Search

The roller mill orientation instrument (RMOI) integrates the processing and the orientation of single-crystal bars. This type of equipment realizes the integration of two processes: grinding of the single-crystal ingot and orientation of the single-crystal bar. This integration enhances production efficiency and the precision of the orientation. However, this type of approach has two significant constraints: 1) there is no quality detection algorithm for the plane of the crystal bar; it is, therefore, currently necessary to randomly find a crystal plane for orientation, which means that there is no guarantee of the quality of the plane after cutting and 2) there is a lack of evaluation methods to determine the overall quality of the crystal bar. This affects the production process and the requirements for the quality of the crystal bar and further affects the popularization and application of the equipment. Here, to overcome these constraints, we first propose a deep learning 1-D convolutional neural network (1-D-CNN) algorithm to perform feature extraction and supervised classification based on the rocking curve of the single crystal. In sharp contrast to the shallow learning method support vector machine (SVM), SVM feature extraction is insufficient, and the accuracy is low. The average accuracy of 1-D-CNN is 92.00%. This realizes the quality requirements for the successful detection of the crystal plane. Next, using the obtained quality detection results of each crystal plane, a hybrid algorithm combining improved Canopy (ICanopy) and ${K}$ -means algorithms is presented, which improved the average accuracy by 10% compared with the traditional Canopy- ${K}$ -means algorithm, and the ${k}$ -nearest neighbor (${k}$ NN) algorithm is further utilized. Finally, an ICanopy- ${K}$ -means- ${k}$ NN algorithm is formed, which realizes the overall quality evaluation of the crystal bar under various situations, and the average accuracy increases by 3.33%–90%. The effectiveness of the proposed algorithms is demonstrated by the analysis of results obtained from practical data. [ABSTRACT FROM AUTHOR]

Using transfer learning (TL) for fault detection and diagnosis has been a hot topic in prognostic and health management (PHM) field. In this article, a systematic framework is established to solve the cross-domain fault diagnosis for rotary machines of the same type using entropy-based TL. The multiscale symbolic dynamic entropy (MSDE) is first proposed to extract features from vibration signals. Then, TL model is trained to get a mapping matrix, which can preserve the structure properties of prior distribution and minimize the discrepancies between different datasets. The mapped features are called multiscale transfer symbolic dynamic entropy (MTSDE). Finally, the support vector machine (SVM) classifier is utilized to accomplish the cross-domain intelligent fault-type recognition. To demonstrate the superiority of the proposed MTSDE method, the comparative experiments of nine methods are carried out under different datasets including gearboxes and bearings. Experimental results demonstrate that our proposed MTSDE method performs best in recognizing various fault types comparing with other nine methods. To the best knowledge of the authors, this is the first attempt of using entropy-based TL for cross-domain fault diagnosis of rotating machinery. [ABSTRACT FROM AUTHOR]

The rolling bearing is the key component of rotating machinery, and it is also a failure–prone component. The intelligent fault diagnosis method has been widely used to accurately diagnose bearing faults. However, in engineering practice, it is difficult to obtain sufficient sample data to train the intelligent diagnosis model. Therefore, in this article, a fusion diagnosis model CGAN-2-D-CNN that combines a conditional generative adversarial network (CGAN) and a two-dimensional convolutional neural network (2-D-CNN) is proposed for bearing fault diagnosis with small samples. Considering the problem of insufficient sample data, CGAN is used to learn the data distribution of real samples to generate new samples with similar data distribution by the confrontation training of the generator and discriminator. Then, 2-D preprocessing is conducted to convert the generated 1-D data into 2-D gray images. Finally, these images are input into the 2-D-CNN to extract the features and classify the bearing fault types. Two experimental cases are implemented to validate the effectiveness and feasibility of the proposed CGAN-2-D-CNN. The experimental results illustrate that the diagnosis accuracy of the proposed method used on the small sample data is close to that of the 2-D-CNN directly used on the enough original sample data whose size is equal to the expanded sample size. In addition, compared with the 1-D-CNN, support vector machine (SVM), and long short-term memory (LSTM) models, the 2-D-CNN model after 2-D preprocessing has the higher fault classification accuracy. [ABSTRACT FROM AUTHOR]

Automated and accurate detection of myocardial infarction (MI) is a momentous task because of its association with damage of heart muscles and sudden cardiac arrest. This work highlights the investigation of discriminative feature representations to distinguish MI episodes from normal sinus rhythms by combining the strength of boosted classification approach. Proposed method analyzes the spectral coherency of 12 lead electrocardiogram (ECG) signal to explore the discrimination of normal and MI data in cross spectral domain. Coherence spectrum reveals the distinguishing characteristics of the pathologically varying ECG signals which in turn identify and localize the MI episode. In this view, two new feature metrics namely spectral coherence indices (SCIs) and phase coherence indices (PCIs) are introduced to capture the significant coherence variations. The effectiveness of the extracted features is determined by the Fisher score (F-score) estimation. Finally, a boosting framework using the ensemble of support vector machine (SVM) classifiers is presented to improve the performances of the proposed detection model. The boosted SVM augments the boosting technique by introducing a new cluster-based training sample set selection algorithm. The proposed detection model is empirically evaluated using the open access Physikalisch-Technische Bundesanstalt (PTB) diagnostic ECG database. The accuracy of 98.96% and 98.85% is obtained for detecting and identifying the different types of MI pathology, respectively. Comparative study with the existing work exhibits encouraging performances with reduced feature dimension. [ABSTRACT FROM AUTHOR]

Hyperspectral images (HSIs) contain many levels of spatial structures, and thus, feature extraction techniques have been broadly studied in hyperspectral data processing to mine its structural information while preserving strong edge details. However, spatial characteristics in different scales unavoidably mix and overlap, which makes separation among pixels very difficult and yields misclassification. This article proposes a feature extraction method by using multitask superpixel-based auxiliary learning (MSAL) that consists of two main stages. We first design a novel multitask learning (MTL) model from the perspective of task input by constructing a set of auxiliary feature cubes as inputs. Under the MTL framework, the spatial information and edge information are refined in the superpixel-based auxiliary learning feature extraction stage. Specifically, the auxiliary rolling guidance features are extracted from auxiliary cubes via rolling guidance filter, which can remove the unwanted structure. Then, the spatial feature within superpixels is refined by applying structure density-based weighted operation, and the edge feature is stacked with the superpixel auxiliary features to supply the edge information. Finally, the obtained feature cubes are fed into a pixelwise classifier, and a weighted decision rule is introduced to optimize the initial results by utilizing the inherent relation between auxiliary cubes. Experimental results demonstrate that the MSAL can significantly improve the quality of the extraction feature and outperform other state-of-the-art approaches in terms of classification accuracies. [ABSTRACT FROM AUTHOR]

Sign languages (SLs) can aid in improving the communication between hearing impaired and hearing communities, but only a limited number of hearing individuals understand it. SL recognition systems tackle this issue by using sensors to acquire gesture data later converted into spoken or written language by means of machine-learning algorithms. Within this frame, we propose a wearable electronic-based system for achieving automatic signer-independent recognition of Italian SL. The system is used for recognizing ten signs performed by 17 inexperienced signers. The wearable electronics measure hand, arm, and forearm gestures, by means of a sensory glove (equipped with ten flex sensors) and six inertial measurement units (IMUs). We adopt three classifiers: artificial neural network (ANN), $k$ -nearest neighbors, and support vector machine, with ANN outperforming the others with an accuracy of 95.07%. However, the real key and novel aspect of this work concerns exploring the importance of each sensor for the overall classification accuracy, with the aim of determining the minimum number of sensors (and so minimizing the complexity of the wearable system) necessary to guarantee an acceptable accuracy. As a result, we could drop out three of the six IMUs, obtaining an accuracy as high as 93.91%. In addition, this valuable result was obtained using a signer-independent approach, which makes the application closer to a real-life situation. [ABSTRACT FROM AUTHOR]

Gait phase analysis is widely used in disease diagnosis, rehabilitation training, and other fields by studying the characteristics of human gait. It systematically evaluates human body’s skeletal muscles and nerves with combined disciplines. Microelectromechanical system (MEMS) inertial sensors are extensively used in attitude detection because of its high-precision, portability, and good real-time performance in time-domain analysis. In this article, we presented a multi-degree-of-freedom MEMS gait detection method, which resolved the problems of single sensor and limited gait phase. We designed a sensor-based gait signal acquisition system, in which gait data acquisition program and feature analysis algorithm were compiled to verify the feasibility of the proposed method. We performed coordinate transformation and corrected position information to eliminate the gait phase detection error caused by random noise interference. Acceleration and angular velocity information were collected from 20 experimenters. We applied an adaptive threshold gait phase detection algorithm to classify the gait information collected by single sensor. To improve the results of gait phase classification, we used multisensor redundant measurement to analyze characteristics of five gait phases. The acceleration and angular velocity information collected by the three sensors placed at instep, ankle, and thigh were input into support vector machine (SVM). The classification results of the five gait phases are approximately 90%. Lastly, we built a human body structure model to simulate human motion in real time, realizing the real-time gait phase detection, which proves effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Online continuous measurement of the mass flow rate of pneumatically conveyed solids is desirable in the monitoring and optimization of a range of industrial processes, such as food processing, chemical engineering, and power generation. This article introduces a technique for the mass flow rate measurement of pneumatically conveyed solids based on multimodal sensing and data-driven modeling. The multimodal sensing system is comprised of an array of ring-shaped electrostatic sensors, four arrays of arc-shaped electrostatic sensors, and a differential-pressure (DP) transducer. Data-driven models, including artificial neural network (ANN), support vector machine (SVM), and convolutional neural network (CNN), are established through training with statistical features extracted from the postprocessed data from the sensing system. Statistical features are shortlisted based on their importance by calculating the partial mutual information (PMI) between the features and the corresponding reference mass flow rate of solids. Experimental work was conducted on a laboratory-scale rig to train and test the models on both horizontal and vertical pipelines with particle velocity ranging from 10.1 to 36.0 m/s and mass flow rate of solids from 3.2 to 35.8 g/s. Experimental results suggest that the ANN, SVM, and CNN models predict the mass flow rate of solids with a relative error within ±18%, ±14%, and ±8%, respectively, under all test conditions. The predicted mass flow rate measurements with the ANN, SVM, and CNN models are repeatable with a normalized standard deviation within 14%, 8%, and 5%, respectively, under all test conditions. [ABSTRACT FROM AUTHOR]

This article presents a highly secure and convenient biometric system for user recognition based on body channel characteristics for electric signal transmission. In the proposed framework, the user can provide reliable biometric features of body channel responses (BCRs) simply by touching an electrode surface on the device with a finger. To realize and verify the proposed approach, we acquired the BCR data from 15 subjects for approximately six weeks through experiments conducted in a customized measurement setup suited to the principle of signal transmission in the human body channel. The proposed BCR-based biometric feature (BBF) comprises a series of envelope vectors of the received BCR when applying up- and down-chirp signals to the human body, which is extracted by an interpolative method based on peak detection. The BBFs are effectively separable according to the subjects because the features magnify quantitative differences in the aspect of path losses and power delay profiles of individual BCRs for the frequency range between 1 and MHz. The classification performance was evaluated by splitting the dataset into ${80\%}$ and 20% for the training and testing, respectively, using conventional machine learning algorithms with uncorrelated 40 session datasets for the respective subjects. The highest average classification accuracy was achieved by the kernel-based support vector machine approximately 95.8% without observable and biased misidentification cases among the subjects. In addition, the analysis of receiver operating characteristic curves shows that the proposed classifiers are robust to decision boundaries at various threshold settings. [ABSTRACT FROM AUTHOR]

Lung cancer is the most common and lethal cancer in many parts of the world. The establishment of lung cancer screening by low-dose computerized tomography (CT) scan has led to finding lung cancers in early stages as very small nodules. However, finding those nodules, particularly when located deep in the lung parenchyma, could be impossible during lung surgeries without preoperative or intraoperative localization. This study introduces a simple and safe method having the potential to localize in-depth pulmonary nodules intraoperatively. In this regard, a bioimpedance probe with four spherical electrodes was designed and built. By an in vitro study, the bioimpedance data of 286 lung tissue samples obtained from 38 patients in a frequency range of 50 kHz–5 MHz were collected and analyzed with Nyquist curves and boxplot charts. Finally, a smart system was designed based on the bioimpedance phase and magnitude to differentiate healthy lung tissue from the tumoral lung tissue. Our proposed system consists of two parts: the feature reduction with principal component analysis (PCA) and the classification with support vector machine (SVM), linear discriminant analysis (LDA), and K-nearest neighbors (KNN). Classifier analysis showed that the accuracy of all classifiers was more than 95% for 15 principal components; the SVM classifiers had the highest accuracy above 98%. This research sheds light on the feasibility of designing a real-time, safe, and smart system to localize the invisible/impalpable pulmonary nodules by the bioimpedance spectrum of the lung tissue. [ABSTRACT FROM AUTHOR]

Partial discharge (PD) detection and localization is one of the most effective methods in gas-insulated transmission lines (GIL) insulation fault diagnosis, which is important to the early troubleshooting and safe operation. Compared with the widely used ultrahigh frequency (UHF) and acoustic PD localization methods, the PD localization method based on optical signals has good resistance to electromagnetic and acoustic interference. However, the current optical localization method comes with several shortcomings: narrow detection range, requiring field experiments to accumulate data, and installing a large number of sensors, which renders low feasibility. Therefore, this article proposes a PD localization method utilizing virtual sensors (VSs) and optical simulation fingerprint. Based on the idea of the digital twin, this method constructs a fingerprint database through optical PD simulation in an equal-sized GIL simulation model, which solves the difficulty of fingerprint collection in actual equipment. Meanwhile, this article predicts the detection value of the VS through the adaptive neuro-fuzzy inference system (ANFIS) to greatly reduce the installation of the actual sensor (AS). Finally, through the support vector machines (SVMs) algorithm, the detection fingerprint that includes the virtual and actual detection values matches with the fingerprint database. The location corresponding to the optimal matching result is recorded as the localization result. As experiments have shown, the average localization error of the proposed method is 19.69 mm, which is 54.8% lower than the one without VSs under the same conditions. The results reveal that this method compensates for the loss of localization accuracy caused by the reduction of ASs, which embodies value in practice. [ABSTRACT FROM AUTHOR]

Effective feature extraction is crucial for accurate fault diagnosis of rolling bearings. A novel feature extraction method called hierarchical dispersion entropy (HDE) based on hierarchical analysis is proposed in this study. The proposed method includes the following three steps: 1) bearing vibration signal is decomposed into a series of subband signal components; 2) dispersion entropies of the components in different frequency bands are calculated as the original feature vector; and 3) joint approximate diagonalization of eigenmatrices (JADE) is used to extract fusion features from the original features. The main contributions of the proposed method are as follows: 1) the HDE method can characterize the complexity and uncertainty of the signal in full frequency band; 2) the JADE method further eliminates redundant information while greatly retaining fault-relevant information; and 3) the proposed method combines the advantages of HDE’s hierarchical analysis and JADE’s information fusion capabilities, so the fusion features extracted by the proposed method can be more effective for the establishment of fault pattern identification model. In the analysis of two experimental cases, the feature extracted by the proposed method shows better feature clustering effect and higher recognition rate than other methods. The results show that compared with other methods, the proposed method can more accurately characterize the health condition of the bearing. [ABSTRACT FROM AUTHOR]

Accurate flow regimes identification is of great significance for the deep understanding of the flow structure and ensuring the safe and economic operation of actual production in the multiphase flow process. For the flow regimes identification of the gas–liquid two-phase flow in the horizontal pipe, a multidomain features processing scheme is proposed. The water holdup data are acquired by conductance rings sensor, and processed by the improved empirical wavelet transform (EWT) to obtain the Fourier spectrum in the frequency domain and decomposition components in the multiscale domain. The obtained multidomain feature information is quantified to get the high-dimensional feature vectors of different flow regimes. To make the features more representative and realize the visualization of the structural relationship between the flow regime samples, the isometric feature mapping (Isomap) method in manifold learning is used to synthesize the feature structure relationships and map them to low-dimensional space. The obtained low-dimensional feature vectors of training samples are input into the support vector machine (SVM) to acquire the flow regimes identification model. Compared with other flow regimes identification schemes that feature reduction uses different manifold learning methods, or feature extraction is not in multidomain, or data preprocessing uses traditional wavelet transform, the proposed scheme can better retain the global structural relationship of the flow regimes, has the characteristics of strong stability, and the identification rate of the gas–liquid two-phase flow regimes in the horizontal pipe is over 95%. [ABSTRACT FROM AUTHOR]

Remaining useful life (RUL) prediction has been receiving increasing attention in industry and academia due to its advantages in ensuring safety, reducing costs, and improving efficiency. For industrial manufacturing, the ball screw is an indispensable high-precision servo component. There is an urgent need to predict its RUL to guarantee machining precision. However, establishing a reliable and accurate physical model is a thorny issue as sophisticated mechanical characteristics are difficult to obtain. Besides, the limited degradation data reduces the prediction accuracy of data-driven methods. This article presents a hybrid RUL prediction architecture of ball screws. First, to track the precision degradation process, the backlash is proposed as a novel nonlinear precision indicator. Then, physical-and data-driven methods are strategically leveraged to reveal the precision degradation process of the ball screw, where the stochastic degradation model is constructed and its initial parameters are estimated by data-driven methods. The effectiveness of the method is experimentally demonstrated through the designed platform and compared with commonly used methods. [ABSTRACT FROM AUTHOR]

Electrocardiogram (ECG) has attracted intense research interests and contests due to its great value in biometric applications, especially with the technological progress of smart instrumentation and artificial intelligence nowadays. For the issue of ECG-based human identification, distance measure in a suitable data space can be an effective solution. Almost all conventional distance measures rely on the independent data samples for classification. However, such measuring mechanism is vulnerable to the variation and bias of data distribution, which are easily caused by the noisy artifacts in this issue. To tackle this problem, we suggest doing distance measurement at the level of multiple-set bundle that consists of multiple sample sets obtained under different conditions. More specifically, we propose a novel method, “Collaborative-Set Measurement (CSM),” which creatively extends distance measure from the sample level through the set level to the bundle level. CSM not only captures the representative information from within-set distribution but also exploits the discriminative information from between-set collaboration, and meanwhile incorporates the robust information from in-bundle fusion. Attributing to these information, this method can largely enhance the identification performance in this issue despite the serious noisy artifacts influencing the data samples. The experimental results have demonstrated the capability of CSM for ECG-based human identification, by means of the public challenging database, DREAMER, on which the proposed method produces an accuracy of 91.30%. In summary, this proposal has initially brought forward a new thought of “collaborative sets.” This thought may provide inspirations for further researches on biometric recognition and general signal classification topics. [ABSTRACT FROM AUTHOR]

Quantifying phenolic compound in peated barley malt and discriminating its origin are essential to maintain the aroma of high-quality Scottish whisky during the manufacturing process. The content of the total phenol varies in peated barley malts, which is critical in measuring the associated peatiness level. Existing methods for measuring such phenols are destructive and/or time-consuming. To tackle these issues, we propose in this article a novel nondestructive system for fast and effective estimating the phenolic concentrations and discriminating their origins with the near-infrared hyperspectral imagery and machine learning. First, novel ways of data acquisition and normalization are developed for robustness. Then, the principal component analysis (PCA) and folded PCA are fused for extracting the global and local spectral features, followed by the support vector machine (SVM)-based origin discrimination and deep neural network-based phenolic measurement. In total, 27 categories of peated barley malts from eight suppliers are utilized to form thousands of spectral samples for modeling. A classification accuracy up to 99.5% and a squared correlation coefficient up to 98.57% are achieved, outperforming a few state of the art. These have fully demonstrated the efficacy of our system in automated phenolic measurement and origin discrimination to benefit the quality monitoring in the whisky industry. [ABSTRACT FROM AUTHOR]

Series arc fault is one of the important causes of electrical fire in industrial and mining enterprises. It is of great significance to study the series arc fault detection and phase selection feature extraction method to ensure safe and stable operation of electrical equipment and to guide line maintenance. Arc fault experiments under different current and circuit conditions with a three-phase motor and inverter load were carried out. A new arc fault detection and phase selection method based on single-phase current was proposed. First, wavelet threshold noise reduction, piecewise linear fitting, and first-order difference processing were performed on single-phase current signals to filter out noise interference and highlight fault features. Second, fractional Fourier transform (FRFT) was applied to the first-order differential signal to construct the amplitude matrix of the signal from the time domain to the frequency domain. The local features of the amplitude matrix were effectively extracted, and the feature vector of arc fault with lower dimension was established by combining the two-level block singular value decomposition (SVD) method. Finally, an arc fault detection and phase selection model was established using a support vector machine (SVM) optimized by grid search (GS) and particle swarm optimization (PSO) algorithm. The applicability of the model in single-phase multiload was analyzed. The results showed that the proposed method could realize series arc fault detection and phase selection in three-phase motor and inverter circuits, and it can also be used to single-phase multiload circuits. [ABSTRACT FROM AUTHOR]

Batch processes are important manufacturing approaches widely used in modern industry. During the manufacturing process, quality prediction is essential. Data-driven-based soft sensors have widely used for quality prediction because of the advantages of simple application and high flexibility. However, they do not perform well due to the characteristics of dynamic, nonlinear, and multiphases in batch processes. To address these issues, a soft sensor based on sequential kernel fuzzy partitioning (SKFP) and just-in-time relevance vector machine (JRVM), named SKFP-JRVM, is proposed in this work. First, an SKFP algorithm is proposed to divide batches into phases. In the SKFP algorithm, the inner and the sequential membership are constructed and compared to obtain highly accurate phase partitioning results. Meanwhile, a partitioning evaluation index is applied to automatically determine the optimal number of phases. Moreover, a soft sensor model based on the JRVM is built for each phase. The JRVM model not only considers the modeling samples of adjacent phases but also effectively addresses the problem of soft sensor modeling of process data with highly nonlinear and dynamic characteristics. The effectiveness of the SKFP-JRVM method is demonstrated on a numerical simulation and a penicillin fermentation process. [ABSTRACT FROM AUTHOR]

For the study of various retinal diseases, an accurate quantitative analysis of the retinal layer is essential for assessing the severity of the disease and diagnosing the progression of the disease. Optical coherence tomography (OCT) images can clearly show each layer of the retinal structure and detect subtle early lesions, thus providing a gold standard for the diagnosis of retinal diseases. In this article, we propose a coarse-to-fine retinal layer boundary segmentation method based on the embedded residual recurrent network (ERR-Net) and the graph search (GS). Considering the integrity of information transmission and the dependence of features, we first design a novel end-to-end residual recurrent network to roughly segment the retinal layer boundaries. The proposed ERR-Net not only solves the gradient problem brought by the depth but also fully captures the global spatial structure of the image. Second, we employ a GS algorithm for boundary continuous to make the layer boundary segmentation results more accurate. Finally, we evaluate the effectiveness of the proposed method on three publicly available datasets and compare it with the state-of-the-art methods on each dataset. The quantitative results and visual effects show that the proposed method outperforms the state-of-the-art approaches in the segmentation of retinal layer boundaries in OCT images. Moreover, the results also show that the proposed method has good algorithm stability on three datasets with different sizes, characteristics, and segmentation difficulties. [ABSTRACT FROM AUTHOR]

The idea of track-side acoustic detection technology is to extract fault-related information from the sound signal emitted by train bearings collected by microphones installed on the sides of the railway. The signal distortion caused by the Doppler effect is a barrier to efficient fault diagnosis. Currently, signal correction is the main way to solve this problem. Alternatively, this study attempts to directly construct the functional relationship between the Doppler-shifted signal and the diagnosis decision. Specifically, a two-stage parameter-driven learning model named kinematic-parameter-driven safety region model (KPD-SRM)/kinematic-parameter-driven backpropagation neural network (KPD-BPNN) is proposed, which provides a novel way for track-side acoustic fault diagnosis and it has the following merits. First, this is a breakthrough to the existing methods based on signal correction, the diagnosis decision does not require signal correction as a prerequisite. Second, with the employment of machine learning methods, historical data can be used to improve the diagnostic accuracy and it will be continuously improved along with the increase in monitoring samples. Finally, the proposed two-stage learning model can solve the problem of sample imbalance, so it has a good prospect of practical engineering application. Both simulation and experimental analysis prove the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Quick and accurate ellipse detection is essential for computer vision, automatic manufacturing, and other fields. In order to obtain more accurate ellipse parameters, the least-squares support vector regression is employed to establish the objective function and further convert the ellipse fitting problem into an optimization problem. This article proposes a hybrid BFGS algorithm integrated with an alternating direction method of multipliers (ADMM) to reduce the number of iterations and improve the algorithm convergence speed. The effectiveness and accuracy of the proposed algorithm are then verified in a series of numerical simulations by using the actual data of crystal images taken by charge-coupled device (CCD) cameras within the Czochralski silicon growth. Finally, the effectiveness is verified by comparing Hough transform (HT), maximum correntropy criterion (MCC), sparsity-based method (SBM), and other methods. [ABSTRACT FROM AUTHOR]

Schizophrenia (SZ) is a psychiatric disorder characterized by cognitive dysfunctions, hallucinations, and delusions, which may lead to lifetime disability. Detection and diagnosis of SZ by visual inspection is subjective, difficult, and time-consuming. Electroencephalogram (EEG) signals are widely used to detect the SZ as they reflect the conditions of the brain. Conventional machine learning methods involve many lengthy manual steps, such as decomposition, feature extraction, feature selection, and classification. In this article, automated identification of SZ is proposed using a combination of time–frequency analysis and convolutional neural network (CNN) to overcome the limitations of feature extraction-based methods. Three press button tasks are analyzed to segregate normal subjects and SZ patients. The EEG signals are subjected to continuous wavelet transform, short-time Fourier transform, and smoothed pseudo-Wigner–Ville distribution (SPWVD) techniques to obtain scalogram, spectrogram, and SPWVD-based time–frequency representation (TFR) plots, respectively. These 2-D plots are fed to pretrained AlexNet, VGG16, ResNet50, and CNN. We have obtained an accuracy of 93.36% using the SPWVD-based TFR and CNN model. In comparison to the benchmark AlexNet, ResNet50, and VGG16 networks, the developed CNN model with four convolutional layers not only requires fewer learnable parameters but also is computationally efficient and fast. This clearly indicates that our proposed method combining the SPWVD-CNN model has performed better than the state-of-the-art transfer learning techniques. Our developed model is ready to be tested with more EEG data and can aid psychiatrists in their diagnosis. [ABSTRACT FROM AUTHOR]

In the actual scenario, it is difficult for an automatic diagnosis system to detect each heartbeat when facing new patients. In this article, a novel active dual-scale residual convolutional long short-term memory neural network is proposed to classify heartbeats interactively and automatically. First, each heartbeat and electrocardiogram segment including current heartbeat are combined to build dual-scale input information. In addition, convolution layers, residual block, and bidirectional long short memory are integrated to extract features from dual-scale inputs. Furthermore, active learning approach with random breaking-ties selection strategy is introduced to choose the most representative unlabeled samples for labeling and fine-tune the trained model with these representative labeled samples. The experimental results show that the proposed method uses only 5.14% of the additional data set and improves the model accuracy and macro F1 score by 13.23% and 35.62%, respectively. The proposed structure has good representative ability and our selection strategy can reduce the workload of experts and improve the performance greatly. Thus, the proposed interactive and incremental method has good potential in real applications. [ABSTRACT FROM AUTHOR]

This article investigates a real-time nondestructive measurement method of the seam strength using the optical spectroscopy and ensemble learning in laser beam welding. First, a linear positive relationship between the seam strength and acquired optical spectrum signal was established. Then, a new feature reduction method was proposed to extract the independent feature set with low correlation and high diversity. Finally, a hybrid classification model is proposed based on fast independent component analysis (ICA) and extremely randomized trees (ET). The model was thoroughly verified with various welding experiments and careful comparison with decision tree (DT), extreme gradient boosting DT (XGBoost), and support vector machine (SVM). More importantly, driven by the monitoring data and random forest with fast independent component analysis (RFICA)-ET model, it was found that Al I at 669.8673 nm and Ar I at 610.5635 nm were the two key elements in the dynamic plasma for accurate seam strength measurement. [ABSTRACT FROM AUTHOR]

Autonomous underwater vehicle (AUV) mostly relies on an integrated navigation system, which consists of a strap-down inertial navigation system (SINS) and Doppler velocity log (DVL). The integrated system provides continuous and accurate navigation information when compared to stand-alone SINS or DVL. However, the dependence of DVL signals on the acoustic environment may cause any DVL malfunction due to marine organisms or strong wave-absorbing material. This article introduces a novel method utilizing Dempster–Shafer (DS) theory augmented by least squares support vector machines (LSSVMs) known as DS-LSSVM. The SINS and DVL data fusion are designed by DS theory whereas LSSVM models the SINS error. The virtual DVL is built by the proposed DS-LSSVM method, which makes AUV navigation purposes possible during the long-term DVL outage. The test results demonstrate the effectiveness of the proposed virtual DVL signal estimation method. In addition, the positioning accuracy of the proposed method outperforms that of the LSSVM method. [ABSTRACT FROM AUTHOR]

In this study, we propose the use of electroencephalography (EEG), electrooculography (EOG), and kinematic motion data captured through wearable sensors to classify emotional states, while individuals are playing a serious computer game (Whack-a-Mole). Twenty-one participants wore an OpenBCI headset and JINS MEME eyewear while playing the Whack-a-Mole game at three levels of difficulty. We used a variety of classifiers [i.e., a support vector machine (SVM), logistic regression (LR), random forest (RF), and ensemble classifier (EC)] to classify the participants’ emotional states based on their EEG, EOG, and kinematic motion data. The classifiers were trained using the International Affective Picture System (IAPS). The EC and RF showed the best results in terms of their overall performance. Using tenfold cross-validation for all the subjects, the accuracies obtained were 73% for Arousal and 80% for Valence. Our results suggest that EEG and EOG biosignals, as well as kinematic motion data acquired using off-the-shelf wearable sensors in combination with machine-learning techniques such as EC, can be used to classify emotional states, while the individuals were playing the Whack-a-Mole game. [ABSTRACT FROM AUTHOR]

The quantity security of stored grain in warehouses is crucial for a country. Since manual inspection of stored grain quantity is inefficient, some online automeasurement methods have been proposed recently, but they are either time-consuming or too expensive to be popularized. This article proposes a deep learning-based method to automatically monitor changes in grain quantity of granaries. First, the image of the same scene in a granary is taken at different moments using a device that integrates a camera and an infrared laser rangefinder. Then, a deep semantic segmentation model based on an encoder–decoder framework is developed to extract the grain loading line and grain surface of the image. Finally, the distance and area between the extracted grain loading line and grain surface are calculated and compared with the previous measured one to determine whether the grain quantity of the granary has changed. To improve the accuracy of the segmentation results, a novel reverse attention model is proposed to provide guidance information to fuse low-level features, which calculates multiscale attention maps based only on the output of the last layer of the encoder. Furthermore, the proposed method and infrared laser rangefinder-based method are combined to get more accurate grain volume after an abnormal change of grain quantity is detected by our method. Experimental results show that our method is effective and feasible for monitoring changes in grain quantity and outperforms the state-of-the-art methods on semantic segmentation. [ABSTRACT FROM AUTHOR]

Due to the power quality problems and the use of some nonlinear loads such as soft starters, rectifiers, and frequency converters, the circuit current will contain complicated harmonic components, which may affect the identification accuracy of arc fault. Aiming at the interferences of power supply harmonics and nonlinear load noise, a kind of recognition method based on kernel principal component analysis (KPCA) and firefly algorithm optimized support vector machine (FA-SVM) was proposed. KPCA was used to separate the harmonics and load noise interferences in the voltage and current signals. Kurtosis and skewness of the fifth and sixth principal components were used as arc fault features. The FA-SVM was designed to recognize arc fault. The arc fault experiments were carried out under complicated harmonic conditions. The effectiveness of the presented method was verified by experiments. [ABSTRACT FROM AUTHOR]

Metering data from the advanced metering infrastructure can be used to find abnormal electricity behavior for the detection of electricity theft, which causes huge financial losses to electric companies every year. This article proposes an electricity theft detector using metering data based on extreme gradient boosting (XGBoost). The metering data are preprocessed, including recover missing and erroneous values and normalization. The classification model based on XGBoost is trained using both benign and malicious samples. Simulations are done by using the Irish Smart Energy Trails data set with six certain attack types. Compared with the support vector machine, decision tree, and other eight machine learning methods, the proposed method can detect electricity theft with either higher accuracy or lower false-positive rate. Experiment results also demonstrate that the proposed method is robust when the data are imbalanced. Our codes are available at https://github.com/WenHe-Hnu/Electric_Theft_XGBoost. [ABSTRACT FROM AUTHOR]

Atrial fibrillation (AF) can cause a variety of heart diseases and its detection is insufficient in outside hospital. We proposed three methods for AF diagnosis in ambulatory settings. The first method is a convolutional neural network (CNN) trained on modified frequency slice wavelet transform (MFSWT) data. The second is a support vector machine (SVM) classifier trained on multiple AF features data. The third method is an SVM trained on the same feature set but extended by the predictive probability of the CNN. The proposed method (the third one) achieved the highest detection accuracy. MIT-BIH AF database was used as a training set with an accuracy of 97.87% for 30-s ECG episodes and 96.09% for 10-s ECG episodes from fivefold cross-validation. The trained model was tested on the PhysioNet/Computing in Cardiology (CinC) Challenge 2017 database, achieving an accuracy of 93.21% for 30-s episodes and 93.03% for 10-s ECG episodes. When tested on the China Physiological Signal Challenge (CPSC) 2018 database, the corresponding accuracies were 98.48% and 98.61%. The results on the wearable ECGs from a clinical AF patient were 99.21% and 97.04%. We retrained the model on the PhysioNet/CinC Challenge 2017 data set and tested on the other database to explore the generalization ability of the proposed method. Corresponding test results on the MIT-BIH AF database showed accuracies of 96.84% and 95.13%, on the CPSC 2018 database were 96.21% and 98.45%, on the wearable ECGs were 99.08% and 96.43%. The results proved that the proposed method could provide high accuracy and reliable recognition for AF events. [ABSTRACT FROM AUTHOR]

Gearbox failure is one of top-ranked factors leading to the unavailability of wind turbines (WTs). Existing data-driven studies of gearbox failure detection (GFD) focus on improving detection accuracies while reducing false alarms has not received sufficient discussions. In this article, we propose a deep joint variational autoencoder (JVAE)-based monitoring method using wind farm supervisory control and data acquisition (SCADA) data to more effectively detect WT gearbox failures. The JVAE-based monitoring method includes two parts. First, a novel JVAE that takes a chunk of multivariate time series derived from collected SCADA data as inputs is developed. The JVAE utilizes two types of predefined parameters, behavior parameters (BPs) and conditional parameters (CPs), to produce reconstruction errors (REs) of the BP, which reflects the gearbox abnormality. Next, a statistical process control chart is developed to monitor REs and raise alarms. To validate advantages of the proposed method in GFD, five methods, the joint latent variational autoencoder (JLVAE)-, the variational autoencoder (VAE)-, full-dimensional VAE (FDVAE)-, recurrent autoencoder (RAE)-, and one-class support vector machine (OCSVM)-based monitoring methods, are considered as benchmarks. SCADA data with field reports of gearbox failure events collected from four commercial wind farms are utilized to demonstrate the effectiveness of the JVAE-based monitoring method on GFD and its stronger ability to resist false alarms. [ABSTRACT FROM AUTHOR]

This article investigates how the creeping wave propagation around the human thigh could be used to monitor the leg movements. The propagation path around the human thigh gives information regarding leg motions that can be used for the classification of activities. The variation of the transmission coefficient is measured between two on-body polyethylene terephthalate (PET) flexible antennas for six different leg-based activities that exhibit unique time-varying signatures. A discrete wavelet transform (DWT) along with different classifiers, such as support vector machine (SVM), decision trees, naive Bayes, and K-nearest neighbors (KNN), is applied for feature extraction and classification to evaluate the efficiency for classifying different activity signals. Additional algorithms, such as dynamic time warping (DTW) and deep convolutional neural network (DCNN), have also been implemented, and in each case, SVM with DWT outperforms the others. Simulation to evaluate a specific absorption rate (SAR) is carried out as the antenna is positioned on the human thigh leaving no gap. The results show that the SAR is within the threshold as per the Federal Communications Commission (FCC) standard. [ABSTRACT FROM AUTHOR]

Condition monitoring for wind turbine (WT) planetary gearbox is of great significance to the reliable operation of WT. A novel condition monitoring method for WT planetary gearbox is proposed in this article. This method can identify and predict the fault in the initial stage. Raw vibration signals are gathered from WT planetary gearbox through an enhanced piezoelectric self-powered wireless sensor. First, after the data processing back-end receives the original samples transmitted by the wireless sensor, the high-dimensional raw vibration signals are compressed and collected by adopting a random Bernoulli matrix to obtain the compressed samples containing the characteristics of the raw signals by compressed sensing (CS). Moreover, the operation of reducing the signal dimension not only eliminates noise pollution but also greatly reduces the overall calculation. Second, the training part of compressed samples is first exploited to optimize the deep belief network (DBN) by the chaotic quantum particle swarm optimization (CQPSO) algorithm. In the optimization process, the CQPSO algorithm can avoid local optimal problem better compared with traditional quantum particle swarm optimization (QPSO) algorithm, which ensures that the optimized DBN architecture can extract distinguishing and deep features from compressed samples. Furthermore, the choice of compression ratio (CR) is realized in this process with the least-squares support vector machine (LSSVM) classifier. Third, testing samples are input into the DBN structure. A regression layer added on the last hidden layer, which stores extracted features, achieves the prediction of fault. In addition, LSSVM is exploited to distinguish fault types of features. The experimental results show that the modified wireless sensor can collect and transmit signals stably and achieve better performance in transmission interval and energy management. More importantly, the experimental results of prediction and diagnosis show that the proposed approach achieves excellent performance in terms of condition monitoring for the WT planetary gearbox. [ABSTRACT FROM AUTHOR]

In this article, the fault diagnosis problem for power converters is considered. Given that the existing fault diagnosis models rarely address the problems of the data noise and the new faults that are never emerged in the database, thus, an optimized fault diagnosis model for power converters based on temporal convolutional network (TCN) is proposed. Our contributions include the following: 1) unknown faults can be efficiently distinguished with an optimized classifier; 2) the proposed model has good robustness and reliability under noisy environment without any subsidiary predenoising algorithm; and 3) it can realize adaptive feature extraction, and the parameters are small. Experimental results on a three-phase voltage inverter platform demonstrate that the proposed approach is efficient and can be adaptively applied to various real applications. [ABSTRACT FROM AUTHOR]

To enhance the reliability of analog circuits in complex electrical systems, a novel incipient fault diagnosis method is presented in this article. The wavelet packet feature quantities, which consist of the energy, fluctuation coefficient, skewness, and margin factor, are obtained via multiscale time-frequency analysis with wavelet packet transform (WPT). Then, generalized discriminant analysis (GDA) is employed to realize the fusion of wavelet packet feature quantities because it can handle the data nonlinearity and eliminate redundant information. Furthermore, the generalized multiple kernel learning support vector machine (GMKL-SVM), which has the advantages of a strong generalization ability and high accuracy, is developed to identify the incipient fault classes of analog circuits. Moreover, a new particle swarm intelligent optimization algorithm, the sine cosine algorithm (SCA), is adopted to optimize key parameters of GMKL-SVM because of its high convergence speed and strong global optimization ability. The method is fully evaluated with the Sallen-Key bandpass filter circuit, the four-op-amp biquad high-pass filter circuit, and the leapfrog filter circuit. The experimental results demonstrate that the proposed incipient fault diagnosis method for analog circuits can produce higher diagnosis accuracy than other typical incipient fault diagnosis methods. [ABSTRACT FROM AUTHOR]

This article focuses on the development and validation of an innovative method for estimating the degree of saturation (DOS) in limestone. A protocol is designed to control, homogenize, and validate such an estimation. The complex dielectric permittivity is acquired from the coaxial electromagnetic transition line on a limited frequency bandwidth. Calibration curves connecting the dielectric permittivity and the DOS are obtained. The use of variants of Jonscher’s model makes it possible to overcome the frequency dependence and propose a relationship between Jonscher model parameters and the DOS. A supervised learning method [ $v$ -support vector regression (SVR)] has been successfully introduced to estimate the DOS from these parameters. The results have been experimentally validated on a wide range of limestone samples. [ABSTRACT FROM AUTHOR]

Given that there exists input uncertainty caused by the noise embedded in industrial data, this study proposes a relevance vector machine (RVM) prediction model with input noise. Due to the fact that the marginal likelihood cannot be analytically calculated when introducing the input uncertainty, a Gaussian approximation is proposed in this study on the basis of the law of total expectation and the law of total covariance. Furthermore, to approximate the posterior distribution over the model weights, this study employs the Markov chain Monte Carlo algorithm, where a Gaussian proposal distribution is designed to draw new samples. In the prediction stage, a Gaussian approximation is also designed for a new testing input in order for the input uncertainty to be reflected in the estimation of output variance. To verify the effectiveness of the proposed method, four synthetic data sets, four benchmark data sets, and two industrial data sets are employed in the comparative experiments. The results indicate that the proposed RVM with uncertain input outperforms other approaches, and it also performs better on the time series prediction issue. [ABSTRACT FROM AUTHOR]

Bridge health assessment has been a challenging problem due to great evaluation errors caused by heterogeneous and different dimensional bridge factors. Most approaches employ monitoring data of one bridge factor to simplify the evaluation problem resulting in poor assessment performance. To address this issue, this article proposes an end-to-end framework to evaluate the health of bridges by exploring objective features and correlations of multiple monitoring factors. This model aims to learn representative features from raw monitoring data of bridge factors (i.e., strain, temperature, traffic flow, and heavy vehicle number) and classify the comprehensive features into different health degrees in a single framework. Especially, in terms of characteristics of bridge factors (e.g., time sequence and heterogeneity), a hierarchical learning structure with multiple convolutional, pooling, and dense layers is presented to learn the representations of the whole bridge monitoring data. The structure contributes to capturing rich information from different observed factors and improving feature learning ability through designing particular neural networks for each bridge factor in accordance with their corresponding data structure. In addition, a classification scheme with multiple fully connection layers and support vector machines in the framework is designed to achieve higher evaluation accuracy. Experiments on real-world monitoring data of a specific bridge validate the superiority of the proposed model. [ABSTRACT FROM AUTHOR]

The reliable classification of coin sets from different countries and counterfeit coins is an essential means of securing cash circulation. Especially bimetallic coins, such as 2-Euro coins, are often subjected to counterfeit by mixing them with coins of other countries or by imitations. In this article, a real-time embedded sensor system is proposed, based on inductance spectroscopy to characterize and identify bimetallic coins having similar geometric properties and looking similar at a first view. The system is based on inductance spectroscopy varying the penetration depth of the magnetic field in the bimetallic structure of the coin, which generally contains buried layers of other metals. The experimental evaluation shows that the bimetallic coins from different countries can be identified and classified by the use of the support vector machine, a machine learning algorithm. Experimental results showed that the system reached a classification accuracy up to 100% within a response rate of 36.62 ms per five bimetallic coins. [ABSTRACT FROM AUTHOR]

The nonidentical and nonindependent distribution problem, caused by various working conditions of gears, inevitably makes the effectiveness of gear fault diagnosis (GFD) degrade because of insufficient or low-quality data. Although some domain adaptation learning (DAL) models exist, most of them only aim at minimizing the global distributional mean discrepancy between the source domain (SD) and the target domain (TD), while the contribution of individual data is neglected. To address this issue, a new DAL, aiming to reduce the discrepancy on extracted features under a least square support vector machine (LSSVM) framework, is studied to exploit SD signals from another working conditions or adjacent mechanical parts to assist and boost target gear fault diagnostic performance in this article. In addition, the white cosine similarity criterion is adopted to quantize the distributional weights of SD and TD feature data, and then the weights are added into a regularization function that measures the projected distributional discrepancy in the LSSVM framework. Related experiments prove that the proposed method has higher diagnostic accuracies than other models. So, this strategy is expected to be a useful tool to transfer from SD to TD, and to boost the GFD performance under various working conditions. [ABSTRACT FROM AUTHOR]

Emotion is a neuronic transient that drives a person to a certain action. Emotion recognition from electroencephalogram (EEG) signals plays a vital role in the development of a brain–computer interface (BCI). Extracting the important information from raw EEG signals is difficult due to its nonstationary nature. Fixing a factual predefined basis function for efficient decomposition using a tunable $Q$ wavelet transform is an arduous task. In this article, an adaptive tunable $Q$ wavelet transform is proposed for the automatic selection of tuning parameters. Optimum tuning parameters are obtained using gray wolf optimization (GWO). Tuning parameters obtained by GWO are used to decompose the EEG signals into subbands (SBs). The set of time-domain features elicited from the SBs are used as an input to multiclass least-squares support vector machine. Classification accuracy of four basic emotions, namely, happy, fear, sad, and relax, is tested and compared with existing methods. An accuracy of 95.70% is achieved with a radial basis function kernel that is about 5% more than the existing methods using the same data set. This article proposes the development of a nonparameterized decomposition method for efficient decomposition of EEG signals. This method can be used with machine learning algorithms to take a step forward in the development of BCI systems. [ABSTRACT FROM AUTHOR]

In this article, Stockwell transform (ST) is used to analyze the stator current signals for diagnosis of various motor conditions such as healthy, stator winding interturn shorts, and phase to ground faults. ST decomposes the current signals into complex ST matrix whose magnitude has been utilized for the fault detection. The nature of the fault, that is, ground or interturn is identified using the zero sequence currents followed by postfault detection. Two separate frequency bands are defined to extract the features which are fed to two different support vector machine (SVM) models for faulty phase detection for both types of faults. Under both cases, a heuristic feature selection approach is utilized to find the optimal features for classification purposes. Average classification accuracy of 96% has been achieved for both types of faults. [ABSTRACT FROM AUTHOR]

Sensor signals have been frequently used for activity recognition and gender classification in the literature. In this study, a new multikernel local diamond pattern (MK-LDP) is proposed as a novel local descriptor method for feature extraction. The proposed MK-LDP is aimed to generate distinctive features from a signal or image using vertical and horizontal diamond-like patterns and multikernel functions. These kernels are signum, ternary, and quaternary. MK-LDP is utilized as a feature extraction technique for human activity recognition (HAR). The proposed HAR method has four fundamental phases, namely, preprocessing, feature generation with MK-LDP, informative features selection using hybrid ReliefF and neighborhood component analysis (RFNCA), and classification using support vector machine (SVM). The proposed MK-LDP extracts 2560 features from the raw sensor signals, RFINCA selects 512 most meaningful ones from the extracted 2560 features, and then the selected 512 features are used as input of the SVM for HAR. Three cases are defined to test the proposed MK-LDP- and RFNCA-based HAR method. These cases are gender classification, activity classification, and both gender and activity classification, respectively. The achieved best accuracy rates are 99.47%, 99.71%, and 99.36% for gender, daily sports activities, and both gender and daily sports activities recognition, respectively. The proposed MK-LDP-based method is also compared to the state-of-the-art and deep learning techniques. The obtained results revealed that the proposed MK-LDP- and RFNCA-based framework is successful HAR using sensor signals. [ABSTRACT FROM AUTHOR]

Railway turnouts require high-performance condition monitoring to prevent disastrous railway accidents. In industrial practice, turnouts’ monitoring is usually done by railway workers who visually inspect the operating current curves. This results in huge labor costs and prone to human mistakes. Thus, automating the process of turnouts’ monitoring via fault-detection algorithms is imperative. The available turnout field data bring three difficulties to fault detection: 1) large amounts of data do not have any labels; 2) data collected in normal condition have multiple unknown modes; and 3) there are only a small number of samples in some modes. To address these difficulties, this article develops a novel unsupervised fault-detection method by using deep autoencoders, which is composed of an unknown modes’ mining stage and a multimode fault-detection stage. First, unknown modes are identified through clustering and employing engineer expertise. Then, an ensemble monitoring model, consisting of local monitoring models developed with individual fault-free modes and a global monitoring model developed by merging the data in all fault-free modes, is proposed to improve the overall fault-detection performance. In addition, to construct local models for the modes with a small number of samples, a one-class transfer learning algorithm is presented. In online monitoring, the decision of a newly arrived sample exploits both local models and the global model. Using both the simulated turnout data and the field data collected from a high-speed railway in China, the efficacy and robustness of the proposed approach are demonstrated by comparisons with other methods. [ABSTRACT FROM AUTHOR]

A highly wearable single-channel instrument, conceived with off-the-shelf components and dry electrodes, is proposed for detecting human stress in real time by electroencephalography (EEG). The instrument exploits EEG robustness to movement artifacts with respect to other biosignals for stress assessment. The single-channel differential measurement aims at analyzing the frontal asymmetry, a well-claimed EEG feature for stress assessment. The instrument was characterized metrologically on human subjects. As triple metrological references, standardized stress tests, observational questionnaires given by psychologists, and performance measurements were exploited. Four standard machine learning classifiers (SVM, k-NN, random forest, and ANN), trained on 50% of the data set, reached more than 90% accuracy in classifying each 2-s epoch of EEG acquired from the stressed subjects. [ABSTRACT FROM AUTHOR]

Point-of-care Ultrasound (PoCUS) is a safe, repeatable, and inexpensive bedside diagnostic tool. Over the years, PoCUS services are adopted in resource-limited settings for faster and useful outcomes. For a cost-effective and power-efficient solution, a smartphone-based portable continuous-wave Doppler ultrasound (US) system has been developed for diagnosing peripheral arterial diseases based on the hemodynamic feature values. The proposed system includes the analog front end (AFE), signal processing and display unit (SPDU), and smartphone application. The AFE acquires blood flow signal from the brachial artery using an 8-MHz pencil probe, extracts the Doppler shift frequency, and transfers to the SPDU through 12-bit analog-to-digital converter. To provide an area and power-efficient solution, SPDU is embedded in a field-programmable gate array (FPGA)-based single chip. A COordinate Rotation DIgital Computer (CORDIC)-based custom-designed 512-point fast Fourier transform is implemented in that FPGA for displaying the blood flow spectrogram in real time. For back-end processing, the smartphone application receives a spectrogram through Bluetooth, removes noise, extracts hemodynamic features, and diagnoses using a machine learning framework. The device has been examined on 18 volunteers (normal: 17 and abnormal: 1), while the accuracy is found to be 94% in the pretrained support vector machine classifier. For validation, the spectrogram of the normal and abnormal subjects and parameter values are compared with the commercial device. Overall, the handheld device is minimally trained operator-dependent and consumes < 4 W of power for real-time processing. Such smartphone-based feature extraction and automated diagnosis can facilitate the point-of-care system and provide a baseline for early assessment. [ABSTRACT FROM AUTHOR]

In this article, a noise-robust self-adaptive support vector machine (namely, NSSVM) is developed for the multiclass classification problem in the residual oxygen concentration measurement of pharmaceutical vials. Two key elements are emphasized by the NSSVM framework: First, targeting the fast time-varying noises, two mutually compensated signal enhancement measures of synchrosqueezed wavelet transform (SWT)-based filtering and adaptive iterative reweighted penalty least squares (AIRPLS)-based baseline corrector are contributed to obtain stable features being fed to classifier. Second, most typical slow time-varying interferences are dynamically avoided by the self-adaptive thresholding scheme learned from the intrinsic priori of production line. The experimental results indicate that our NSSVM-based measurement method achieves a considerable performance, with an average accuracy of 97.24% when applied on an automated visual inspection (AVI) machine of Truking Tech., which indicates that the proposed method is promising to the in situ instrumentation for the residual oxygen concentration measurement of pharmaceutical glass vials. [ABSTRACT FROM AUTHOR]

This article aims to propose a thermography-based fault detection approach for a brushless synchronous generator. In this approach, the energy of 2-D wavelet coefficients is utilized to extract features from thermal images. In order to select the appropriate wavelet basis function and decomposition level (DL), a novel filter feature selection called neighbors class solidarity (NCS) is propounded. To this purpose, the energy of the approximation coefficients is obtained for 65 wavelet basis functions and 12 DLs. Subsequently, the best feature is selected by using NCS. The selected feature is then applied to k-nearest neighbors and support vector machine in order to classify the images into three operational states. These states include healthy, one diode open-circuit fault of the diode rectifier, and one open-phase fault of the exciter armature, while the tests are conducted under three different load conditions for each state. Results imply the efficiency of the proposed approach for the fault diagnosis in the brushless synchronous generator with an accuracy of 95.1%. To evaluate the performance of the NCS, it is compared with other conventional filter feature selection methods, and also for different test analysis. [ABSTRACT FROM AUTHOR]

The variability of machinery fault signatures causes the data samples to follow different distributions under various operating conditions, which poses significant challenges on autonomous diagnosis based on machine learning techniques. This article presents a new transfer learning method for cross-domain feature learning by mitigating the domain difference caused by various operating conditions for machinery autonomous diagnosis. More specifically, a factor analysis (FA)-based transfer learning method is formulated and named as probabilistic transfer FA (PTFA). It seeks a new feature space across different domains corresponding to various operating conditions and then transfers the original features into a low-dimensional latent space via feature extraction to minimize domain difference and preserve data properties. The learned features by PTFA minimize the domain difference and are then used to construct the machinery autonomous diagnosis model based on machine learning techniques [e.g., support vector machine (SVM)]. The effectiveness of the PTFA method is demonstrated in the experimental tests for a gearbox diagnosis under various operating conditions comparing with traditional feature extraction and transfer learning techniques. [ABSTRACT FROM AUTHOR]

Effective fault diagnosis of rotating machinery plays a pretty important role in the enhanced reliability and improved safety of industrial informatics applications. Although traditional intelligent fault diagnosis techniques, such as support vector machine, extreme learning machine, and convolutional neural network, might achieve satisfactory accuracy, a very high price is caused by marking all samples manually. In this article, a novel fault diagnosis method of the rotating machinery is proposed by integrating semisupervised generative adversarial nets with wavelet transform (WT-SSGANs). The proposed WT-SSGANs’ method involves two parts. In the first part, WT is adopted to transform 1-D raw vibration signals into 2-D time–frequency images. In the second part, the 2-D time–frequency images are inputted into the built SSGANs’ model to realize fault diagnosis with little labeled samples. The advantage of the built model is that the unlabeled samples might be made full use of through an adversarial learning mechanism. Finally, two case studies are implemented to verify the proposed method. The results indicate that it can achieve higher accuracy and use less labeled samples than the other existing methods in the literature. In addition, its performance in stability is pretty good as well. Competitive and promising results are still achieved when working conditions are changed. [ABSTRACT FROM AUTHOR]