Sleep apnea is a common sleep disorder that occurs due to repetitive obstruction of the airflow in human body that seriously affect the lives and health of people. This article aims to present an improved time–frequency transformation aided deep learning framework employing convolutional neural network (CNN) for automated diagnosis of apnea using single channel electrocardiogram (ECG) signals. The recorded ECG signals are further processed and converted into two-dimensional time frequency (TF) images via an iterative multisynchrosqueezing transform (I-MSST) for necessary feature extraction. After that the TF images are fed to a cascaded deep dense CNN (DCN). The advantages of the proposed module are twofold. Firstly, the I-MSST based images possess a concentrated TF plot superior to other conventional TF plots, and secondly, the cascaded DCN is able to concatenate features throughout the convolutional layers with the help of a conversion block. Moreover, an optimal DL framework has been formulated by tuning the hyperparameters via Bayesian optimization. The proposed method has been validated on two benchmark datasets for comparative analysis, and it is found that the proposed module can effectively diagnose SA with increased performance for an improved computer aided diagnosis system.

The motion state estimation is indispensable for unmanned ground vehicles. Most traditional methods estimate individual states and utilize high-dimensional dynamics models that necessitate extensive prior information to improve accuracy. However, these approaches are beset with numerous challenges in the accuracy, complexity, and computational burden. In this paper, a novel method is proposed to address these problems with two novelties. Firstly, it possesses the capability to simultaneously estimate 10 fundamental states, including vehicle velocities, wheel slip ratios, and wheel sideslip velocities. The estimation accuracy is augmented since the coupling relationships among states are considered. Secondly, the method avoids the utilization of the complete vehicle model, instead employing the constraints derived from the models, thereby the requisite prior information and computational burden can be reduced. The accuracy of the proposed method was demonstrated through simulation experiments and field tests, as well as its robustness in challenging situations, such as the wheel slip.

This article presents a method for three-dimensional reconstruction based on line laser-assisted virtual stereovision, elaborating on the equipment selection strategy for virtual stereovision under actual working conditions of cranes, the principles of distance measurement and its calibration methods, as well as the procedures for material 3D reconstruction and point cloud data splicing updates. In a coal stockyard scenario, we conducted a 3D reconstruction experiment on coal inventory using an intelligent bridge crane. Tailored to the bridge crane’s operational environment, we selected an appropriate number of industrial cameras and corresponding lens focal lengths to achieve the 3D point cloud reconstruction of the crane’s working area. The experimental results demonstrate that within the range of 9977.1 to 13037.1 mm, the fluctuation in distance measurement error was between −12.5 mm and 13.1 mm, with an average error of approximately 1.99 mm and an average relative error of about 0.022 %.

Carbon fiber-reinforced polymer (CFRP) composites are widely used in aerospace but are susceptible to damage, threatening structural safety. Thus, developing an effective structural health monitoring system is essential. This study presents CFRP-former, a novel deep learning framework that detects damage in CFRP composites by constructing spatio-temporal representation graphs of Lamb waves (LWs). The CFRP-former comprises three key modules: graph convolutional neural network (GCNN), a multi-layer perceptron (MLP), and a Transformer Encoder. The GCNN models the topological relationships between LW nodes and sensors. The MLP reduces data dimensionality while preserving key information for subsequent analysis. Meanwhile, the Transformer Encoder, employing multi-head attention mechanisms, captures global time-series patterns in the LW signals. Additionally, the CFRP-former framework includes a parallel module that separately predicts both the size and location of damage. Experiments using only four sensors show that CFRP-former achieves high accuracy in detecting and quantifying damage, with robustness confirmed through ablation tests.

This study addresses the challenge of diagnosing motor faults in long-tailed data distributions, characterized by dominant healthy states and rare fault types. We propose the LT-CVAE-GAN model, which integrates a Conditional Variational Autoencoder (CVAE) with a Conditional Generative Adversarial Network (CGAN) to enhance long-tailed fault diagnosis. Initially, we train the CVAE-GAN model using traditional CVAE and CGAN losses such as Kullback–Leibler (KL) divergence loss, reconstruction loss, and adversarial loss. Additionally, we introduce mean feature matching loss and pairwise feature matching loss to mitigate mode collapse and improve model stability, thereby enhancing the generation ability of less frequent fault samples under long-tail conditions. Subsequently, the pre-trained Generator is used to produce infrequent fault mode data to rebalance the dataset. Classifier parameters are fine-tuned in this step to improve fault diagnosis accuracy. Experimental results demonstrate that our LT-CVAE-GAN surpasses state-of-the-art models in diverse long-tailed conditions.

Currently, instruments capable of simultaneously measuring airflow velocity and direction are costly and bulky, which presents a challenge in high spatial and temporal resolution in airflow measurement. To address these issues, this paper explores a method that uses artificial neural networks to calculate airflow velocity and direction, capable of measuring the velocity and direction of airflow within the range of 2–12 m/s with high spatiotemporal resolution. Based on the results of comparing the pressure difference response to wind speed in flow simulation, the airflow sensor is designed in a hexagonal prism shape, using low-cost, tiny MEMS pressure sensor chips, with an overall size of only 13 mm × 13 mm × 9 mm, and a power consumption of less than 1.2 mW. Performance tests in the wind tunnel showed that the average error in airflow velocity measurement was 0.20 m/s, and the average error in airflow direction measurement was 0.74°.

Impedance control is a commonly used approach for robots to interact with environments. At present, the impedance control is mostly implemented at the foot end, while the impedance control at the joint space offers faster response speed and is a more favorable option. However, due to the influence of installation position and moment arm of hydraulic drive unit (HDU), it is more complicated to establish the mapping relationship of equivalent stiffness and equivalent damping between the foot end and joint HDU. To address this challenge, this paper provides a general calculation expression of mapping matrices between the foot end and joint HDU, which also considers the influence of above nonlinear factors. Additionally, the accuracy of mapping matrices is verified through the impedance control, and the “finite difference method” (FDM) is used to enhance the accuracy of validation. Finally, the accuracy of mapping matrices is validated through the single leg of a 2-DOF series hydraulic robot.

The trade-off between sensitivity and bandwidth has limited the clinical application of spin exchange relaxation-free (SERF) atomic magnetometers in magnetocardiac measurements. This research addresses the issue by modeling the SERF magnetometer as a first-order inertial link and proposing a method that integrates transient response with tracking signals. The tracking-differential link enables the simultaneous acquisition of both transient and tracking signals, leading to effective noise suppression. Empirical mode decomposition is employed to extract features and filter out noise, ensuring high signal quality. This method extends the bandwidth of dual-beam magnetometers from 9.77 Hz to 240 Hz and single-beam magnetometers from 139 Hz to 289 Hz, while preserving sensitivities of 1 fT/Hz1/2 and 10 fT/Hz1/2, respectively. This research enables high-bandwidth measurements with sub-femtotesla magnetic field sensitivity, presenting promising application opportunities.

The documentation of historical structures is considered as a crucial subject in international regulations. The accuracy and speed of documentation are among the significant factors for the qualified transmission of cultural values to future generations. In this study, the accuracy and position precision of ViDoc device, which integrates with an iPad Pro equipped with LIDAR system, have been determined. Furthermore, the advantages and disadvantages of this device have been elaborated in detail. The aim of the study is to determine whether the ViDoc device and iPad Pro can be used in documenting historical structures. The study covers the traditional method of documentation based on measurement with meter and analog devices, documentation with a total station, and documentation with the ViDoc device, using two historical fountains selected as examples. Two cultural heritage fountains located in Kastamonu province, Turkey, have been selected for documentation. Measurements of the fountains were taken using a precision meter and a laser meter in the classical method, and corner point coordinates were obtained with a total station. The measurement process was carried out in one go using a ViDoc device and a smartphone. As a result of the study, it was understood that the ViDoc device made very positive contributions to the scanning accuracy of smart devices with LIDAR sensors. When comparing the measurement values in the documentation of the structures, the accuracy precision was found to be less than 1 cm (0.7 cm), and the positioning accuracy was less than 5 cm. In the measurement conducted without using the ViDoc device, differences in position distances reached up to 79 m. This indicates that the device can be used for documentation in structures without creating disadvantages. However, it has been determined through the study that the device may not be usable for every structure. Nevertheless, the study also evaluates the device as a new alternative alongside traditional documentation, photogrammetry, and terrestrial laser scanning systems. The study is significant as it is the first to holistically determine the measurement accuracy of the device and the positioning accuracy specifically in relation to historical structures.

Partial discharge (PD) identification is critical for the insulation diagnosis of cable conductors; however, there is still scope for enhancement in the existing adaptive extraction capabilities and feature utilization for PD signals. In this context, this paper introduces a method that integrates a one-dimensional convolutional neural network (1D-CNN) with a residual graph convolutional neural network (ResGCN) to recognize PD signals. Noise analysis is performed using various combinations of mother wavelets, and Bayesian optimization is employed to mitigate background noise. Key features of PD signals are progressively abstracted through a 1D-CNN-based multilevel automatic feature learning method, while signal timing attributes are maintained to minimize manual intervention. The graph data is constructed using the signal feature matrix and the signal timing feature similarity matrix. This is followed by the development of a ResGCN utilizing a graph attention mechanism to integrate node feature information and the topology of the PD graph data. This approach aims to fully exploit the correlation between local regions of the feature space and the temporal numerical properties of the signals. Additionally, it jointly optimizes feature extraction and model classification to facilitate adaptive diagnosis. The method is validated with extensive real experimental data obtained from medium voltage overhead power lines. It demonstrates exceptional performance and practicality, achieving an accuracy rate of 97.3% and a recognition rate of 96.1% for PD samples, thus offering reliable theoretical support for effective PD detection.