查看更多>>摘要：Thermopile infrared (IR) sensors are thermal-type sensors. It is of great significance to evaluate its four thermal parameters: Seebeck coefficient, thermal conductance, heat capacitance, and thermal time constant. In this work, an effective self-test method to measure these thermal parameters is proposed that takes advantages of ther-mopile IR sensors' structural and electrical characteristics and does not need to add heater in structure. The technique to accurately extract thermal parameters is analyzed carefully, and its validity is verified by a ther-mopile IR sensor. The experimental Seebeck coefficient is consistent with the tested value of on-chip test structure in the same die, and the experimental thermal conductance, heat capacitance, and thermal time con-stant results agree well with the theoretical analysis of structure. These demonstrate that this method is effective and accurate, as well as simple for researchers to apply.
查看更多>>摘要：It has become a challenge to accurately extract weak bearing fault features from early fault stage. To solve this problem, a novel fault features extraction method called improved Kurtogram and Hyper-Laplacian Sparse Coding (KurHLSC) based on probability sparse coding is proposed in this paper. The originality of the present article lies in the construction of a sparse coding model considering probability and wavelet dictionary, which can effectively decompose sparse fault features even in strong noise. Moreover, in order to eliminate the interference of random pulse on sparse coding model, the improved kurtogram method successfully achieved filtering. The effectiveness of KurHLSC in rolling bearing fault diagnosis is verified by simulation studies and run to-failure experiments, and the comparison studies showed that KurHLSC has better estimation results than other existing methods.
查看更多>>摘要：This work reports a study aimed at identifying the operating state of gear through an enhanced multiclass support vector data description (eMSVDD) model where the multiple hyperspheres corresponding to different operating states of gears are established. The model addresses the following issues: (1) the overlap of multiple hyperspheres deteriorates the diagnostic performance for boundary samples and severely weakens the generalization ability of the model; (2) the original support vector data description can only be applied to handle binary classification tasks; and (3) the performance of the SVDD depends heavily on the penalty factor C and kernel parameter delta. In the proposed model, we first implement the multi-label classification task by building multiple hyperspheres. Then, a novel fitness function for adaptive simplified chaotic particle swarm optimization is proposed to determine the model parameters, which considers the empirical risk minimization and model generalization capability. Finally, experimental results demonstrate that the eMSVDD outperforms deep neural networks and support vector machine by 3.4% and 0.5%, respectively, and this performance improvement respectively reaches 17.8% and 2.1% in the case of a very small sample size. The proposed approach provides a means for gear fault diagnosis with small samples.
查看更多>>摘要：In this paper an innovative spring-shaped fiber modulation (SSFM) based on bending loss characteristic of the optical fiber and spatial helical structure with simple construction and cheap cost was developed for linear displacement monitoring. The structure design and sensing principle of the light intensity with displacement variation were reported. The structural analysis, parameter selection and monitoring performance of the SSFM constructed with two kinds of host materials (nylon carbon fiber spring and spring steel) were investigated through theoretical analysis and laboratory tests. The results showed that the SSFM can be used to modulate the optical fiber loss, and the three equal-pitch coils SSFM made of steel spring material with an initial diameter of 15 mm, an initial pitch of 15 mm would be a good choice. It was characterized by a high linearity, low hysteresis of 4.47%, suitable sensitivity of 0.0202 dB/mm and measurement range of 30 mm, respectively. Additionally, the SSFM can be connected in series with other structures to fabricate a sensor for detection of subsurface properties with a larger measurement range and it has a good multiplexing capacity. The capability supports the promising prospect of the SSFM in civil engineering for deformation monitoring of fill slopes and foundation pits.
查看更多>>摘要：Compressive strength (CS) is the maximum resistance of concrete against axial compressive loading in standard conditions. Estimation of this parameter is essential for the proper design of concrete mixture. Considering the complexity of this task as a burden for traditional approaches, machine learning models like artificial neural network (ANN) have been successfully used for analyzing the nonlinear relationship between the CS and concrete ingredients. This study implements two ANN-based scenarios to approximate the uniaxial CS of manufactured sand concrete. First, the ANN is trained by nine regular algorithms, and the best one is selected to represent the conventional ANN (CNN). For the second scenario, two improved ANNs are created with metaheuristic algorithms, namely biogeography-based optimization (BBO) and multi-tracker optimization algorithm (MTOA). The first scenario revealed that Levenberg-Marquardt is the strongest regular trainer. Comparing the performance of the CNN with hybrid models showed that both BBO and MTOA can construct a more accurate ANN. In this sense, root mean square error of the CNN experienced 8.77 and 8.84% reduction in the training phase, and more effectively, 13.05 and 11.46% in the testing phase by applying the BBO and MTOA, respectively. Hence, the suggested hybrids can act as promising alternatives to traditional models for predicting the CS of concrete. Two explicit formulas optimized by the MTOA and BBO are derived for practical applications. Also, importance analysis revealed the high contribution of curing age and water to binder ratio to the compressive behavior of the concrete.
查看更多>>摘要：This paper proposes a method of an explainable ensemble tree model in an optimized feature space, which is based on the wave propagation model and the leakage noise mechanism. Specifically, the vibration signal is analyzed and the piecewise power spectrum entropy is proposed and used to construct the feature space.The Boruta algorithm is used for feature reduction; then, four ensemble tree models are applied to build leakage identification models. Furthermore, Shapley Additive explanation method is used to select the optimal feature space. In addition, four groups of experiments were designed with different aperture, and 13 features were obtained after dimensionality reduction with the measured distance as the variable, the XGBoost model with the highest accuracy was selected, and 7 features were obtained using SHAP. Finally, the performance of the methodology is evaluated with different pipeline leakage scenarios and different algorithms, and the results demonstrate its application capability in the field.
查看更多>>摘要：In this paper, we experimentally propose and theoretically investigate a two-dimensional vector curvature sensor based on asymmetrically polished long period fiber grating (AP-LPFG). This sensor is fabricated by a high-frequency CO2 laser in order to polish periodic asymmetrical V-shaped grooves on a single mode fiber (SMF). Such grooves efficiently enhances the bending sensitivity of sensor and shorten the length of the structure to 7.39 mm. Experimental results show that the sensitivity of AP-LPFG reaches 30.85 nm/m(-1) in the direction of the X-axis and 0.97 nm/m(-1) in the direction of the Y-axis. Meanwhile, the temperature response of AP-LPFG is carried out, which is 76 pm/degrees C in range of 30-160 degrees C. Moreover, the AP-LPFG has the features of compact length and simple structure, which shows the potential in curvature sensing.
查看更多>>摘要：Fusion object detection using camera and LiDAR information in autonomous driving is still a challenging task, the difference between sensor data increases the difficulty of data fusion. To address this issue, we propose a multi-scale selective kernel fusion(MSSKF) method and demonstrate its practical utility by using LiDAR-camera fusion in object detection network. Specifically, a multi-scale feature fusion module that uses multi-scale convolution to separate the feature expression of multi-modal information and calculates the weight of each modal feature channel is proposed. We use the idea of multi-scale convolution and selection kernel to complete multi-modal fusion in object detection, which is conducive to solving the problem that the image and point cloud fusion are difficult to match due to the difference in data structure, and the complementarity of multi-modal information has been fully utilized. To verify the effectiveness of MSSKF, experiments on the KITTI object detection benchmark dataset are conducted. It has been observed that the proposed method achieves more accurate detection for pedestrians and vehicles, with a 1.6% gain in AP(50) compared to the values of the original fusion method, reaching a score of 90.1%, and the mAP reached 60.9%. Experiments show that the proposed method introduces a new optimization idea for multi-modal fusion in the field of autonomous driving object detection, and the fusion detection efficiency is at over 12 fps on a single GPU.
查看更多>>摘要：The existing, universal approaches to sensory system calibration do not consider target application during calibration. We propose a new approach to calibration that uses the target application processing pipeline to measure the calibration quality. This approach is exemplified and tested in the calibration of an onboard vision system used to localize a city bus to an electric charging station. Our calibration procedure determines the parameters applying optimization and automatic features detection from a deep learning-based vision processing pipeline. The parameters are calibrated to produce localization estimates matching the ground truth pose measurements. We measure the calibration's performance in a target application on over 10000 poses recorded from a real city bus over several days. The automatic calibration procedure results in numerous optimization constraints, which can be used to calibrate more parameters and reduce manual labor. The proposed approach reduces the localization errors by almost 50% in the presented target application.
查看更多>>摘要：Video object segmentation (VOS) has been a research hot-spot these years. However, evaluating the performance of different VOS methods requires labor-intensive and time-consuming manually labeled mask annotations, making it hard to validate the algorithm quality in field tests. In this paper, we tackle the problem of automatically measuring the mask quality for video object segmentation tasks without accessing manual annotations. We propose that with an elaborately designed network structure, we can extract quality sensitive features to predict mask quality scores without ground-truth labels. To achieve this, we train an end-to-end convolutional neural network to capture the quality-sensitive features with both spatial reference and temporal reference. In the proposed Video Object Segmentation Evaluation Network, the VOSE-Net, the corresponding video frame and motion amplitude information are used for spatial and temporal references respectively. Instead of directly concatenating features for mask and references, we extract spatial quality cues with feature correlation, which is more rational and effective in this specific task. Taking in the segmented mask, its corresponding frame image and optical flow map, the VOSE-Net can provide an accurate quality estimation without the need for human intervention. To train and verify the proposed network, we construct a new dataset by using the DAVIS video segmentation benchmark and results from many public video object segmentation algorithms. We also demonstrate the robustness and usefulness of the proposed method on several applications, i.e. proposal selection, parameter optimization, arbitrary video mask evaluation. The experimental results and analysis show that the VOSE-Net is fast, effective and of practical use.
NSTL
Elsevier