Time Series Signals Research Articles

1D-CNN: Classification of normal delivery and cesarean section types using cardiotocography time-series signals

Abstract Cardiotocography (CTG) is considered the gold standard for monitoring fetal heart rate (FHR) during pregnancy and labor to estimate the danger of oxygen deprivation. Visual interpretation of CTG traces is complex and frequently results in high rates of false positives and false negatives, leading to unfavorable and unwanted outcomes such as fetal mortality or needless cesarean surgery. If the data are well-balanced, which is uncommon in medical datasets, machine learning techniques can be helpful in interpretation. This study is designed to determine classification performance under various data balance approaches. We propose a robust methodology for the automated extraction of features that use a deep learning model based on the one-dimensional convolutional neural network (1D-CNN). We used a public database containing 552 intrapartum CTG recordings. Due to the imbalance in the dataset, the experiments were conducted under a variety of conditions such as (i) an unbalanced dataset, (ii) undersampling, (iii) a weighted binary cross-entropy approach, and (iv) oversampling utilizing the synthetic minority oversampling technique (SMOTE). We found an excellent sensitivity (99.80% for the unbalanced dataset, 96.25% for the weighted binary cross-entropy approach, and 99.81% with SMOTE) except for the under sampling situation, in which the sensitivity was 85.71%. Moreover, the 1D-CNN model incorporating SMOTE yielded promising results in 88% specificity, 93.72% quality index (QI), and 95.10% area under the curve. The model exhibited excellent performance in terms of sensitivity in every scenario except for undersampling. The oversampling of training data with SMOTE yielded a decent level of specificity, demonstrating the model’s strong predictive capacity. In addition, the SMOTE scenario resulted in fewer training epochs, which is another accomplishment.

HeMTAN: Hybrid task-adapted experts-based multi-task attention network for unseen compound fault decoupling diagnosis of rotating machinery

In a rotating machinery system, a single fault of one component often causes damage to other related components, thus inducing compound faults. Without compound fault data to train intelligent models, the realization of decoupling diagnosis and accurate identification of unseen compound faults is not only of great practical significance for the safety management of equipment operation and maintenance, but also remains a challenging topic. Considering some shortcomings in the current intelligent diagnosis of compound faults, as well as the relatedness and difference between different fault features in compound fault signals, a hybrid task-adapted experts-based multi-task attention network (HeMTAN) model is investigated in this paper, which can be used for identify single faults and unseen compound faults in mechanical transmission systems. Firstly, variational mode decomposition is combined with Hilbert-Huang transform to obtain time–frequency graphs of time series signal as model input, so as to better characterize different fault features. Secondly, a hybrid task-adapted expert module is designed to extract the common and some private feature information of different learning tasks from different multi-perspective, and then the important information related to the specific learning task is further mined by the constructed private feature attention-based densely connected module. Finally, the diagnosis results can be obtained by fusing the outputs of the classifier of the two learning tasks. The performance of the investigated HeMTAN model is analyzed by the gearbox compound fault dataset and rolling bearing compound fault dataset, and the results demonstrate that the investigated HEMTAN method has significantly improved diagnosis accuracy and generalization performance.

Singular spectrum analysis for the time-variable seasonal signals from GPS in Yunnan Province

Studying the seasonal deformation in GPS time series is important to interpreting geophysical contributors and identifying unmodeled and mismodeled seasonal signals. Traditional seasonal signal extraction used the least squares method, which models seasonal deformation as a constant seasonal amplitude and phase. However, the seasonal variations are not constant from year to year, and the seasonal amplitude and phase are time-variable. In order to obtain the time-variable seasonal signal in the GPS station coordinate time series, singular spectrum analysis (SSA) is conducted in this study. We firstly applied the SSA on simulated seasonal signals with different frequencies 1.00 cycle per year (cpy), 1.04 cpy and with time-variable amplitude are superimposed. It was found that SSA can successfully obtain the seasonal variations with different frequencies and with time-variable amplitude superimposed. Then, SSA is carried out on the GPS observations in Yunnan Province. The results show that the time-variable amplitude seasonal signals are ubiquitous in Yunnan Province, and the time-variable amplitude change in 2019 in the region is extracted, which is further explained by the soil moisture mass loading and atmospheric pressure loading. After removing the two loading effects, the SSA obtained modulated seasonal signals which contain the obvious seasonal variations at frequency of 1.046 cpy, it is close with the GPS draconitic year, 1.040 cpy. Hence, the time-variable amplitude changes in 2019 and the seasonal GPS draconitic year in the region could be discriminated successfully by SSA in Yunnan Province.

MEET: A Multi-Band EEG Transformer for Brain States Decoding.

Electroencephalography (EEG) is among the most widely used and inexpensive neuroimaging techniques. Compared to the CNN or RNN based models, Transformer can better capture the temporal information in EEG signals and focus more on global features of the brain's functional activities. Importantly, according to the multiscale nature of EEG signals, it is crucial to consider the multi-band concept into the design of EEG Transformer architecture. We propose a novel Multi-band EEG Transformer (MEET) to represent and analyze the multiscale temporal time series of human brain EEG signals. MEET mainly includes three parts: 1) transform the EEG signals into multi-band images, and preserve the 3D spatial information between electrodes; 2) design a Band Attention Block to compute the attention maps of the stacked multi-band images and infer the fused feature maps; 3) apply the Temporal Self-Attention and Spatial Self-Attention modules to extract the spatiotemporal features for the characterization and differentiation of multi-frame dynamic brain states. The experimental results show that: 1) MEET outperforms state-of-the-art methods on multiple open EEG datasets (SEED, SEED-IV, WM) for brain states classification; 2) MEET demonstrates that 5-bands fusion is the best integration strategy; and 3) MEET identifies interpretable brain attention regions. MEET is an interpretable and universal model based on the multiband-multiscale characteristics of EEG. The innovative combination of band attention and temporal/spatial self-attention mechanisms in MEET achieves promising data-driven learning of the temporal dependencies and spatial relationships of EEG signals across the entire brain in a holistic and comprehensive fashion.

Microseismic source location using deep learning: A coal mine case study in China

Microseismic source location is crucial for the early warning of rockburst risks. However, the conventional methods face challenges in terms of the microseismic wave velocity and arrival time accuracy. Intelligent techniques, such as the full convolutional neural network (FCNN), can capture spatial information but struggle with complex microseismic sequence. Combining the FCNN with the long short-term memory (LSTM) network enables better time-series signal classification by integrating multi-scale information and is therefore suitable for waveform location. The LSTM-FCNN model does not require extensive data preprocessing and it simplifies the microseismic source location through feature extraction. In this study, we utilized the LSTM-FCNN as a regression learning model to locate the seismic focus. Initially, the method of short-time-average/long-time-average (STA/LTA) arrival time picking was employed to augment spatiotemporal information. Subsequently, oversampling the on-site data was performed to address the issue of data imbalance, and finally, the performance of LSTM-FCNN was tested. Meanwhile, we compared the LSTM-FCNN model with previous deep-learning models. Our results demonstrated remarkable location capabilities with a mean absolute error (MAE) of only 7.16 m. The model can realize swift training and high accuracy, thereby significantly improving risk warning of rockbursts.

Experimental study on the disturbance of microbubble liquid jets

The presence of microbubble significantly influences the rheological properties of a jet, thereby considerably increasing the complexity of the jet instability problem. However, the instability of microbubble jets has been rarely studied, and their breakup mechanism remains obscure. The present study was focused on conducting a comprehensive analysis of bubble jet flows. A high-precision experimental system was constructed to extract the time-series signals of each cross-sectional profile along the axial direction of a jet. Glycerol solutions with three different bubble volume fractions were prepared, and the standard deviation, spectral relationship, and nonlinear dynamical behavior (i.e., autocorrelation, phase space trajectory, Poincaré section, and Lyapunov exponent) were studied. The detailed development process of small surface disturbances in space was assessed, and the influence of variations in the bubble volume fraction on the nonlinear dynamics of the jet was examined. The results demonstrate that microbubbles can significantly change the energy transfer relationship between various sections of a jet and enhance the periodicity of the signal.

Interpretable Recurrent Variational State-Space Model for Fault Detection of Complex Systems Based on Multisensory Signals

It is necessary to develop a health monitoring system (HMS) for complex systems to improve safety and reliability and prevent potential failures. Time-series signals are collected from multiple sensors installed on the equipment that can reflect the health condition of them. In this study, a novel interpretable recurrent variational state-space model (IRVSSM) is proposed for time-series modeling and anomaly detection. To be specific, the deterministic hidden state of a recursive neural network is used to capture the latent structure of sensor data, while the stochastic latent variables of a nonlinear deep state-space model capture the diversity of sensor data. Temporal dependencies are modeled through a nonlinear transition matrix; an automatic relevance determination network is introduced to selectively emphasize important sensor data. Experimental results demonstrate that the proposed algorithm effectively captures vital information within the sensor data and provides accurate and reliable fault diagnosis during the steady-state phase of liquid rocket engine operation.

Meta-learning-based approach for tool condition monitoring in multi-condition small sample scenarios

Tool Condition Monitoring (TCM) technology in machining is crucial for maintaining safety and optimizing costs. However, its practical application faces two significant challenges: difficulties in data collection and a decline in generalization performance across different monitoring tasks. To this end, a hybrid feature boundary-enhanced meta-learning network with adaptive gradients (HFBEAML) is proposed. This method combines cutting process parameters with time-series signal features, employing a multimodal one-dimensional Convolutional Neural Network (1D-CNN) based on the DenseNet architecture and a multi-head self-attention mechanism (MHSA) to mine multi-dimensional features sensitive to tool wear. To further enhance the model’s feature discrimination capability, a multi-loss joint optimization strategy is introduced, combining task-level loss with an improved triplet loss. This approach incorporates strategies for adaptive boundaries and handling the hardest positive and negative sample sets, thereby enhancing the robustness of feature metrics. Additionally, this research innovatively proposes an adaptive meta-level gradient update mechanism, dynamically adjusting gradient weights according to the characteristics of multiple tasks, aiming to improve the model’s generalization ability in multi-task learning environments. The effectiveness of the proposed method is demonstrated through experiments in two different scenarios, comparing its results with five other models showcasing its significant advantages in multi-task, small-sample data monitoring environments.

Fault Diagnosis Method of Flexible Converter Valve Equipment Based on Ensemble Empirical Mode Decomposition and Temporal Convolutional Networks

The flexible converter valve is a crucial component of the flexible transmission system, and its proper functioning is directly related to the power system's reliability. This study proposes a method for diagnosing faults in flexible converter valve equipment based on ensemble empirical mode decomposition and temporal convolutional networks. The method involves measuring voltage signal data in the power submodule of the flexible converter valve, decomposing and reconstructing the voltage signal using ensemble empirical mode decomposition to extract frequency variation patterns and fault features. Subsequently, a temporal convolutional network is introduced, and a device fault diagnosis model is constructed by learning the evolution law of voltage signals in time series. The experimental results demonstrate that the proposed method has high fault diagnosis accuracy and robustness with an average F1-score of 89.58% and an average area under the curve (AUC) of 94.38%, which are higher than other methods by at least 1.39% and 1.03%.

Resting-state MRI reveals spontaneous physiological fluctuations in the kidney and tracks diabetic nephropathy in rats.

The kidneys maintain fluid-electrolyte balance and excrete waste in the presence of constant fluctuations in plasma volume and systemic blood pressure. The kidneys perform these functions to control capillary perfusion and glomerular filtration by modulating the mechanisms of autoregulation. An effect of these modulations are spontaneous, natural fluctuations in glomerular perfusion. Numerous other mechanisms can lead to fluctuations in perfusion and flow. The ability to monitor these spontaneous physiological fluctuations in vivo could facilitate the early detection of kidney disease. The goal of this work was to investigate the use of resting-state magnetic resonance imaging (rsMRI) to detect spontaneous physiological fluctuations in the kidney. We performed rsMRI of rat kidneys in vivo over 10 min, applying motion correction to resolve time series in each voxel. We observed spatially variable, spontaneous fluctuations in rsMRI signal between 0 and 0.3 Hz, in frequency bands associated with autoregulatory mechanisms. We further applied rsMRI to investigate changes in these fluctuations in a rat model of diabetic nephropathy. Spectral analysis was performed on time series of rsMRI signals in the kidney cortex and medulla. The power from spectra in specific frequency bands from the cortex correlated with severity of glomerular pathology caused by diabetic nephropathy. Finally, we investigated the feasibility of using rsMRI of the human kidney in two participants, observing the presence of similar, spatially variable fluctuations. This approach may enable a range of preclinical and clinical investigations of kidney function and facilitate the development of new therapies to improve outcomes in patients with kidney disease.NEW & NOTEWORTHY This work demonstrates the development and use of resting-state MRI to detect low-frequency, spontaneous physiological fluctuations in the kidney consistent with previously observed fluctuations in perfusion and potentially due to autoregulatory function. These fluctuations are detectable in rat and human kidneys, and the power of these fluctuations is affected by diabetic nephropathy in rats.

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.

A research on an improved fuzzy approximate entropy algorithm for EMG-based shoulder and neck muscle fatigue detection

A robust muscle fatigue algorithm plays a pivotal role in depicting the degree of muscle fatigue in both time-series EMG signal graphs and spectral graphs, aligning with human perception. While the fuzzy approximate entropy (fApEn) algorithm has been enhanced from the foundation of approximate entropy (ApEn) through the incorporation of fuzzy affiliation, concerns persist regarding the threshold value and the algorithm’s application range. This study extracts EMG signals across varied time durations and head-down angles, employing enhanced signal preprocessing techniques and optimizing the fApEn algorithm. Furthermore, real-time fatigue perceptions of subjects were recorded using the rating of perceived exertion. Experimental outcomes reveal that the EMG signal, post-wavelet analysis preprocessing, demonstrates promising noise reduction capabilities. Notably, the fApEn algorithm exhibits considerable enhancements through the identification of an optimal threshold using the gradient descent algorithm and a machine learning strategy.

An optoacoustic field-programmable perceptron for recurrent neural networks.

Recurrent neural networks (RNNs) can process contextual information such as time series signals and language. But their tracking of internal states is a limiting factor, motivating research on analog implementations in photonics. While photonic unidirectional feedforward neural networks (NNs) have demonstrated big leaps, bi-directional optical RNNs present a challenge: the need for a short-term memory that (i) programmable and coherently computes optical inputs, (ii) minimizes added noise, and (iii) allows scalability. Here, we experimentally demonstrate an optoacoustic recurrent operator (OREO) which meets (i, ii, iii). OREO contextualizes the information of an optical pulse sequence via acoustic waves. The acoustic waves link different optical pulses, capturing their information and using it to manipulate subsequent operations. OREO's all-optical control on a pulse-by-pulse basis offers simple reconfigurability and is used to implement a recurrent drop-out and pattern recognition of 27 optical pulse patterns. Finally, we introduce OREO as bi-directional perceptron for new classes of optical NNs.

An ensemble maximal feature subset selection for smartphone based human activity recognition

Smartphone-based Human Activity Recognition (HAR) uses spatiotemporal time series data collected from a smartphone’s in-built accelerometer, gyroscope, and magnetometer sensor. Real-time HAR datasets such as UCI-HAR and UCI-HAPT extract time and frequency domain statistical features for activity recognition from the collected time series signals. Various Machine Learning techniques have been proposed in the perspective of feature selection through ensemble, traditional, hybrid, and meta-heuristic-based techniques to improve the recognition performance. Though the feature selection techniques have shown promising results, the computational time and selection of valid significant features for better classification performance still need to be included. Moreover, these techniques are highly parameter-dependent and critical to threshold values used for experimentation. This research proposes a novel method of parameter-free ensemble feature selection by combining traditional feature selection algorithms with the concept of association rule mining to obtain the Maximal Feature Subset (MFS), which is used for training and testing the classification model for HAR. Mathematical association rule mining is framed rather than the traditional one to select the associated frequent feature(s) from the feature set evolved by traditional techniques. Experimentation has been carried out on UCI-HAR and UCI-HAPT, and the proposed method has increased the classification efficiency by selecting significant features and reduces computation time significantly.

Modulate the impact of the drowsiness on the resting state functional connectivity.

This research explores different methodologies to modulate the effects of drowsiness on functional connectivity (FC) during resting-state functional magnetic resonance imaging (RS-fMRI). The study utilized a cohort of students (MRi-Share) and classified individuals into drowsy, alert, and mixed/undetermined states based on observed respiratory oscillations. We analyzed the FC group difference between drowsy and alert individuals after five different processing methods: the reference method, two based on physiological and a global signal regression of the BOLD time series signal, and two based on Gaussian standardizations of the FC distribution. According to the reference method, drowsy individuals exhibit higher cortico-cortical FC than alert individuals. First, we demonstrated that each method reduced the differences between drowsy and alert states. The second result is that the global signal regression was quantitively the most effective, minimizing significant FC differences to only 3.3% of the total FCs. However, one should consider the risks of overcorrection often associated with this methodology. Therefore, choosing a less aggressive form of regression, such as the physiological method or Gaussian-based approaches, might be a more cautious approach. Third and last, using the Gaussian-based methods, cortico-subcortical and intra-default mode network (DMN) FCs were significantly greater in alert than drowsy subjects. These findings bear resemblance to the anticipated patterns during the onset of sleep, where the cortex isolates itself to assist in transitioning into deeper slow wave sleep phases, simultaneously disconnecting the DMN.

Virtual tomography as a novel method for segmenting machining process phases with the use of machine learning-supported measurement

A new idea of machine learning-based technological process segmentation with the use of multi-sensor measurement is proposed in this article. The proposed segmentation of the machining process through appropriate measurement data modelling provides valuable insight that is necessary for technological parameters optimization or predictive maintenance. By combining multi-sensor industrial measurement and data science, this new solution is a more advanced and effective way of determining qualitative characteristics of the machining process that must be taken into account when developing smart analytical approaches, such as digital twins. An experimental measuring system consisting of accelerometers and current transducers is described. The system is implemented on an industrial CNC machine in order to assess operating conditions of the spindle and axis drives during the milling process. A novel method of detecting and segmenting the milling phases is proposed, involving data preprocessing and time-series signals data analysis to detect specific patterns or features that are indicative of each phase of the milling process. When applied in smart structures or digital twins, the proposed segmentation method provides similar information to that obtained by tomography imaging; therefore, the new method is called as virtual tomography. The developed phase segmentation method is of vital practical importance in terms of industrial implementation of technological process measurement and diagnostic systems. Especially as it can provide valuable information concerning elementary cutting zones and their influence on the process efficiency or influence of the composite structure layers machining on the tool wear, not available using hitherto known methods.

Cardiac Arrhythmia Classification Using Advanced Deep Learning Techniques on Digitized ECG Datasets.

ECG classification or heartbeat classification is an extremely valuable tool in cardiology. Deep learning-based techniques for the analysis of ECG signals assist human experts in the timely diagnosis of cardiac diseases and help save precious lives. This research aims at digitizing a dataset of images of ECG records into time series signals and then applying deep learning (DL) techniques on the digitized dataset. State-of-the-art DL techniques are proposed for the classification of the ECG signals into different cardiac classes. Multiple DL models, including a convolutional neural network (CNN), a long short-term memory (LSTM) network, and a self-supervised learning (SSL)-based model using autoencoders are explored and compared in this study. The models are trained on the dataset generated from ECG plots of patients from various healthcare institutes in Pakistan. First, the ECG images are digitized, segmenting the lead II heartbeats, and then the digitized signals are passed to the proposed deep learning models for classification. Among the different DL models used in this study, the proposed CNN model achieves the highest accuracy of ∼92%. The proposed model is highly accurate and provides fast inference for real-time and direct monitoring of ECG signals that are captured from the electrodes (sensors) placed on different parts of the body. Using the digitized form of ECG signals instead of images for the classification of cardiac arrhythmia allows cardiologists to utilize DL models directly on ECG signals from an ECG machine for the real-time and accurate monitoring of ECGs.

New graphical models for sequential data and the improved state estimations by data-conditioned driving noises

A prevalent problem in statistical signal processing, applied statistics, and time series analysis arises from the attempt to identify the hidden state of Markov process based on a set of available noisy observations. In the context of sequential data, filtering refers to the probability distribution of the underlying Markovian system given the measurements made at or before the time of the estimated state. In addition to the filtering, the smoothing distribution is obtained from incorporating measurements made after the time of the estimated state into the filtered solution. This work proposes a number of new filters and smoothers that, in contrast to the traditional schemes, systematically make use of the process noises to give rise to enhanced performances in addressing the state estimation problem. In doing so, our approaches for the resolution are characterized by the application of the graphical models; the graph-based framework not only provides a unified perspective on the existing filters and smoothers but leads us to design new algorithms in a consistent and comprehensible manner. Moreover, the graph models facilitate the implementation of the suggested algorithms through message passing on the graph.

Enhancing Magnetic Material Data Analysis with Genetic Algorithm-Optimized Variational Mode Decomposition

As the application of machine learning technology in predicting and optimizing material performance continues to grow, handling the electromagnetic data of magnetic materials, especially in removing unavoidable data noise and accurately extracting resonance peaks in the imaginary part of electromagnetic information, has become a significant challenge. These steps are crucial for revealing the deep electromagnetic behavior of materials and optimizing their performance. In response to this challenge, this study introduces an innovative approach—Genetic Algorithm-Optimized Variational Mode Decomposition for Signal Enhancement (GAO-VMD-SE). This method, through the Variational Mode Decomposition (VMD) technique optimized by genetic algorithms, not only effectively reduces noise in the data, thereby improving the Signal-to-Noise Ratio (SNR) and reducing the Mean Absolute Error (MAE), but also significantly enhances the hidden resonance peak information in complex permittivity and permeability data to achieve a comprehensive improvement in key performance indicators. Experimental results prove that this method surpasses traditional analysis techniques in key performance metrics such as the peak width ratio, peak overlap ratio, and the number of peaks. Especially in identifying characteristic peaks related to the Snoek limit, GAO-VMD-SE can effectively reveal the peak features hidden in complex data, thus providing important insights for evaluating the performance of materials at specific frequencies. Moreover, the effectiveness of this method in denoising not only enhances the quality and accuracy of material data analysis but also achieves a 1% to 10% enhancement in peak information extraction. This optimized data processing capability and versatility make GAO-VMD-SE not only suitable for evaluating the performance of magnetic materials but also show significant practical application value in processing spectral data and other time series signal data applications.

A Real-Time Fault Diagnosis Method for Grid-Side Overcurrent in Train Traction System Using Signal Time Series Feature Pattern Recognition

Grid-side overcurrent (GSOC) is a common anomaly state in traction drive system (TDS). Once GSOC occurs, to protect the train, traction control units (TCU) will trip the main circuit breaker and stop the train in a short time. Since TCU cannot locate the source of the GSOC fault, troubleshooting can only be done by inspecting the components one by one after the train stops, which greatly affects the availability and maintenance efficiency of train. Therefore, this paper proposes a real-time fault diagnosis method based on time series feature pattern recognition of signals in TCU. Firstly, the time series pattern of GSOC caused by different fault sources are established. Secondly, the feature extraction method of time series pattern is carried out based on the fusion of engineering knowledge, variational mode decomposition and signal selection. Thirdly, the time series feature patterns of GSOC are identified to determine the most likely fault type of GSOC by Gaussians mixture model-Hidden Markov model (GMM-HMM). Finally, the proposed real-time fault diagnosis method is verified by the field test data. The results show that the proposed algorithm can realize real-time diagnosis of GSOC and trace the fault source effectively.