidelobe suppression of compressive beamforming but also achieves panoramic acoustic source identification owing to the rotational symmetry of spherical microphone arrays, which is an interesting topic in the field of acoustic source identification. The recent off-grid sparse Bayesian inference-based CSB (OGSBI-CSB) can effectively overcome the basis mismatch of earlier on-grid CSB approaches and shows a higher resolution than the Newtonized orthogonal matching pursuit-based CSB (NOMP-CSB), however, it is severely time-consuming. Therefore, this paper proposes fast OGSBI-CSB (FOGSBI-CSB), which first solves an on-grid CSB model using sparse Bayesian inference to estimate the initial directions of arrival (DOAs), then performs DOA refinement by discretizing the local regions centered on the initial on-grid DOAs into finer grids and searching for candidates that can maximize the cost function, and finally quantifies source strengths utilizing the least squares method. Simulation and experimental results demonstrate that the proposed FOGSBI-CSB could provide a higher resolution than NOMP-CSB and a higher computational efficiency and resolution than OGSBI-CSB.

Accurately identifying the pipeline leak size is crucial for risk assessment and timely rescue. In this study, a Convolutional Neural Network (CNN) based on Continuous Wavelet Transform (CWT) acoustic image transformation is proposed to identify small-sized leak in non-metallic pipes. Firstly, one-dimensional acoustic signals are filtered using the Piecewise Aggregate Approximation (PAA) algorithm to reduce noise and storage resource consumption. Then, the filtered signals are transformed into two-dimensional images by CWT to enrich signal feature information, serving as the input for the CNN. Further, a leak size recognition model based on CWT-CNN is established. The effectiveness of this model is verified using experimental data from a non-metallic pipeline leak test. A comparative analysis is conducted on diverse acoustic image transformation methods, including CWT, Gramian Angular Summation Field (GASF), and Relative Position Matrix (RPM). The results demonstrate the superiority of the CWT-CNN model in pipeline leak size recognition. Finally, the impact of the signal length in an acoustic image on recognition accuracy is also examined. The results demonstrate that when the signal length in an acoustic image is 0.75 s, the accuracy obtained by CWT-CNN can reach 95 %.

Herein, a broadband acoustic metamaterial composed of parallel Helmholtz resonators (PHR) with embedded channels and porous material (PM), is designed for low-to-mid-frequency noise absorption. A theoretical model of acoustic impedance is developed to illustrate the absorption characteristics of PHR–PM. The validity of the present model is confirmed by comparing the experimental results and numerical simulations. The PM may enhance the sound absorption performance of the PHR–PM by satisfying impedance matching conditions, which provides a new strategy for designing resonant systems with tunable sound-absorption characteristics. Both PM and PHR contribute to sound absorption, although their absorption capacities depend on the frequency ranges. The effects of structural and material parameters on sound absorption capacity are also analytically explored. Results indicate that sound absorption in the co-action and PM-dominated regions is mainly affected by material parameters, while that across the entire frequency range is considerably affected by structural parameters. Moreover, the average absorption coefficient of the 13HRs–PM may reach up to 0.6 at the frequency range of 100–1600 Hz, demonstrating its potential in achieving good broadband sound absorption performance and excellent absorption tenability. The proposed novel composite structure offers a new strategy for realizing high sound absorption at low-to-mid frequencies.

In this study, we propose Signal Latent Subspace (SLS), a flexible method that classifies environmental sound events using the subspace representations of latent features obtained from various neural network-based models. Our main goal is to leverage the high expressiveness of neural networks while retaining the advantages of subspace representation, such as its robustness to noise and ability to work under small sample size (SSS) conditions. We also propose an ensemble strategy native to the subspace representation, to achieve increased performance and reduce the generalization error. We do this through product Grassmann manifold (PGM), resulting in SLS-PGM. Each subspace constructed from latent features of a network can be seen as a point on a factor Grassmann manifold (GM) of a neural network; through PGM, it is possible to unify factor manifolds into a singular representation, and perform classification through a similarity metric on the manifold. We further improve SLS and SLS-PGM in two ways: (1) by using generalized difference subspace (GDS) projection to address the lack of between-class discrimination of subspace representation and (2) by leveraging finetuning regimes to better adapt neural network models to the ESC task. We evaluate our proposed methods, factoring various neural networks, on ESC-10, ESC-50 and UrbanSound environmental sound datasets, and provide extensive ablation experiments and notes for practical use.