Signal-based Features Research Articles

Optimizing porosity detection in wire laser metal deposition processes through data-driven AI classification techniques

Additive manufacturing (AM) is an attractive solution for many companies that produce geometrically complex parts. This process consists of depositing material layer by layer following a sliced CAD geometry. It brings several benefits to manufacturing capabilities, such as design freedom, reduced material waste, and short-run customization. However, one of the current challenges faced by users of the process, mainly in wire laser metal deposition (wLMD), is to avoid defects in the manufactured part, especially the porosity. This defect is caused by extreme conditions and metallurgical transformations of the process. And not only does it directly affect the mechanical performance of the parts, especially the fatigue properties, but it also means an increase in costs due to the inspection tasks to which the manufactured parts must be subjected. This work compares three operational solution approaches, product-centric, based on signal-based feature extraction and Topological Data Analysis together with statistical and Machine Learning (ML) techniques, for the early detection and prediction of porosity failure in a wLMD process. The different forecasting and validation strategies demonstrate the variety of conclusions that can be drawn with different objectives in the analysis of the monitored data in AM problems.

Comparison of Signal Processing Methods for Bearing Fault Detection

Abstract: Bearings with rolling elements are a typical part of rotating machinery. Vibration analysis is one of them that is frequently used to gauge the general wellbeing and condition of rotating machinery. Finding bearing defects is important in the pursuit of extremely reliable operations. The main relationship between vibration signal-based features and measured signals is that they can be used in order to assess the condition of a bearing. In order to examine the effectiveness of non-parametric harmonic approaches for defect identification using real motor data, the study recommends utilizing Thomson's Multitaper Estimation and Welch Periodogram Estimation. The researches show that this approach can identify outer race faults and inner race fault. The conclusive evidence points to Thomson's Multitaper Periodogram Estimation method as a successful technique for bearing fault identification.

UroSound: A Smartwatch-Based Platform to Perform Non-Intrusive Sound-Based Uroflowmetry.

Leveraging consumer technology such as smartwatches to objectively and remotely assess people with voiding dysfunction could capture unique features for prompt diagnosis of a disease. This paper presents the UroSound, the first platform that performs non-intrusive sound-based uroflowmetry with a smartwatch. We study the feasibility of using a smartwatch to assess how well the urinary tract functions by processing the sound generated when the urine stream hits the water level in the toilet bowl, which can be modelled through the sound envelope. Signal-based features related to the sound envelope were extracted from a smartwatch's built-in microphone. The constructed model achieves a good correlation between acoustic and standard uroflowmetry in terms of the voiding shape and it can extract relevant voiding parameters. This indicates that accurate and remote measurement of the ambulatory characteristics of voiding dysfunction can be achieved with smartwatch-based uroflowmetry. UroSound also facilitates the collection of a voiding diary by measuring multiple uroflows during daytime and nighttime. Finally, the performance of 6 commercial smartwatches was analysed while recording a voiding event. The results demonstrate that the presence of an automatic gain control in the smartwatch microphone has a negative impact on the signal envelope, and should be avoided. Overall, this work demonstrates the potential for the use of smartwatches in the assessment of voiding dysfunction, to deliver more personalized and effective health care at home with less waste of time and resources, in particular in rural or less developed areas where access to a urology specialist is more difficult.

심전도 기반의 심장 질환 분류를 위한 신호 특징 추출 시스템 개발

In this paper, we propose a signal-based feature detection system for the early diagnosis of heart disease. The purpose of this study is to develop a compact healthcare system that extracts features from ECG signals. Therefore, the performance of the Tompkins algorithm, which is widely known as a signal-based feature detection algorithm, and the deep learning model of prior research are compared. In addition, by verifying the performance of the deep learning model by applying the Haar wavelet transform to the preprocessing, we determine the optimal feature detection model applicable to the healthcare system in terms of speed and accuracy. All algorithms and models were developed using Matlab on Window, and performance was compared on Jetson nano and Raspberry pi embedded boards. As a result, the best model in terms of speed and accuracy is the Haar wavelet bidirectional long short-term memory model, which quickly performs classification prediction with almost 97% accuracy. The advantage of this study is that it quickly performs classification prediction with high accuracy compared to existing healthcare systems that use algorithms that cause signal distortion. Therefore, cardiovascular diseases can be predicted and monitored based on the feature regions detected in this study. The data used are the MIT-BIH Arrhythmia Database and the Normal Sinus Rhythm Database. If additional patient data are established and verified in the future, it will be possible to use them not only in real life but also in clinical settings.

Signal-based feature extraction for Makhraj emission point classification

Due to the similar sound of one letter to the others, mistakes might happen when pronouncing a hijaiyah letter. The reciter will not read the Quran correctly if they do not understand the relationship between the hijaiyah letter sound and its point of articulation. This study addresses the issue to recognize the nine points of articulation (throat, uvular, molar, palatal, alveolar, dental, alveolar dental, lip, and interdental) from makhraj recitation using speech processing technique. As much as 181 non-distributive audio samples recorded in control environment. The input speech is a sukun combination of the Hijaiyah letter from an expert reciter. The research uses 5 type of signal-based feature extraction methods (MFCC, chroma, Mel spectrogram, spectral contract, and Tonnetz) and three type of classification methods (ANN, kNN, and SVM). The result shows the proposed method obtained a fair accuracy with the highest accuracy is 56% using ANN.

Human Silhouette and Skeleton Video Synthesis Through Wi-Fi Signals.

The increasing availability of wireless access points (APs) is leading toward human sensing applications based on Wi-Fi signals as support or alternative tools to the widespread visual sensors, where the signals enable to address well-known vision-related problems such as illumination changes or occlusions. Indeed, using image synthesis techniques to translate radio frequencies to the visible spectrum can become essential to obtain otherwise unavailable visual data. This domain-to-domain translation is feasible because both objects and people affect electromagnetic waves, causing radio and optical frequencies variations. In the literature, models capable of inferring radio-to-visual features mappings have gained momentum in the last few years since frequency changes can be observed in the radio domain through the channel state information (CSI) of Wi-Fi APs, enabling signal-based feature extraction, e.g. amplitude. On this account, this paper presents a novel two-branch generative neural network that effectively maps radio data into visual features, following a teacher-student design that exploits a cross-modality supervision strategy. The latter conditions signal-based features in the visual domain to completely replace visual data. Once trained, the proposed method synthesizes human silhouette and skeleton videos using exclusively Wi-Fi signals. The approach is evaluated on publicly available data, where it obtains remarkable results for both silhouette and skeleton videos generation, demonstrating the effectiveness of the proposed cross-modality supervision strategy.

Multilingual broad phoneme recognition and language-independent spoken term detection for low-resourced languages

Language-independent spoken term detection (LI-STD) refers to the process of locating the occurrences of spoken queries from speech databases of any language. This paper alization of a multilingual broad phoneme classifier (BPC) and its application for the development of an LI-STD system. This work proposes a multi-stage architecture to address the task of LI-STD for low-resourced languages, where there is limited amount of labelled training data. The proposed LI-STD system contains three stages; one label sequence matching stage and two template matching stages. A deep neural network (DNN) based BPC trained using 16 handcrafted, signal-based features is the backbone of the proposed LI-STD system. In LI-STD system, stage 1 performs a broad phoneme sequence matching, while stage 2 and 3 perform template matching on posteriorgram and feature sequence, respectively. Concatenation of multiple stages results in search space reduction for the later computationally intensive template matching stages. In order to adapt to a new/unseen language, the BPC gets retrained using selected broad phoneme labelled data of the language generated by itself. The effectiveness of the proposed system is demonstrated on a set of low-resourced Indian languages.

Open-Circuit Fault Diagnosis of Three-Phase PWM Rectifier Using Beetle Antennae Search Algorithm Optimized Deep Belief Network

Effective open-circuit fault diagnosis for a two-level three-phase pulse-width modulating (PWM) rectifier can reduce the failure rate and prevent unscheduled shutdown. Nevertheless, traditional signal-based feature extraction methods show poor distinguishability for insufficient fault features. Shallow learning diagnosis models are prone to fall into local extremum, slow convergence speed, and overfitting. In this paper, a novel fault diagnosis strategy based on modified ensemble empirical mode decomposition (MEEMD) and the beetle antennae search (BAS) algorithm optimized deep belief network (DBN) is proposed to cope with these problems. Initially, MEEMD is applied to extract useful fault features from each intrinsic mode function (IMF) component. Meanwhile, to remove features with redundancy and interference, fault features are selected by calculating the importance of each feature based on the extremely randomized trees (ERT) algorithm, and the dimension of fault feature vectors is reduced by principal component analysis. Additionally, the DBN stacked with two layers of a restricted Boltzmann machine (RBM) is selected as the classifier, and the BAS algorithm is used as the optimizer to determine the optimal number of units in the hidden layers of the DBN. The proposed method combined with feature extraction, feature selection, optimization, and fault classification algorithms significantly improves the diagnosis accuracy.

Smartphone- and Smartwatch-Based Remote Characterisation of Ambulation in Multiple Sclerosis During the Two-Minute Walk Test

Leveraging consumer technology such as smartphone and smartwatch devices to objectively assess people with multiple sclerosis (PwMS) remotely could capture unique aspects of disease progression. This study explores the feasibility of assessing PwMS and Healthy Control's (HC) physical function by characterising gait-related features, which can be modelled using machine learning (ML) techniques to correctly distinguish subgroups of PwMS from healthy controls. A total of 97 subjects (24HC subjects, 52 mildly disabled (PwMSmild, EDSS [0-3]) and 21 moderately disabled (PwMSmod, EDSS [3.5-5.5]) contributed data which was recorded from a Two-Minute Walk Test (2MWT) performed out-of-clinic and daily over a 24-week period. Signal-based features relating to movement were extracted from sensors in smartphone and smartwatch devices. A large number of features (n = 156) showed fair-to-strong (R 0.3) correlations with clinical outcomes. LASSO feature selection was applied to select and rank subsets of features used for dichotomous classification between subject groups, which were compared using Logistic Regression (LR), Support Vector Machines (SVM) and Random Forest (RF) models. Classifications of subject types were compared using data obtained from smartphone, smartwatch and the fusion of features from both devices. Models built on smartphone features alone achieved the highest classification performance, indicating that accurate and remote measurement of the ambulatory characteristics of HC and PwMS can be achieved with only one device. It was observed however that smartphone-based performance was affected by inconsistent placement location (running belt versus pocket). Results show that PwMSmod could be distinguished from HC subjects (Acc. 82.2 ± 2.9%, Sen. 80.1 ± 3.9%, Spec. 87.2 ± 4.2%, F 1 84.3 ± 3.8), and PwMSmild (Acc. 82.3 ± 1.9%, Sen. 71.6 ± 4.2%, Spec. 87.0 ± 3.2%, F 1 75.1 ± 2.2) using an SVM classifier with a Radial Basis Function (RBF). PwMSmild were shown to exhibit HC-like behaviour and were thus less distinguishable from HC (Acc. 66.4 ± 4.5%, Sen. 67.5 ± 5.7%, Spec. 60.3 ± 6.7%, F 1 58.6 ± 5.8). Finally, it was observed that subjects in this study demonstrated low intra- and high inter-subject variability which was representative of subject-specific gait characteristics.

Automatic damage type classification and severity quantification using signal based and nonlinear model based damage sensitive features

Structural health monitoring (SHM) is an emerging technology designed to automate the inspection process undertaken to assess the health condition of structures. The SHM process is classically decomposed into four sequential steps: damage detection, localization, classification, and quantification. This paper addresses damage type classification and severity quantification issues as classification problems whereby each class corresponds to a given damage type or a certain damage extent. A Support Vector Machine (SVM) is used to perform multi-class classification task. Classically, Signal Based Features (SBF) are used to train SVMs when approaching SHM from a machine learning perspective. In this work, starting from the assumption that damage causes a structure to exhibit nonlinear response, it is investigated whether the use of Nonlinear Model Based Features (NMBF) increases classification performance. NMBF are computed based on parallel Hammerstein models which are identified with an Exponential Sine Sweep (ESS) signal. A study of the sensitivity of classification performance to the noise contained in output signals is also conducted. Dimension reduction of features vector using Principal Component Analysis (PCA) is carried out in order to find out if it allows robustifying the classification/quantification process suggested in this work. Simulated data on a cantilever beam with various damage types and severities as well as experimental data coming from a composite aeronautic plate with various damage severities generated with a unique and original laser process are considered for demonstration. For both application cases, results show that by introducing NMBF, classification performance is improved. Furthermore, PCA allows for high recognition rates while reducing features vector dimension.

Accelerometer-based predictive models of fall risk in older women: a pilot study

Current clinical methods of screening older adults for fall risk have difficulties. We analyzed data on 67 women (mean age = 77.5 years) who participated in the Objective Physical Activity and Cardiovascular Health (OPACH) study within the Women’s Health Initiative and in an accelerometer calibration substudy. Participants completed the short physical performance battery (SPPB), questions about falls in the past year, and a timed 400-m walk while wearing a hip triaxial accelerometer (30 Hz). Women with SPPB ≤ 9 and 1+reported falls (n = 19) were grouped as high fall risk; women with SPPB = 10–12 and 0 reported falls (n = 48) were grouped as low fall risk. Random Forests were trained to classify women into these groups, based upon traditional measures of gait and/or signal-based features extracted from accelerometer data. Eleven models investigated combined feature effects on classification accuracy, using 10-fold cross-validation. The models had an average 73.7% accuracy, 81.1% precision, and 0.706 AUC. The best performing model including triaxial data, cross-correlations, and traditional measures of gait had 78.9% accuracy, 84.4% precision, and 0.846 AUC. Mediolateral signal-based measures—coefficient of variance, cross-correlation with anteroposterior accelerations, and mean acceleration—ranked as the top 3 features. The classification accuracy is promising, given research on probabilistic models of falls indicates accuracies ≥80% are challenging to achieve. The results suggest accelerometer-based measures captured during walking are potentially useful in screening older women for fall risk. We are applying algorithms developed in this paper on an OPACH dataset of 5000 women with a 1-year prospective falls log and week-long, free-living accelerometer data.

Early Hospital Mortality Prediction using Vital Signals.

Early hospital mortality prediction is critical as intensivists strive to make efficient medical decisions about the severely ill patients staying in intensive care units (ICUs). As a result, various methods have been developed to address this problem based on clinical records. However, some of the laboratory test results are time-consuming and need to be processed. In this paper, we propose a novel method to predict mortality using features extracted from the heart signals of patients within the first hour of ICU admission. In order to predict the risk, quantitative features have been computed based on the heart rate signals of ICU patients suffering cardiovascular diseases. Each signal is described in terms of 12 statistical and signal-based features. The extracted features are fed into eight classifiers: decision tree, linear discriminant, logistic regression, support vector machine (SVM), random forest, boosted trees, Gaussian SVM, and K-nearest neighborhood (K-NN). To derive insight into the performance of the proposed method, several experiments have been conducted using the well-known clinical dataset named Medical Information Mart for Intensive Care III (MIMIC-III). The experimental results demonstrate the capability of the proposed method in terms of precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). The decision tree classifier satisfies both accuracy and interpretability better than the other classifiers, producing an F1-score and AUC equal to 0.91 and 0.93, respectively. It indicates that heart rate signals can be used for predicting mortality in patients in the care units especially coronary care units (CCUs), achieving a comparable performance with existing predictions that rely on high dimensional features from clinical records which need to be processed and may contain missing information.

Automatic Damage Quantification Using Signal Based And Nonlinear Model Based Damage Sensitive Features

Structural Health Monitoring (SHM) can be defined as the process of acquiring and analyzing data from on-board sensors to evaluate the health of a structure. Classically, an SHM process can be performed in four steps: detection, localization, classification and quantification. This paper addresses damage quantification issue as a classification problem whereby each class corresponds to a certain damage extent. Starting from the assumption that damage causes a structure to exhibit nonlinear response, we investigate whether the use of nonlinear model based features increases classification performance. A support Vector Machine (SVM) is used to perform multi-class classification task. Two types of features are used as inputs to the SVM algorithm: Signal Based Features (SBF) and Nonlinear Model Based Features (NMBF). SBF are rooted in a direct use of response signals and do not consider any underlying model of the test structure. NMBF are computed based on parallel Hammerstein models which are identified with an Exponential Sine Sweep (ESS) signal. A study of the sensitivity of classification performance to the noise contained in output signals is also conducted. Dimension reduction of features vector using Principal Component Analysis (PCA) is carried out in order to find out if it allows robustifying the quantification process suggested in this work. Simulation results on a cantilever beam with a bilinear torsion spring stiffness are considered for demonstration. Results show that by introducing NMBF, classification performance is improved. Furthermore, PCA allows for higher recognition rates while reducing features vector dimension. However, classifiers trained on NMBF or on principal components appear to be more sensitive to output noise than those trained on SBF.

An Enhanced Power Spectral Density Transmissibility (EPSDT) approach for operational modal analysis: Theoretical and experimental investigation

The transmissibility based operational modal analysis (TOMA) has drawn great attention due to its insensitivity to the nature of external excitation. Different from conventional modal identification approaches working in the frequency domain, it opens up a new path to find the system poles by utilizing mode shape information. Recently, the power spectral density transmissibility (PSDT) driven peak picking (PP) approach was proposed by authors for large-scale linear engineering structures under a single operational loading condition. As a nonparametric method, however, the PSDT-driven PP approach identifies natural frequencies by just observing signal-based features subjectively without fitting a parametric model. An enhanced PSDT approach is proposed in this study. The system poles are identified by employing the least-squares complex frequency-domain (LSCF) estimator, following which the mode shapes are evaluated by utilizing singular value decomposition (SVD) technique for PSDT matrix. As a result, a stabilization diagram containing the information of frequency, damping ratio and mode shape can be constructed to separate the stable system poles. Furthermore, the asymptotic analysis is implemented to investigate the robustness of PSDT to noise effect. The presented approach is able to maintain the benefits of PSDT and at the same time leads to an improvement of the modal parameter estimation. It is computationally efficient, simple, stable and robust to noise. A numerical example subjected to ground motion and two structures tested in the field are employed to demonstrate the efficiency and accuracy of the proposed method.

Anomaly Detection Framework Based on Matching Pursuit for Network Security Enhancement

In this paper, a framework for recognizing network traffic in order to detect anomalies is proposed. We propose to combine and correlate parameters from different layers in order to detect 0-day attacks and reduce false positives. Moreover, we propose to combine statistical and signal-based features. The major contribution of this paper are: novel framework for network security based on the correlation approach as well as new signal based algorithm for intrusion detection using matching pursuit.

Improving Music Genre Classification Using Automatically Induced Harmony Rules

We present a new genre classification framework using both low-level signal-based features and high-level harmony features. A state-of-the-art statistical genre classifier based on timbral features is extended using a first-order random forest containing for each genre rules derived from harmony or chord sequences. This random forest has been automatically induced, using the first-order logic induction algorithm TILDE, from a dataset, in which for each chord the degree and chord category are identified, and covering classical, jazz and pop genre classes. The audio descriptor-based genre classifier contains 206 features, covering spectral, temporal, energy, and pitch characteristics of the audio signal. The fusion of the harmony-based classifier with the extracted feature vectors is tested on three-genre subsets of the GTZAN and ISMIR04 datasets, which contain 300 and 448 recordings, respectively. Machine learning classifiers were tested using 5 × 5-fold cross-validation and feature selection. Results indicate that the proposed harmony-based rules combined with the timbral descriptor-based genre classification system lead to improved genre classification rates.