Noise Robustness Research Articles

Applying matrix regularization to identification of traffic-induced equivalent loads and multi-vehicle weight on beam-like bridge

Identification of traffic-induced forces is an important topic in the field of bridge structural health monitoring (B-SHM). It can provide reliable data to support the great demand for monitoring and evaluating the working conditions of bridges. Bridges may serve under consistent moving vehicles and this requires that the methods of moving force identification (MFI) should be applicable under circumstances such as multiple-vehicle loading and long-term monitoring. Centering on cases containing multiple vehicles, this study proposed a matrix regularization-based method for the identification of traffic-induced equivalent loads and multi-vehicle weights. Herein, multiple moving forces in the spatial domain are converted into a certain number of equivalent loads which virtually act at fixed locations on the bridge as a kind of coordinate transformation. The MFI problem then can be transformed into tasks of identifying some concentrated forces. To solve the problem of equivalent load identification, the structural input-output relationship is firstly reorganized in a matrix-matrix-matrix form with the help of segmentation via time windows and consideration of unknown structural initial conditions. Then the matrix regularization with a sparse penalty term is introduced to formulate an identified equation so that, it can ensure the strong noise robustness of the identified results. Finally, the total weight of multiple vehicles can be empirically estimated from the sum of the identified equivalent loads. Numerical simulations and model experiments are performed to verify the proposed method's performance. The effects of vehicle speeds, measuring noise, number of modal orders, number of vehicles, and vehicle spacings are investigated. The illustrated results show that the proposed method turns out to be effective and precise in identifying the traffic-induced equivalent loads and the weight of multiple vehicles.

SHM data compression and reconstruction based on IGWO-OMP algorithm

The long-term operating structural health monitoring (SHM) system generates massive monitoring data, whose transmission and storage are challenging tasks. To address this issue, this study proposed an improved grey wolf optimizer-orthogonal matching pursuit (IGWO-OMP) algorithm for SHM data compression and reconstruction. The effectiveness of the proposed algorithm was validated using four SHM datasets including ultrasonic guided wave (UGW), acceleration, and audio signals. Firstly, the optimization results of IGWO were compared with those of other classical algorithms to verify its superiority in reconstruction accuracy. Secondly, the reconstruction accuracy under various compression factors was investigated and the optimal compression factor was determined. Finally, the relationship between reconstruction accuracy and signal sparsity as well as the performance of IGWO-OMP in dealing with noise-containing signals were discussed. The results demonstrate the generalizability and excellent noise robustness of IGWO-OMP in SHM data compression and reconstruction. Compared with other optimization algorithms, IGWO exhibits the best global optimal solution searching ability and obtains the reconstructed signal closest to the original one. Comprehensively considering the compressed signal length, reconstruction accuracy and computational efficiency, the optimal compression factor is suggested to be 0.2. In IGWO-OMP, the signal reconstruction accuracy is related to its sparsity, the variation of the CCD index for the reconstructed signal with relative sparsity follows an exponential growth relationship.

Quantum computing quantum Monte Carlo with hybrid tensor network for electronic structure calculations

Quantum computers have a potential for solving quantum chemistry problems with higher accuracy than classical computers. Quantum computing quantum Monte Carlo (QC-QMC) is a QMC with a trial state prepared in quantum circuit, which is employed to obtain the ground state with higher accuracy than QMC alone. We propose an algorithm combining QC-QMC with a hybrid tensor network to extend the applicability of QC-QMC beyond a single quantum device size. In a two-layer quantum-quantum tree tensor, our algorithm for the larger trial wave function can be executed than preparable wave function in a device. Our algorithm is evaluated on the Heisenberg chain model, graphite-based Hubbard model, hydrogen plane model, and MonoArylBiImidazole using full configuration interaction QMC. Our algorithm can achieve energy accuracy (specifically, variance) several orders of magnitude higher than QMC, and the hybrid tensor version of QMC gives the same energy accuracy as QC-QMC when the system is appropriately decomposed. Moreover, we develop a pseudo-Hadamard test technique that enables efficient overlap calculations between a trial wave function and an orthonormal basis state. In a real device experiment by using the technique, we obtained almost the same accuracy as the statevector simulator, indicating the noise robustness of our algorithm. These results suggests that the present approach will pave the way to electronic structure calculation for large systems with high accuracy on current quantum devices.

Multi-scale quaternion CNN and BiGRU with cross self-attention feature fusion for fault diagnosis of bearing

In recent years, deep learning has led to significant advances in bearing fault diagnosis (FD). Most techniques aim to achieve greater accuracy. However, they are sensitive to noise and lack robustness, resulting in insufficient domain adaptation and anti-noise ability. The comparison of studies reveals that giving equal attention to all features does not differentiate their significance. In this work, we propose a novel FD model by integrating multi-scale quaternion convolutional neural network (MQCNN), bidirectional gated recurrent unit (BiGRU), and cross self-attention feature fusion (CSAFF). We have developed innovative designs in two modules, namely MQCNN and CSAFF. Firstly, MQCNN applies quaternion convolution to multi-scale architecture for the first time, aiming to extract the rich hidden features of the original signal from multiple scales. Then, the extracted multi-scale information is input into CSAFF for feature fusion, where CSAFF innovatively incorporates cross self-attention mechanism to enhance discriminative interaction representation within features. Finally, BiGRU captures temporal dependencies while a softmax layer is employed for fault classification, achieving accurate FD. To assess the efficacy of our approach, we experiment on three public datasets (CWRU, MFPT, and Ottawa) and compare it with other excellent methods. The results confirm its state-of-the-art, which the average accuracies can achieve up to 99.99%, 100%, and 99.21% on CWRU, MFPT, and Ottawa datasets. Moreover, we perform practical tests and ablation experiments to validate the efficacy and robustness of the proposed approach. Code is available at https://github.com/mubai011/MQCCAF.

EDUNet++: An Enhanced Denoising Unet++ for Ice-Covered Transmission Line Images

New technology has made it possible to monitor and analyze the condition of ice-covered transmission lines based on images. However, the collected images are frequently accompanied by noise, which results in inaccurate monitoring. Therefore, this paper proposes an enhanced denoising Unet++ for ice-covered transmission line images (EDUNet++). This algorithm mainly comprises three modules: a feature encoding and decoding module (FEADM), a shared source feature fusion module (SSFFM), and an error correction module (ECM). In the FEADM, a residual attention module (RAM) and a multilevel feature attention module (MFAM) are proposed. The RAM incorporates the cascaded residual structure and hybrid attention mechanism, that effectively preserve the mapping of feature information. The MFAM uses dilated convolution to obtain features at different levels, and then uses feature attention for weighting. This module effectively combines local and global features, which can better capture the details and texture information in the image. In the SSFFM, the source features are fused to preserve low-frequency information like texture and edges in the image, hence enhancing the realism and clarity of the image. The ECM utilizes the discrepancy between the generated image and the original image to effectively capture all the potential information in the image, hence enhancing the realism of the generated image. We employ a novel piecewise joint loss. On the dataset of ice-covered transmission lines, PSNR (peak signal to noise ratio) and SSIM (structural similarity) achieved values of 29.765 dB and 0.968, respectively. Additionally, the visual effects exhibited more distinct detailed features. The proposed method exhibits superior noise suppression capabilities and robustness compared to alternative approaches.

A comparative study on hydrocarbon detection using cepstrum–based methods

Amplitude anomaly caused by a high–frequency component is attenuated more rapidly than a low–frequency component for a seismic trace in sediments susceptible to gas influence and is always used as a direct indicator of hydrocarbons. Recently, cepstrum–based hydrocarbon detection methods have become an important exploration pathway for detecting amplitude anomalies in the gas–bearing formation more sensitively and accurately. In this study, we compared three cepstrum–based hydrocarbon detection methods. We compared their fluid identification abilities in gas reservoirs and the noise robustness. They were further compared with the traditional absorption coefficient estimation method. We also indicate how the choice of the selection of the sliding–window length influences the performance of the three cepstrum–based hydrocarbon detection methods. Model test results and real data applications show that for thick gas reservoirs, the hydrocarbon detection methods using the Berthil–based cepstrum and the wavelet–based cepstrum can better discriminate the upper and lower interfaces than the hydrocarbon detection method using the Fourier–based cepstrum. For thin gas reservoirs, the hydrocarbon detection method using the wavelet–based cepstrum can identify the lower interface of the gas–bearing reservoir more obviously than the upper interface. The hydrocarbon detection method using the Berthil cepstrum cannot identify the upper and lower interfaces of the gas–bearing reservoir, but it shows higher temporal and spatial resolution compared to the hydrocarbon detection method using the Fourier–based cepstrum. For noise robustness, the hydrocarbon detection method using the wavelet–based cepstrum has the strongest noise robustness property, and the hydrocarbon detection method using the Fourier–based cepstrum is the most sensitive method to the noise. For time–consumption, the hydrocarbon detection method using the wavelet–based cepstrum is the most time–consuming, and the hydrocarbon detection method using the Fourier–based cepstrum is the fastest.

VAD system under uncontrolled environment: A solution for strengthening the noise robustness using MMSE-SPZC

VAD system under uncontrolled environment: A solution for strengthening the noise robustness using MMSE-SPZC

WBUN: an interpretable convolutional neural network with wavelet basis unit embedded for fault diagnosis

Convolutional Neural Network (CNN) is extensively applied in mechanical system fault diagnosis. However, the absence of transparent decision mechanisms in CNNs hinders credibility. To address these challenges, this paper proposes an interpretable wavelet basis unit convolutional network (WBUN). This network incorporates meticulously designed wavelet basis unit (WBU) functions into convolutional layer, creating the interpretable wavelet basis unit convolutional (WBUConv) layer. Convolutional kernels with clear physical significance enable the WBUConv layer to extract fault-related features in both time and frequency domains, enhancing diagnostic performance, and interpreting the CNN’s attention frequency along with the convolutional kernel’s training outcomes. In this paper, three WBU functions are designed to construct the corresponding WBUNs, and their effectiveness and interpretability are verified through three sets of mechanical fault diagnosis experiments. Meanwhile, experimental results demonstrate the WBUConv layer’s remarkable advantages in noise robustness, convergence speed, and strong generalization ability.

Improved motion correction in brain MRI using 3D radial trajectory and projection moment analysis.

To develop a generalized rigid body motion correction method in 3D radial brain MRI to deal with continuous motion pattern through projection moment analysis. An assumption was made that the multichannel coil moves with the head, which was achieved by using a flexible head coil. A two-step motion correction scheme was proposed to directly extract the motion parameters from the acquired k-space data using the analysis of center-of-mass with high noise robustness, which were used for retrospective motion correction. A recursive least-squares model was introduced to recursively estimate the motion parameters for every single spoke, which used the smoothness of motion and resulted in high temporal resolution and low computational cost. Five volunteers were scanned at 3 T using a 3D radial multidimensional golden-means trajectory with instructed motion patterns. The performance was tested through both simulation and in vivo experiments. Quantitative image quality metrics were calculated for comparison. The proposed method showed good accuracy and precision in both translation and rotation estimation. A better result was achieved using the proposed two-step correction compared to traditional one-step correction without significantly increasing computation time. Retrospective correction showed substantial improvements in image quality among all scans, even for stationary scans. The proposed method provides an easy, robust, and time-efficient tool for motion correction in brain MRI, which may benefit clinical diagnosis of uncooperative patients as well as scientific MRI researches.

Recursive demodulated synchro spline-kernelled chirplet extracting transform: a useful tool for non-linear behavior estimation of non-stationary signal and application to wind turbine fault detection

Non-linear behavior is widespread in many kinds of signals from nature and engineering fields. Although the high energy-concentration level of various advanced time-frequency (TF) analysis (TFA) techniques currently developed ensure a fine representation of non-linear behavior of time-varying component (TVC) of the signal, it is far from sufficient to solely consider the single aspect of energy-concentration level, because the actual signal composition is always more complicated, especially for some thorny difficulties such as strong noise interference and the early weak TVC, etc., these negative factors bring significant challenges to reveal the non-linear behavior of TVC of practical signals. A new TFA method aimed at this issue, called recursive demodulated synchro spline-kernelled chirplet extracting transform (RDSSCET), is proposed in this paper. The proposed RDSSCET is developed on the frame of synchro spline-kernelled chirplet extracting transform (SSCET) and a newly designed external-internal nested double iteration mechanism, which effectively addresses the limitation of SSCET in handling multicomponent signals while also exhibiting superior high energy concentration and noise robustness. As such, the proposed RDSSCET can yield a more favorable outcome when revealing the non-linear behavior of TVC, particularly for weak TVC with strong noise interference. Comparison analysis results in numerical simulations verified the performance of RDSSCET. Its effectiveness in real applications is fully tested via two real-world sound signals and a practical case of wind turbine fault detection.

Topologically Defective Lattice Potential‐Based Gain‐Dissipative Ising Annealer with Large Noise Margin

AbstractGain‐dissipative Ising machines (GIMs) are annealers inspired by physical systems such as Ising spin glasses to solve combinatorial optimization problems. Compared to traditional quantum annealers, GIM is relatively easier to scale and can save on additional power consumption caused by low‐temperature cooling. However, traditional GIMs have a limited noise margin. Specifically, their normal operation requires ensuring that the noise intensity is lower than their saturation fixed point amplitude, which may result in increased power consumption to suppress noise‐induced spin state switching. To enhance the noise robustness of GIM, in this study a GIM based on a topologically defective lattice potential (TDLP) is proposed. Numerical simulations demonstrate that the TDLP‐based GIM can accurately simulate the bifurcation spin evolution in the Ising model. Furthermore, through the MAXCUT benchmark based on G‐set graphs, the optimal performance of TDLP‐based GIM is shown to surpass that of traditional GIMs. Additionally, the proposed TDLP‐based GIM successfully solves the MAXCUT benchmark and domain clustering dynamics benchmark based on G‐set graphs when the noise intensity exceeds its saturation fixed‐point amplitude. This indicates that the proposed system provides a promising architecture for breaking the small noise constraints required by traditional GIMs.

Automatic Lyric Transcription and Automatic Music Transcription from Multimodal Singing

Automatic lyric transcription (ALT) refers to transcribing singing voices into lyrics, while automatic music transcription (AMT) refers to transcribing singing voices into note events, i.e., musical MIDI notes. Despite these two tasks having significant potential for practical application, they are still nascent. This is because the transcription of lyrics and note events solely from singing audio is notoriously difficult due to the presence of noise contamination, e.g., musical accompaniment, resulting in a degradation of both the intelligibility of sung lyrics and the recognizability of sung notes. To address this challenge, we propose a general framework for implementing multimodal ALT and AMT systems. Additionally, we curate the first multimodal singing dataset, comprising N20EMv1 and N20EMv2, which encompasses audio recordings and videos of lip movements, together with ground truth for lyrics and note events. For model construction, we propose adapting self-supervised learning models from the speech domain as acoustic encoders and visual encoders to alleviate the scarcity of labeled data. We also introduce a residual cross-attention mechanism to effectively integrate features from the audio and video modalities. Through extensive experiments, we demonstrate that our single-modal systems exhibit state-of-the-art performance on both ALT and AMT tasks. Subsequently, through single-modal experiments, we also explore the individual contributions of each modality to the multimodal system. Finally, we combine these and demonstrate the effectiveness of our proposed multimodal systems, particularly in terms of their noise robustness.

Motion-guided large aperture ULA to enhance DOA estimation: An inverse synthetic aperture perspective

Direction-of-arrival (DOA) estimation is a critical issue in array signal processing, especially in radar applications. However, the array aperture of high-frequency radars is constrained by the site and platform, which in turn leads to limited DOA estimation performance. In this paper, inspired by the inverse synthetic aperture radar, we design a novel DOA estimation method based on motion-guided large-aperture uniform linear array (ULA). Referring to the concept of inverse synthetic aperture, the novel method utilizes the relative motion between the radar and the target to construct a larger ULA. We demonstrate that the key condition for reconstructing ULAs is suitable data segmentation and reshaping within the coherent integration time. However, the correct segmentation position is related to the target's motion trajectory. For non-cooperative targets, the uncertain segmentation position is regarded as a mismatch problem between the steering and weighting vectors, such that the spatial spectrum becomes a product of two sinc-like functions. With the analysis of the main lobe and grating lobes, the maximum peak-to-grating-lobe ratio criterion is introduced to find the ambiguity-free DOA. The experimental results demonstrate the performance advantages of the proposed DOA estimation algorithm in terms of noise robustness and resolution.

Weight extracting transform for instantaneous frequency estimation and signal reconstruction

This paper introduces a novel time–frequency analysis method called the Weight Extracting Transform (WET). The primary objective of WET is to enhance the energy concentration in linear time–frequency representations for time-varying signals. By aggregating the time–frequency blur produced by the short-time Fourier transform onto the actual instantaneous frequency, WET improves the readability and accuracy of the time–frequency representation. Additionally, the algorithm is extended to a more general linear time–frequency transform known as the chirplet transform, which effectively handles fast-varying signals. The WET method excels at distinguishing components with closely spaced instantaneous frequencies and can reconstruct the original signal from its time–frequency representation. Experimental results using simulated and real-world signals demonstrate that WET achieves superior energy concentration and noise robustness compared to existing methods.

A Hybrid Neural Coding Approach for Pattern Recognition With Spiking Neural Networks.

Recently, brain-inspired spiking neural networks (SNNs) have demonstrated promising capabilities in solving pattern recognition tasks. However, these SNNs are grounded on homogeneous neurons that utilize a uniform neural coding for information representation. Given that each neural coding scheme possesses its own merits and drawbacks, these SNNs encounter challenges in achieving optimal performance such as accuracy, response time, efficiency, and robustness, all of which are crucial for practical applications. In this study, we argue that SNN architectures should be holistically designed to incorporate heterogeneous coding schemes. As an initial exploration in this direction, we propose a hybrid neural coding and learning framework, which encompasses a neural coding zoo with diverse neural coding schemes discovered in neuroscience. Additionally, it incorporates a flexible neural coding assignment strategy to accommodate task-specific requirements, along with novel layer-wise learning methods to effectively implement hybrid coding SNNs. We demonstrate the superiority of the proposed framework on image classification and sound localization tasks. Specifically, the proposed hybrid coding SNNs achieve comparable accuracy to state-of-the-art SNNs, while exhibiting significantly reduced inference latency and energy consumption, as well as high noise robustness. This study yields valuable insights into hybrid neural coding designs, paving the way for developing high-performance neuromorphic systems.

On the Compositional Step of the IC-GN Digital Image Correlation Algorithm

Digital Image Correlation (DIC) is a well-known experimental optical technique for measuring the displacement field based on the assumption that pixel intensity does not change with motion.DIC can be implemented using various alternative approaches. The most used in the Experimental Mechanics field are the Forward-Additive Gauss-Newton (FA-GN) and the Inverse-Compositional Gauss-Newton (IC-GN) formulations. The former corresponds to the original formulation proposed by Lucas and Kanade, while the latter was proposed by Baker and Matthews twenty years later. Although both formulations give the same results at the first order, their speed, convergency characteristic and noise robustness differ considerably. Nowadays, the IC-GN method is usually preferred because of its lower computational load and the significantly better noise sensitivity. However, the Inverse Compositional approach, as its name states, requires the inversion and composition of the displacement field, thus enforcing the use of invertible displacement fields. This can be a significant limitation because it may introduce an under-matching error in the solution.This work shows that a simple modification of the viewpoint makes the compositional step simpler, thus giving a DIC formulation as fast as that of the IC-GN, with the same noise-bias characteristic and without requiring an invertible displacement field.

Weak-form latent space dynamics identification

Recent work in data-driven modeling has demonstrated that a weak formulation of model equations enhances the noise robustness of a wide range of computational methods. In this paper, we demonstrate the power of the weak form to enhance the LaSDI (Latent Space Dynamics Identification) algorithm, a recently developed data-driven reduced order modeling technique.We introduce a weak form-based version WLaSDI (Weak-form Latent Space Dynamics Identification). WLaSDI first compresses data, then projects onto the test functions and learns the local latent space models. Notably, WLaSDI demonstrates significantly enhanced robustness to noise. With WLaSDI, the local latent space is obtained using weak-form equation learning techniques. Compared to the standard sparse identification of nonlinear dynamics (SINDy) used in LaSDI, the variance reduction of the weak form guarantees a robust and precise latent space recovery, hence allowing for a fast, robust, and accurate simulation. We demonstrate the efficacy of WLaSDI vs. LaSDI on several common benchmark examples including viscid and inviscid Burgers’, radial advection, and heat conduction. For instance, in the case of 1D inviscid Burgers’ simulations with the addition of up to 100% Gaussian white noise, the relative error remains consistently below 6% for WLaSDI, while it can exceed 10,000% for LaSDI. Similarly, for radial advection simulations, the relative errors stay below 16% for WLaSDI, in stark contrast to the potential errors of up to 10,000% with LaSDI. Moreover, speedups of several orders of magnitude can be obtained with WLaSDI. For example applying WLaSDI to 1D Burgers’ yields a 140X speedup compared to the corresponding full order model.Python code to reproduce the results in this work is available at (https://github.com/MathBioCU/PyWSINDy_ODE) and (https://github.com/MathBioCU/PyWLaSDI).

A sparse time-frequency reconstruction approach from the synchroextracting domain

Synchroextracting, including frequency-extracting and transient-extracting, offers a sparse time-frequency (TF) representation for multi-component signals. However, it is susceptible to amplitude modulation (AM) and frequency modulation (FM) when reconstructing the raw signal through the TF ridge. This paper proposes a new signal reconstruction approach (SRA) in the synchroextracting domain to address this limitation. This approach performs synchroextracting on the short-time Fourier transform (STFT) domain. We incorporate the extracting operator in the inverse STFT to establish the mathematical relationship between the synchroextracting representation and the retrieved signal, thus enabling the reconstruction of each mode in the synchroextracting domain. The proposed SRA is theoretically derived, and its reconstruction performance is evaluated in the frequency-extracting and transient-extracting domains. Two synthetic signals are utilized to demonstrate that compared with some advanced reconstruction methods, the SRA has a high reconstruction accuracy, strong noise robustness, and is less affected by AM-FM features and the window length. The application of the bat signal further clarifies the advantages of the SRA and its feasibility in practical applications. Moreover, this study also provides a new reconstruction framework and reference for other synchroextracting-type methods.

A novel diagnostic framework based on vibration image encoding and multi-scale neural network

Intelligent fault diagnosis employing CNN-based techniques has demonstrated promising results in rotating machinery maintenance and management. However, most approaches that rely on time-series vibration signals fail to incorporate the physical knowledge of faults into feature learning, and are subject to obtaining the correlation between input and output characteristics. Furthermore, it is challenging for the neural network to improve feature extraction due to the rigid convolution kernel and the limited receptive field. To address these issues, an intelligent fault diagnosis framework is developed based on vibration image encoding and multi-scale neural networks. Firstly, a novel image encoding rule is designed to convert time-series signals into vibration images, which incorporate the impact characteristics of faults into the model inputs to enhance fault information. Secondly, a multi-scale feature fusion network with dilatated and irregular sampling convolution block is constructed for feature extraction and classification. In addition, a new regularization term is designed to avoid training overfitting. Finally, Grad-CAM is introduced to visualize the region of interest of the model. The proposed method is verified on the datasets with different fault types from our laboratory and DIRG bearing dataset with different fault severities from Politecnico di Torino. The experimental results demonstrate that our diagnostic framework exhibits good generalization and noise robustness, and can achieve high-precision fault diagnosis.

An inversion problem for optical spectrum data via physics-guided machine learning

We propose the regularized recurrent inference machine (rRIM), a novel machine-learning approach to solve the challenging problem of deriving the pairing glue function from measured optical spectra. The rRIM incorporates physical principles into both training and inference and affords noise robustness, flexibility with out-of-distribution data, and reduced data requirements. It effectively obtains reliable pairing glue functions from experimental optical spectra and yields promising solutions for similar inverse problems of the Fredholm integral equation of the first kind.