Time-domain Signal Research Articles

A Tacholess Order Tracking Method Based on the STFTSC Algorithm for Rotor Unbalance Fault Diagnosis Under Variable-Speed Conditions

Abstract Due to the fact that rotors usually operate in a non-stationary mode with changing speeds, the conventional rotor unbalance detection method based on the stationary signal will produce a major “spectrum ambiguity issue” and affect the accuracy of rotor unbalance detection. To this end, a tacholess order tracking method based on the STFTSC algorithm is suggested in this study, where the STFTSC algorithm is developed by combining the short-time Fourier transform and the seam carving algorithm. First, the STFTSC algorithm is utilized to accurately extract the instantaneous frequency (IF) of the rotor and calculate the instantaneous phase under variable-speed conditions. Subsequently, the original signal is resampled in the angular domain to transform the non-stationary time domain signal into a stable angle domain signal, eliminating the effect of the speed variations. Finally, the angular domain signal is transformed into the order domain signal, which uses the discrete Fourier transform and the discrete spectrum correction method to identify the amplitude and phase corresponding to the fundamental frequency component of the signal. The simulation results show that the IF extracted by the STFTSC algorithm has higher extraction accuracy compared with the traditional STFT spectral peak detection method and effectively eliminates the effect of speed fluctuations. A rotor dynamic-balancing experiment shows that the unbalance correction effect based on the STFTSC algorithm is remarkable, with the average unbalance amount decrease rate on the left and right sides being 90.02% and 92.56%, respectively, after a single correction.

Deep transfer learning rolling bearing fault diagnosis method based on convolutional neural network feature fusion

Current deep-learning methods are often based on significantly large quantities of labeled fault data for supervised training. In practice, it is difficult to obtain samples of rolling bearing failures. In this paper, a transfer learning-based feature fusion convolutional neural network approach for bearing fault diagnosis is proposed. Specifically, the raw vibration signal features and the corresponding time-frequency image features of the input data are extracted by a one-dimensional convolutional neural network and a pre-trained ConvNeXt, respectively, and connected by a feature fusion strategy. Then, the fine-tuning method based on transfer learning can effectively reduce the reliance on labeled samples in the target domain. A wide convolution kernel is introduced in the time-domain signal feature extraction to increase the receptive field, which is combined with the channel attention mechanism to further optimize the feature quality. Finally, two common bearing datasets are utilized for fault diagnosis experiments. The experimental results show that the proposed model achieves an average accuracy of more than 98.63% in both cross-working conditions and cross-device diagnosis tasks. Meanwhile, anti-noise experiments and ablation experiments further validate the accuracy and robustness of the proposed method.

Prediction of the belt drive contamination status based on vibration analysis and artificial neural network

Belt drive contamination is considered one of the most common failure modes that could be developed in the belts due to harsh operation conditions, high humidity, and sunlight exposure, reducing the belt’s performance. If the belt failure has not been detected early, a sudden shutdown may happen, producing safety and economic consequences. However, most maintenance personnel use their senses of sight, hearing, smell, and touch to identify the cause of the problem while diagnosing a belt drive condition. Hence, this research involves developing an intelligent contamination status detection system based on vibration signal analysis for a pulley-belt rotating system. Time-domain signal analysis was employed to extract some suggestive features such as the root mean square, kurtosis, and skewness from the vibration data. An artificial neural network (ANN) model was built to detect the simulated different operating conditions. The vibration data was gathered with the help of two MEMS accelerometers (ADXL335) interfaced with an NI USB-6009 data acquisition device. A signal capture, analysis, and feature extraction system was developed using Matlab Simulink. The simulated operating conditions include clean, wet, and powder-contaminated belts. The results showed that the designed system could identify the pulley-belt operation conditions with 100% overall accuracy.

Frequency analysis of passive sonar signals using multiple-measurement physics-informed neural network

We propose a deep learning based frequency analysis for passive sonar signals by considering a linear system based on a relation between time and frequency domain signal representations. An adaptive learned iterative shrinkage thresholding algorithm (Ada-LISTA) is utilized as the deep learning architecture to solve the linear system effectively and to ensure generalization performance. To obtain more reliable solutions, we involve the linear system in loss function during training. Futhermore, in order to reduce noise effectively, we employ multiple measurements from time and space domains, which have common frequency components of passive sonar signals. Thus, the loss function is modified for solutions from the multiple measurements to have shared frequency components. We conduct experiments using synthetic and underwater in-situ data to examine performance of frequency analysis considering the linear system in designing the architecture and finding the optimal weights. The frequency analysis demonstrates superior performance in detecting frequency components and reducing noise, compared to fast Fourier transform and sparse Bayesian learning with low computational burden during its application to the test measurements.

Semi‐blind demodulation algorithm for short burst FM‐MFSK based on STFT

AbstractIn the blind reception of non‐cooperative signals in short‐time burst communication, signal time domain detection is the premise and key of signal modulation parameter estimation, signal demodulation, decoding, interpretation, information acquisition and interference guidance. In this paper, the semi‐blind demodulation method of short‐time burst FM‐MFSK signal is studied under the condition of low signal‐to‐noise ratio. A short‐time burst FM‐MFSK semi‐blind demodulation method based on STFT is proposed. When SNR is higher than 3 dB, the correct demodulation probability for FM‐MFSK signal can reach 90%.

Evaluation of skin sympathetic nervous activity for classification of intracerebral hemorrhage and outcome prediction

Classification and outcome prediction of intracerebral hemorrhage (ICH) is critical for improving the survival rate of patients. Early or delayed neurological deterioration is common in ICH patients, which may lead to changes in the autonomic nervous system (ANS). Therefore, we proposed a new framework for ICH classification and outcome prediction based on skin sympathetic nervous activity (SKNA) signals. A customized measurement device presented in our previous papers was used to collect data. 117 subjects (50 healthy control subjects and 67 ICH patients) were recruited for this study to obtain their 5-min electrocardiogram (ECG) and SKNA signals. We extracted the signal's time-domain, frequency-domain, and nonlinear features and analyzed their differences between healthy control subjects and ICH patients. Subsequently, we established the ICH classification and outcome evaluation model based on the eXtreme Gradient Boosting (XGBoost). In addition, heart rate variability (HRV) as an ANS assessment method was also included as a comparison method in this study. The results showed significant differences in most features of the SKNA signal between healthy control subjects and ICH patients. The ICH patients with good outcomes have a higher change rate and complexity of SKNA signal than those with bad outcomes. In addition, the accuracy of the model for ICH classification and outcome prediction based on the SKNA signal was more than 91% and 83%, respectively. The ICH classification and outcome prediction based on the SKNA signal proved to be a feasible method in this study. Furthermore, the features of change rate and complexity, such as entropy measures, can be used to characterize the difference in SKNA signals of different groups. The method can potentially provide a new tool for rapid classification and outcome prediction of ICH patients.Index Terms—intracerebral hemorrhage (ICH), skin sympathetic nervous activity (SKNA), classification, outcome prediction, cardiovascular and cerebrovascular diseases.

Validation of zero-group-velocity feature guided waves in a welded joint

Extensive research has been conducted on zero-group-velocity (ZGV) Lamb waves in elastic plates, demonstrating significant progress in the field of nondestructive testing. However, there is a scarcity of studies focusing on ZGV modes in complex structures. In this paper, we present our research investigating the presence of ZGV feature guided waves (FGWs) in a welded joint. Our approach follows a similar methodology used to study ZGV Lamb waves in elastic plates. By employing two-dimensional (2D) finite element (FE) modeling, we analyze the response spectra of the welded joint when subjected to a force source, revealing the occurrence of resonance in the response spectra. To investigate resonance modes in the welded joint, we employ the three-dimensional (3D) time-step FE method. By applying spatial 2D and short-time Fourier transforms to the received time-domain signals, we analyze the frequency content and spatial distribution of the signals. This analysis allows us to verify the existence of non-propagation and propagation modes in the welded joint. The non-propagation mode refers to the presence of signals with a zero wavenumber, indicating that they do not propagate or travel along the welded joint. These signals are typically associated with local resonances or vibrations within the welded joint itself. On the other hand, the propagation mode corresponds to signals with nonzero wavenumbers, suggesting that they propagate or travel along the welded joint. Furthermore, by further analyzing the propagation mode in the welded joint, similar to the analysis of ZGV modes in solid plates, we have observed that it also exhibits ZGV characteristics based on the wavenumber-frequency spectra. To further analyze acoustic field distributions at resonance frequencies, we utilize the semi-analytical finite element method in conjunction with the perfectly matched layer method. The results obtained from this analysis are consistent with those obtained from the 2D FE method and 3D time-step FE method, thereby confirming that propagation modes with ZGV characteristics at resonance frequencies correspond to FGWs, which we refer to as ZGV-FGWs. Through this step-by-step analysis, we ultimately establish the existence of ZGV-FGWs in the welded joint. This study introduces fresh ideas and serves as a point of reference for future research on ZGV-FGWs in complex structures.

Data-Driven DOA Estimation Methods Based on Deep Learning for Underwater Acoustic Vector Sensor Array

Abstract Direction of arrival (DOA) estimation is a fundamental problem in underwater acoustic vector sensor array signal processing. Because of the advantages of deep learning technology, this paper proposes two categories of data-driven DOA estimation methods for underwater acoustic vector sensor array, which transform the DOA estimation problem into a neural network classification problem. Specifically, one is the DOA estimation method of convolutional neural network based on teacher-student noise reduction (TS-CNN), which considers the covariance matrix as the training data set; the other is a DOA estimation method based on long-short term memory network and attention mechanism (LSTM-ATT), which applies the time-domain signal as the training data set. The experimental simulation results show that: 1) when the number of array elements is small, the accuracy of the DOA estimation method based on TS-CNN is equivalent to that of traditional methods, and it can effectively suppress the influence of noise when the signal-to-noise ratio (SNR) is low; 2) the accuracy of DOA estimation method based on LSTM-ATT is much higher than that of traditional Multiple Signal Classification method, especially in the case of low SNR, which also proves the importance of temporal characteristics for DOA estimation in a real environment.

Wavelet transform-based frequency self-adaptive model for functional brain network.

The accurate estimation of functional brain networks is essential for comprehending the intricate relationships between different brain regions. Conventional methods such as Pearson Correlation and Sparse Representation often fail to uncover concealed information within diverse frequency bands. To address this limitation, we introduce a novel frequency-adaptive model based on wavelet transform, enabling selective capture of highly correlated frequency band sequences. Our approach involves decomposing the original time-domain signal from resting-state functional magnetic resonance imaging into distinct frequency domains, thus constructing an adjacency matrix that offers enhanced separation of features across brain regions. Comparative analysis demonstrates the superior performance of our proposed model over conventional techniques, showcasing improved clarity and distinctiveness. Notably, we achieved the highest accuracy rate of 89.01% using Sparse Representation based on Wavelet Transform, outperforming Pearson Correlation based on Wavelet Transform with an accuracy of 81.32%. Importantly, our method optimizes raw data without significantly altering feature topology, rendering it adaptable to various functional brain network estimation approaches. Overall, this innovation holds the potential to advance the understanding of brain function and furnish more accurate samples for future research and clinical applications.

Bearing fault diagnosis based on CNN-BiLSTM and residual module

Bearings are key components of rotating machinery, and their fault diagnosis is essential for machinery operation. Bearing vibration signals belong to time series data, but traditional convolutional neural networks (CNNs) or recurrent neural networks cannot fully extract the fault features from these signals. To address the insufficient feature extraction and poor noise resistance, this paper proposes a fault diagnosis model based on continuous wavelet transform (CWT), CNN with channel attention, bidirectional long short-term memory network (BiLSTM) and residual module. Firstly, a parallel dual-path feature extraction mechanism is constructed which takes time-domain signals and time–frequency images transformed via CWT as the input respectively. Then BiLSTM extracts the time features of the signal as one path, and the CNN with efficient channel attention extracts the spatial features as the other path. This parallel neural network contributes to better feature extraction. Then, the residual module is applied to extract the global features to further improve the feature extraction ability and noise immunity. The experimental results demonstrate that the proposed model on the Case Western Reserve University dataset has better diagnostic accuracy under different working conditions and different signal-to-noise ratios than other methods. In addition, the model shows good generalization performance on Jiangnan University dataset.

Research on charge induction law and application of coal samples with different fissures

Indoor testing are performed to explore the charge induction law during the uniaxial compression fracture process of coal samples, and the charge time and frequency domain signals of coal samples with different primary fissures are analyzed in the paper. On-site monitoring of charge in different fissures distribution areas of underground coal tunnels, and the charge signals of different drillingdepths in coal seams are analyzed. The results show that the uniaxial compressive strength and elastic modulus of multi-fissured coal samples are less than those of less fissured coal samples, and the Poisson’s ratio is greater than those of less fissured coal samples. The charge induction signal intensity during the fracture process of multi-fissured coal samples is relatively low, but it is concentrated at the low frequency of 0–50 Hz in the compacting elasticity stage. The charge signal intensity during the fracture process of coal samples with less fissure is relatively high, and the charge frequency during the reinforcement damage stage is concentrated at a low frequency of 0–50 Hz. Therefore, the sudden appearance of low-frequency charge signals is more suitable as effective precursor information for the instability and failure of less fissured coal bodies. The average charge intensity is small in the multi-fissured area with a drilling depth of 1–4 m in the coal seam, and the average charge intensity of the coal body with less fissures is larger in the 5–12 m region. The on-site charge monitoring results have good consistency with the indoor test results. This study has guiding significance in setting up a charge monitoring warning index of instability failure in different coal body fissures regions.

Time-Frequency Analysis of Two-Dimensional Electron Spin Resonance Signals.

Two-dimensional electron spin resonance (2D ESR) spectroscopy is a unique experimental technique for probing protein structure and dynamics, including processes that occur at the microsecond time scale. While it provides significant resolution enhancement over the one-dimensional experimental setup, spectral broadening and noise make extraction of spectral information highly challenging. Traditionally, two-dimensional Fourier transform (2D FT) is applied for the analysis of 2D ESR signals, although its efficiency is limited to stationary signals. In addition, it often fails to resolve overlapping peaks in 2D ESR. In this work, we propose a time-frequency analysis of 2D time-domain signals, which identifies all frequency peaks by decoupling a signal into its distinct constituent components via projection on the time-frequency plane. The method utilizes 2D undecimated discrete wavelet transform (2D UDWT) as an intermediate step in the analysis, followed by signal reconstruction and 2D FT. We have applied the method to a simulated 2D double quantum coherence (DQC) signal for validation and a set of experimental 2D ESR signals, demonstrating its efficiency in resolving overlapping peaks in the frequency domain, while displaying frequency evolution with time in case of non-stationary data.

Sound source localization method based time-domain signal feature using deep learning

Deep learning, as the most commonly used machine learning algorithm, is widely used in various fields. In the field of acoustics, deep learning methods are combined with frequency-domain features of signals to locate sound sources. The commonly frequency domain features include microphones array Cross-spectral-Matrix(CSM) and Short Time Fourier Transform(STFT). However, the use of frequency-domain features often leads to the loss of partial signal information and increases the computational complexity. This paper proposed a novel sound source localization algorithm based on time-domain features, which uses convolutional neural network(CNN) as a medium to achieve mapping from time-domain features to sound source locations. This method does not rely on any basic signal processing algorithm, and directly uses time-domain sampling points as network inputs for sound source localization. The application simulation shows that the proposed method can achieve precise localization and low side-lobe effect under different testing conditions. Once the network training is completed, the testing accuracy under different conditions is above 95%, with a maximum of 100%.

Parameter identification method of nonuniform and under-sampled blade tip timing based on extended DFT and compressed sensing

Vibration amplitude and frequency are the two most important indicators that characterize the health status of high-speed rotating blades, but the signal obtained by blade tip timing (BTT) technology, one of the best rotating blade vibration monitoring methods, is seriously nonuniform and under-sampled, which makes these two indicators difficult to identify. In view of this problem, the paper proposes a parameter identification method for the nonuniform and under-sampled BTT signal based on extended Discrete Fourier transform and compressed sensing (CS), with the Fourier integral transformation as the goal. It realizes the frequency analysis of nonuniform under-sampled signals by constructing and optimizing the transformation basis function instead of the exponential basis in the traditional FFT transformation in the extended frequency range, and then constructs a CS model through the obtained blade vibration frequency. The complete waveform of the blade vibration is restored by using a small number of under-sampled signals, thus obtaining the blade vibration amplitude and vibration frequency. On the one hand, the method proposed in this paper breaks through the limitation of Nyquist’s sampling theorem, and the number of analytical spectral lines is no longer limited to the number of sampling points, which improves the frequency resolution. On the other hand, only a small number of measurement signals can be reconstructed to achieve a complete vibration signal. The feasibility and reliability of the proposed method are verified by mathematical modeling, simulation analysis, and experimental testing. The results indicate that when the number of sensors is greater than or equal to four, the time domain and frequency domain signals of blade vibration can be accurately analyzed based on the proposed method, the vibration amplitude error is less than 0.01 mm, the frequency error is less than 0.1 Hz, and it has good anti-interference performance.

EEG Emotion Recognition based on Multi scale Self Attention Convolutional Networks

A multi-view self-attention module is proposed and paired with a multi-scale convolutional model to builda multi-view self-attention convolutional network for multi-channel EEG emotion recognition. First, timeand frequency domain characteristics are extracted from multi-channel EEG signals, and a three-dimensionalfeature matrix is built using spatial mapping connections. Then, a multi-scale convolutional network extractsthe high-level abstract features from the feature matrix, and a multi-view self-attention network strengthensthe features. Finally, use the multilayer perceptron for sentiment classification. The experimental results revealthat the multi-view self-attention convolutional network can effectively integrate the time domain, frequencydomain, and spatial domain elements of EEG signals using the DEAP public emotion dataset. The multi-viewself-attention module can eliminate superfluous data, apply attention weight to the network to hasten networkconvergence, and enhance model recognition precision.

Assessment of Signal Similarity Index Based Line Protection Scheme for Systems With Inverter Based Resources

This article evaluates the performance of a signal similarity index (SSI) based relaying scheme for fault detection in lines connected to inverter-based resources (IBRs). During fault events, the IBRs respond differently than conventional synchronous machine based sources. Due to various control schemes and power electronic protection implemented for IBRs, fault current is limited leading to failure in fault detection and isolation by traditional protection methods. In this article, six different SSIs are calculated using time-domain current signals at both the ends of the protected line and the performance of SSI-based line protection philosophy is evaluated for various operating conditions. The relaying schemes are extensively tested on two different test systems. The test results are utilized to perform a comparative assessment of the SSI-based protection scheme. The tradeoff between the performance of the relaying scheme based on processing time, fault resistance, threshold selection, presence of noise, and presence of missing data are demonstrated thoroughly on three different test systems.

INSOMNet: Automated insomnia detection using scalogram and deep neural networks with ECG signals.

Sleep is a natural state of rest for the body and mind. It is essential for a human's physical and mental health because it helps the body restore itself. Insomnia is a sleep disorder that causes difficulty falling asleep or staying asleep and can lead to several health problems. Conventional sleep monitoring and insomnia detection systems are expensive, laborious, and time-consuming. This is the first study that integrates an electrocardiogram (ECG) scalogram with a convolutional neural network (CNN) to develop a model for the accurate measurement of the quality of sleep in identifying insomnia. Continuous wavelet transform has been employed to convert 1-D time-domain ECG signals into 2-D scalograms. Obtained scalograms are fed to AlexNet, MobileNetV2, VGG16, and newly developed CNN for automated detection of insomnia. The proposed INSOMNet system is validated on the cyclic alternating pattern (CAP) and sleep disorder research center (SDRC) datasets. Six performance measures, accuracy (ACC), false omission rate (FOR), sensitivity (SEN), false discovery rate (FDR), specificity (SPE), and threat score (TS), have been calculated to evaluate the developed model. Our developed system attained the classifications ACC of 98.91%, 98.68%, FOR of 1.5, 0.66, SEN of 98.94%, 99.31%, FDR of 0.80, 2.00, SPE of 98.87%, 98.08%, and TS 0.98, 0.97 on CAP and SDRC datasets, respectively. The developed model is less complex and more accurate than transfer-learning networks. The prototype is ready to be tested with a huge dataset from diverse centers.

Wide-Area Composite Load Parameter Identification Based on Multi-Residual Deep Neural Network.

Accurate and practical load modeling plays a critical role in the power system studies including stability, control, and protection. Recently, wide-area measurement systems (WAMSs) are utilized to model the static and dynamic behavior of the load consumption pattern in real-time, simultaneously. In this article, a WAMS-based load modeling method is established based on a multi-residual deep learning structure. To do so, a comprehensive and efficient load model founded on combination of impedance-current-power and induction motor (IM) is constructed at the first step. Then, a deep learning-based framework is developed to understand the time-varying and complex behavior of the composite load model (CLM). To do so, a residual convolutional neural network (ResCNN) is developed to capture the spatial features of the load at different location of the large-scale power system. Then, gated recurrent unit (GRU) is used to fully understand the temporal features from highly variant time-domain signals. It is essential to provide a balance between fast and slow variant parameters. Thus, the designed structure is implemented in a parallel manner to fulfill the balance and moreover, weighted fusion method is used to estimate the parameters, as well. Consequently, an error-based loss function is reformulated to improve the training process as well as robustness in the noisy conditions. The numerical experiments on IEEE 68-bus and Iranian 95-bus systems verify the effectiveness and robustness of the proposed load modeling approach. Furthermore, a comparative study with some relevant methods demonstrates the superiority of the proposed structure. The obtained results in the worst-case scenario show error lower than 0.055% considering noisy condition and at least 50% improvement comparing the several state-of-art methods.

Using Dynamic Phasors To Model and Analyze Selective Harmonic Compensated Single-Phase Grid-Forming Inverter Connected to Nonlinear and Resistive Loads

In an islanded microgrid, a grid-forming inverter (GFMI) is required to maintain a stable sinusoidal voltage at nominal frequency and amplitude with minimal distortion regardless of loading conditions. As a result, the accurate design of GFMI control schemes in an islanded microgrid requires accurate understanding of the converter dynamics and system harmonics. Electromagnetic transient (EMT) simulations based on detailed switching models give an accurate picture of the GFMI's behaviour. However, EMT simulations take a relatively long time to produce accurate results in large-scale systems. On the other hand, conventional phasor models are unsuitable for harmonic studies. To address this limitation, this article leverages the dynamic phasor (DP) method to develop a new model of a single-phase GFMI connected to single-phase nonlinear (diode-bridge rectifier) and resistive loads. The benefit of the DP method lies in its inherent feature of transforming time-domain signals into slow-varying or DC signals, thereby enabling faster simulations. In this article, the relationship between the control gains of the proposed DP model and the conventional detailed model is established for the first time through small-signal and frequency-domain-based comparative analyses. User-written codes of the proposed DP model are scripted in MATLAB, and simulations on MATLAB/Simulink show that the proposed DP model results have a high degree of correlation with the detailed switching model results. A significant savings in computation time is exhibited by the proposed DP models. The simulation results are validated against experimental results to confirm the accuracy of the proposed DP model.

Parameters identification of closely spaced modes with the covariance-drivern stochastic subspace and damping ratio dispersion method

Electromechanical equipment with quasi-symmetric and partially symmetrical structures is widespread in the industry, which not only makes the dynamic characteristics of the system more complex but also leads to severe closely spaced modes. The existence of closely spaced modes can lead to more confusion, because severe closely spaced modes have similar frequency eigenvalues and some singular eigenvalues are prone to generate spurious modes in matrix computations. In this paper, the covariance-drivern stochastic subspace and damping ratio dispersion (SSI-COV-DRD) method is proposed, which in comparison with previous methods, identification process directly is performed in the stochastic subspace by solving the system matrix of the time domain signal to overcome modal aliasing of the frequency-based method. Firstly, a Toeplitz matrix is constructed with a time series signal and then the system modal parameters are gained from the system matrix after performing singular value decomposition (SVD) to the Toeplitz matrix. Subsequently, the poles of the continuous model order are classified by a hierarchical clustering algorithm to obtain stabilization diagram and the real physical modes are automatically extracted from the stabilization diagram according to the choice criteria. Finally, the damping ratio dispersion index is calculated to express the nonlinear distribution characteristic of the damping ratio. Numerical examples for the five-order free vibration simulation signals with noise-free and noise indicate that the proposed method can simultaneously identify the modal frequency and damping ratio with high accuracy in severe closely spaced modes, especially for the repeated frequency modes. Since the bolt connection rotor of the aero-engine usually produces severe closely spaced modes in vibration due to the structural symmetry, the frequency and damping ratio are extracted and the damping ratio dispersion index is calculated from different bolt loosening states, which is helpful for the bolt loosening detection of aviation safety assurance and maintenance.