Time Domain Features Research Articles

Exploiting the Electrochemical Impedance Spectroscopy Frequency Profiles for State-of-Health Predication of Lithium-Ion Battery

Battery Management Systems (BMS) are essential for optimizing battery performance and extending lifespan through continuous monitoring and decision-making via control sensors. The State of Health (SOH) is one of the BMS metrics that provides valuable information on battery health and degradation. However, one of the main challenges in the BMS domain development is finding accurate and effective algorithms for battery SOH prediction, especially for electric vehicles and grid-connected energy storage systems. This study introduces a new SOH prediction method using wavelet-convolutional neural regression networks (CNRN) algorithms. The methodology involves extracting detailed frequency profiles from Electrochemical Impedance Spectroscopy (EIS) data, which are processed through wavelet transformation to capture both time and frequency domain features. These transformed profiles are then input into the CNRN model for SOH prediction. The results demonstrate improved SOH prediction accuracy with EIS frequency profiles, evidenced by a reduction in root mean square error (RMSE) compared to the standard EIS profile. This improvement is due to the fact that the wavelet-CNRN algorithm efficiently captures both the time and frequency features of the battery impedance. Moreover, the performance of the proposed algorithm demonstrated robustness in early end-of-life (EOL) prediction, thereby enhancing the reliability and safety of BMS functions.

Development of PIND system based on resonant sensors and research on material identification methods for redundant particles

Abstract Aiming at the problems of the existing particle impact noise detection (PIND) system, such as low detection accuracy, high rate of misjudgment and missed judgment, and the inability to judge the material of internal redundant particles in real time, a method for automatic detection and material identification of redundant particles based on neural network recognition is proposed, and an automatic detection system for redundant particles of sealed electronic devices is developed. The system is equipped with multi-channel piezoelectric resonant sensors, which convert the redundant particle impact noise signal into a voltage signal, and then transmit the signal to the host using a high-speed data acquisition system. The upper computer uses a short-term energy threshold detection method to extract signal pulses; Then, a variety of time and frequency domain features are extracted and combined with wavelet domain features; Finally, by training the multi output neural network prediction model, it can automatically judge whether there are redundant particles in the tested parts, and automatically give the material information of the redundant particles. Experiments show that the accuracy of the system for material recognition of redundant particles with a mass greater than 0.1 mg can reach 88.6%.

MPDDF: Multi perspective dual data flow model for time series forecasting

Abstract Multivariate time series forecasting tasks are widely used in social production and daily life, and making accurate predictions for the future can help enterprises take preventive measures for upcoming events. The prediction method based on deep neural networks has become the main method used in time series forecasting tasks. In order to improve the performance of the prediction model, this paper designs methods to improve the prediction performance of the model from the perspectives of multi-dimensional, time-domain, and data distribution in multivariate time series data. Using graph neural networks to establish dependencies between multi-dimensional data and perform feature extraction, using dual flow learning networks for time-domain feature extraction, and preventing overfitting of the model during training, and using learnable bias values to mitigate the impact of time series data distribution shift. By conducting experiments on the model in five different application scenarios and comparing it with advanced time series forecasting models, an average improvement of 7.50% was achieved in the MSE index and 6.53% in the MAE index, demonstrating the effectiveness of the model designed in this paper.

A Machine Learning Approach to Classifying EEG Data Collected with or without Haptic Feedback during a Simulated Drilling Task.

Artificial Intelligence (AI), computer simulations, and virtual reality (VR) are increasingly becoming accessible tools that can be leveraged to implement training protocols and educational resources. Typical assessment tools related to sensory and neural processing associated with task performance in virtual environments often rely on self-reported surveys, unlike electroencephalography (EEG), which is often used to compare the effects of different types of sensory feedback (e.g., auditory, visual, and haptic) in simulation environments in an objective manner. However, it can be challenging to know which aspects of the EEG signal represent the impact of different types of sensory feedback on neural processing. Machine learning approaches offer a promising direction for identifying EEG signal features that differentiate the impact of different types of sensory feedback during simulation training. For the current study, machine learning techniques were applied to differentiate neural circuitry associated with haptic and non-haptic feedback in a simulated drilling task. Nine EEG channels were selected and analyzed, extracting different time-domain, frequency-domain, and nonlinear features, where 360 features were tested (40 features per channel). A feature selection stage identified the most relevant features, including the Hurst exponent of 13-21 Hz, kurtosis of 21-30 Hz, power spectral density of 21-30 Hz, variance of 21-30 Hz, and spectral entropy of 13-21 Hz. Using those five features, trials with haptic feedback were correctly identified from those without haptic feedback with an accuracy exceeding 90%, increasing to 99% when using 10 features. These results show promise for the future application of machine learning approaches to predict the impact of haptic feedback on neural processing during VR protocols involving drilling tasks, which can inform future applications of VR and simulation for occupational skill acquisition.

Gated recurrent unit and temporal convolutional network with soft thresholding and attention mechanism for tool wear prediction

Predicting tool wear significantly improves cutting efficiency and quality in intelligent manufacturing. Traditional recurrent neural networks suffer from vanishing or exploding gradients and need to be improved prediction accuracy. Therefore, this paper proposes a gated recurrent unit and temporal convolutional network with soft thresholding and attention mechanism (TCN-BiGRU-SA). Monotonicity, robustness, and trendability indicators are used to assess time and frequency domain features. Then, evaluation results of these three indicators are weighted to identify the sensitive features with the degradation trend. The attention mechanism is proposed to adaptively adjust the soft thresholding, and the SA is added to TCN-BiGRU to remain helpful features. Finally, the proposed model is validated using an experimental dataset. Compared to the TCN-BiGRU model without the SA mechanism, the predicted results show that the average mean absolute error and root mean square error are reduced by 48.29 % and 36.39 %, respectively.

Advanced Sensing System for Sleep Bruxism across Multiple Postures via EMG and Machine Learning.

Diagnosis of bruxism is challenging because not all contractions of the masticatory muscles can be classified as bruxism. Conventional methods for sleep bruxism detection vary in effectiveness. Some provide objective data through EMG, ECG, or EEG; others, such as dental implants, are less accessible for daily practice. These methods have targeted the masseter as the key muscle for bruxism detection. However, it is important to consider that the temporalis muscle is also active during bruxism among masticatory muscles. Moreover, studies have predominantly examined sleep bruxism in the supine position, but other anatomical positions are also associated with sleep. In this research, we have collected EMG data to detect the maximum voluntary contraction of the temporalis and masseter muscles in three primary anatomical positions associated with sleep, i.e., supine and left and right lateral recumbent positions. A total of 10 time domain features were extracted, and six machine learning classifiers were compared, with random forest outperforming others. The models achieved better accuracies in the detection of sleep bruxism with the temporalis muscle. An accuracy of 93.33% was specifically found for the left lateral recumbent position among the specified anatomical positions. These results indicate a promising direction of machine learning in clinical applications, facilitating enhanced diagnosis and management of sleep bruxism.

Semi-supervised learning for real-time anomaly detection in pulsed transfer wire arc additive manufacturing

In Wire-Arc Additive Manufacturing (WAAM), ensuring the quality and integrity of components is of key importance and performing anomaly detection during production is an efficient strategy for preventing defects and maintaining an acceptable quality with lower costs compared to the development of complex supervised learning techniques. This paper presents an approach based on semi-supervised learning for real-time anomaly detection in the production of Inconel 718 components via the Pulsed Transfer WAAM process. The workflow is based on the use of training data obtained by employing established process parameters, similar to what is done in Welding Procedure Qualification Records (WPQRs), to develop a semi-supervised anomaly detection application. The proposed approach involves depositing material using the already established process parameters and collecting the corresponding welding data in terms of welding current and voltage sensor signals. The collected data are then employed to train an unsupervised algorithm by using solely good deposition data to learn hidden complex patterns of normality and hence detect anomalies based on deviations from these patterns. With this aim, global and local features are extracted from the welding current and voltage sensor signals in both time and time-frequency domains via Wavelet Transform technique and a deep learning approach based on a residual convolutional autoencoder. To showcase the effectiveness of the proposed approach, a case study involving wire arc additive manufacturing of Inconel 718 components was conducted. The proposed algorithm was tested and achieved an F-score of 0.895, representing a 30 % improvement over state-of-the-art methods that rely on time domain features for anomaly detection. This research work contributes to the improvement of anomaly detection methodologies, offering substantial advantages over alternative techniques for quality control and reliability of additive manufacturing processes such as WAAM.

State of health estimation for lithium-ion batteries based on multi-scale frequency feature and time domain feature fusion method

Abstract Accurately estimating the state of health (SOH) of lithium-ion batteries is important for improving battery safety performance. The single time-domain feature extraction is hard to efficiently extract discriminative features from strongly nonlinear coupled data, leading to difficulties in accurately estimating the battery SOH. To this end, this paper proposes a multi-scale frequency domain feature and time domain feature fusion method for SOH estimation of lithium-ion batteries based on the Transformer model. Firstly, the voltage, current, temperature and time information of the battery are extracted as time domain features; secondly, the battery signal is processed by a multi-scale filter bank based on Mel-frequency cepstral coefficients (MFCC) to obtain the multi-scale frequency domain features; then, a parallel focusing network (PFN) is designed to fuse the time domain features with the frequency domain features, which yields low-coupling complementary discriminative features; finally, constructing the SOH estimation mechanism based on the Transformer deep network model. The algorithm is validated by NASA and Oxford datasets, the mean's absolute error (MAE) and root mean square error (RMSE) are as low as 0.06% and 0.23%, respectively.

A novel speaker verification approach featuring multidomain acoustics based on the weighted city block Minkowski distance

AbstractAccess control is vital in interconnected environments like the Internet of Things, Industry 4.0, and smart connectivity, ensuring authorized access for security. Biometric‐based access, particularly speaker verification (SV), enhances security with unique vocal features, offering nonintrusive authentication with continuous monitoring. Single‐domain features prove insufficient in distinguishing similar traits, prompting latest SV advancements to adopt multidomain‐based speech features. This paradigm addresses the limitations of single‐domain features by amalgamating the merits of individual domains, establishing a cutting‐edge approach. It utilizes cepstral–frequency–time domain feature fusion, achieved via cepstral mean‐variance normalization for generalizability. The weighted city block Minkowski distance is proposed to compare reference and test speech templates. Parameters are computed based on the confusion matrix, template matching distance functions, dynamic acoustic conditions, and additive white Gaussian noise. A deep convolutional neural network classifier is assessed on open‐source LibriSpeech and Speaker in the Wild corpora, surpassing the current methodologies.

A variable modal decomposition and BiGRU-based tool wear monitoring method based on non-dominated sorting genetic algorithm II

ABSTRACT In the metal cutting process, it is important to realize the effective monitoring of tool wear to ensure the quality of parts machining. A variable modal decomposition (VMD) and bi-directional gated recurrent neural network (BiGRU) based on non-dominated sorting genetic algorithm II (NSGA-II) is proposed for tool wear monitoring (TWM) problem. The method takes the envelope entropy and kurtosis of the decomposed signal as the objective function, and iteratively optimizes the number of decomposition layers k and the penalty factor α of the VMD by NSGA-II to obtain the optimal solution set. Meanwhile, an evaluation method of entropy weight- technique for order preference by similarity to ideal solution (EW-TOPSIS) is proposed to obtain the optimal parameter combinations in VMD. The VMD decomposed signals are screened by calculating the correlation coefficients, the time domain and frequency domain features of the screened signals are extracted, and the screened signals and features are mirrored. To realize the in-depth mining of the feature information of time series data, a BiGRU deep learning monitoring model is established. By comparing three classification methods, it has been proven that this method has achieved better results in monitoring tool wear state.

PaFESD: Patterns augmented by Features Epileptic Seizure Detection.

The detection of epileptic seizures can have a significant impact on the patients' quality of life and on their caregivers. In this paper we propose a method for detecting such seizures from electroencephalogram (EEG) data named Patterns augmented by Features Epileptic Seizure Detection (PaFESD). The main novelty of our proposal consists in a detection model that combines EEG signal features with pattern matching. After cleaning the signal and removing artifacts (as eye blinking or muscle movement noise), time-domain and frequency-domain features are extracted to filter out non-seizure regions of the EEG. Jointly, pattern matching based on Dynamic Time Warping (DTW) distance is also leveraged to identify the most discriminative patterns of the seizures, even under scarce training data. The proposed model is evaluated on all patients in the CHB-MIT database, and the results show that it is able to detect seizures with an average F1 score of 98.9%. Furthermore, our approach achieves a F1 score of 100% (no false alarms or missed true seizures) for 20 of the patients (out of 24). Additionally, we automatically detect the most seizure/non-seizure discriminative EEG channel so that a wearable with only two electrodes would suffice to warn patients of seizures.

Wearable gait recognition system based on time domain features of Bluetooth inertial sensor network

Abstract As an important part of medical monitoring, gait recognition has been used in lower limb rehabilitation diagnosis and gait assessment. To meet practical requirements, we designed a wearable gait recognition system based on time domain features of Bluetooth inertial sensor network, which can be used for real-time acquisition and evaluation of gait signals. In this research, we used 6 sensor units to integrate 10 kinds of gait information, and designed a gait monitoring system covering the main joints of lower limbs based on Bluetooth network. The gait monitoring system can collect the inertial sensing data of lower limbs at a maximum rate of 90 Hz through the 6 sensor units, and has good electrical performance. In daily use, each unit can work for more than 3 hours and maintain a received signal strength indicator (RSSI) of more than -90 dBm at a spatial distance of 10 meters. On the gait dataset collected in the experiment, we used sliding window to segment the time domain signals and extract 10 kinds of time domain features for classification algorithm. We achieved 97.1% average identification accuracy of 7 different gait patterns by support vector machine (SVM).

A Novel Unsupervised Machine Learning Approach to Assess Postural Dynamics in Euthymic Bipolar Disorder.

Bipolar disorder (BD) is a mood disorder with different phases alternating between euthymia, manic or hypomanic episodes, and depressive episodes. While motor abnormalities are commonly seen during depressive or manic episodes, not much attention has been paid to postural abnormalities during periods of euthymia and their association with illness burden. We collected 24-hour posture data in 32 euthymic participants diagnosed with BD using a shirt-based wearable. We extracted a set of nine time-domain features, and performed unsupervised participant clustering. We investigated the association between posture variables and 12 clinical characteristics of illness burden. Based on their postural dynamics during the daytime, evening, or nighttime, participants clustered in three clusters. Higher illness burden was associated with lower postural variability, in particular during daytime. Participants who exhibited a mostly upright sitting/standing posture during the night with frequent nighttime postural transitions had the highest number of lifetime depressive episodes. Euthymic participants with BD exhibit postural abnormalities that are associated with illness burden, especially with the number of depressive episodes. Our results contribute to understanding the role of illness burden on posture changes and sleep consolidation in periods of euthymia.

Wearable Surface Deformation Myography (sDMG) System for Recognition of Locomotion Modes.

This study designs a wearable sensing system for locomotion mode recognition using lower-limb skin surface curvature deformation caused by the morphological changes of musculotendinous complexes and soft tissues. Flexible bending sensors are embedded into stretch pants, enabling curvature deformations of specific skin segments above lower-limb muscle groups to be captured in a noncontact manner. To evaluate the performance of this system, we conducted experiments on eight able-bodied subjects completing seven common locomotive activities, including walking, running, ramp ascending/descending, stair ascending/descending, and standing. The system measured seven channels of deformation signals from two cross-sections on the shank and the thigh. The collected signals were distinguishable across different locomotion modes and exhibited consistency when monitoring steps. Using selected time-domain features and a linear discriminant analysis (LDA) classifier enabled the proposed system to continuously recognize locomotion modes with an average accuracy of 96.5%. Furthermore, the system maintains recognition performance with 95.7% accuracy even after removing and reapplying the sensors. Finally, we conducted comparison experiments to analyze how window length, feature selection, and the number of channels affect recognition performance, providing insights for optimization. We believe that this novel signal platform holds great potential as a valuable supplementary tool in wearable human motion detection, enriching the information diversity for motion analysis, and enabling new possibilities for further advancements and applications in fields including biomedical engineering, textiles, and computer graphics.

ZCHSound: Open-Source ZJU Paediatric Heart Sound Database With Congenital Heart Disease.

From 2020 to 2022, we collaborated with experienced cardiac surgeons from three general children's hospitals to collect heart sound signals from 1259 participants using electronic stethoscopes. To ensure the accuracy of the labels, the labels for all data were confirmed by two cardiac experts. To establish the baseline of ZCHsound, we extracted 84 features and used machine learning models to evaluate the performance of the classification task. The ZCHSound database was divided into two datasets: one is a high-quality, filtered clean heart sound dataset, and the other is a low-quality, noisy heart sound dataset. In the evaluation of the high-quality dataset, our random forest ensemble model achieved an F1 score of 90.3% in the classification task of normal and pathological heart sounds. This study has successfully established a large-scale, high-quality, rigorously standardized pediatric CHD sound database with precise disease diagnosis. This database not only provides important learning resources for clinical doctors in auscultation knowledge but also offers valuable data support for algorithm engineers in developing intelligent auscultation algorithms.

Modeling the Intent to Interact With VR Using Physiological Features.

Mixed-Reality (XR) technologies promise a user experience (UX) that rivals the interactive experience with the real-world. The key facilitators in the design of such a natural UX are that the interaction has zero lag and that users experience no excess mental load. This is difficult to achieve due to technical constraints such as motion-to-photon latency as well as false-positives during gesture-based interaction. In this paper, we explored the use of physiological features to model the user's intent to interact with a virtual reality (VR) environment. Accurate predictions about when users want to express an interaction intent could overcome the limitations of an interactive device that lags behind the intention of a user. We computed time-domain features from electroencephalography (EEG) and electromyography (EMG) recordings during a grab-and-drop task in VR and cross-validated a Linear Discriminant Analysis (LDA) for three different combinations of (1) EEG, (2) EMG and (3) EEG-EMG features. We found the classifiers to detect the presence of a pre-movement state from background idle activity reflecting the users' intent to interact with the virtual objects (EEG: 62 % ± 10 %, EMG: 72 % ± 9 %, EEG-EMG: 69 % ± 10 %) above simulated chance level. The features leveraged in our classification scheme have a low computational cost and are especially useful for fast decoding of users' mental states. Our work is a further step towards a useful classification of users' intent to interact, as a high temporal resolution and speed of detection is crucial. This facilitates natural experiences through zero-lag adaptive interfaces.

Research on degradation analysis and health condition assessment method of phase shifter

Abstract In high-power, high-reliability power supply systems, the switching operation of power devices is driven by phase shifters. Degradation or failure of the phase shifters leads to power device shoot-through and other failures, and reduces the power devices’ service lifetime. Aiming at the problem of phase shifter condition monitoring, a degradation model is proposed by analysing the degradation mechanism and aging process of the sensitive key components of the phase shifter, and then a health condition monitoring method with multi-dimensional features is presented in this paper. The multidimensional feature vectors in the time domain are extracted from the output voltage signals of different aging stages, and the time-domain feature separability of different health conditions is verified. Then the deep neural networks identification model based on deep learning technology is proposed to recognize the health condition of phase shifter. At last, the performance degradation simulation and experimental evaluation show that the proposed method can identify the phase shifter health conditions more accurately, and the accuracy rate can reach 95.892%.

An interpretable TFAFI-1DCNN-LSTM framework for UGW-based pre-stress identification of steel strands

Steel strands serve as the key load-bearing components of pre-stressed bridges, yet the identification of effective pre-stress for steel strands is a challenging task. In this study, a time and frequency adaptive fusion input-one dimensional convolutional neural network-long short-term memory (TFAFI-1DCNN-LSTM) framework was proposed for ultrasonic guided wave (UGW)-based effective pre-stress identification of steel strands. In this framework, the time and frequency domain signals of UGW were input into the network as two parallel branches, CNN and LSTM were utilized to extract spatial and serial-related features, and a trainable weight coefficient was introduced for adaptive fusion of time and frequency domain features. Firstly, ablation studies were conducted to investigate the influence of each key component in the proposed framework on the prediction results. Secondly, deep learning (DL) models of Res Net, Dense Net and Inception Net were introduced as comparisons, and the prediction results based on TFAFI-1DCNN-LSTM were compared with those based on the above DL models. Thirdly, the noise-robustness and performance under a few trainable samples of TFAFI-1DCNN-LSTM were also investigated. Finally, the outputs of various key layers of the proposed framework were subjected to dimensionality reduction and visualization to provide an intuitive demonstration of the feature extraction process, and the inherent mechanisms of the proposed framework were interpreted using local interpretable model-agnostic explanations (LIME). Results show that the architecture of TFAFI-1DCNN-LSTM is reasonably designed and all key components are necessary. The weights of features in time and frequency domain branches are rationally assigned due to the addition of an adaptive weight coefficient. Compared with other DL models, the proposed TFAFI-1DCNN-LSTM exhibits higher pre-stress identification accuracy, excellent noise-robustness, and superior performance under only a few trainable samples. The correlation between features extracted by the network and the pre-stress labels significantly increases with the deepening of the network, and the correlation between input UGW signals and predicted pre-stress values is effectively interpreted utilizing LIME, proving that the architecture of the proposed framework is reasonable and effective, the predicted results are reliable and credible.

High-resolution seismic inversion method based on joint data-driven in the time-frequency domain

Seismic inversion can be divided into time-domain inversion and frequency-domain inversion based on different transform domains. Time-domain inversion has stronger stability and noise resistance compared to frequency-domain inversion. Frequency domain inversion has stronger ability to identify small-scale bodies and higher inversion resolution. Therefore, the research on the joint inversion method in the time-frequency domain is of great significance for improving the inversion resolution, stability, and noise resistance. The introduction of prior information constraints can effectively reduce ambiguity in the inversion process. However, the existing model-driven time-frequency joint inversion assumes a specific prior distribution of the reservoir. These methods do not consider the original features of the data and are difficult to describe the relationship between time-domain features and frequency-domain features. Therefore, this paper proposes a high-resolution seismic inversion method based on joint data-driven in the time-frequency domain. The method is based on the impedance and reflectivity samples from logging, using joint dictionary learning to obtain adaptive feature information of the reservoir, and using sparse coefficients to capture the intrinsic relationship between impedance and reflectivity. The optimization result of the inversion is achieved through the regularization term of the joint dictionary sparse representation. We have finally achieved an inversion method that combines constraints on time-domain features and frequency features. By testing the model data and field data, the method has higher resolution in the inversion results and good noise resistance.

Study on the classification of sleep stages in EEG signals based on DoubleLinkSleepCLNet.

The classification of sleep stages based on Electroencephalogram (EEG) changes has significant implications for evaluating sleep quality and sleep status. Most polysomnography (PSG) systems have a limited number of channels and do not achieve optimal classification performance due to a paucity of raw data. To leverage the data characteristics and enhance the classification accuracy, we propose and evaluate a novel dual-link deep neural network model, 'DoubleLinkSleepCLNet'. The DoubleLinkSleepCLNet model performs feature extraction and efficient classification on both the raw EEG and the EEG processed with the Hilbert transform. It leverages the frequency domain and time domain feature modules, resulting in superior performance compared to other models. The DoubleLinkSleepCLNet model, using the 2 Raw/2 Hilbert data modes, achieved the highest classification performance with an accuracy of 88.47%. The average accuracy of the EEG was improved by approximately 4.08% after the application of the Hilbert transform. Additionally, Convolutional Neural Network (CNN) demonstrated superior performance in processing phase information, whereas Long Short-Term Memory (LSTM) excelled in handling time series data. The application of the Hilbert transform to EEG data, followed by processing it with a convolutional neural network, enhances the accuracy of the model. These findings introduce novel concepts for accelerating sleep stage prediction research, suggesting potential applications of these methods to other EEG analyses.