查看更多>>摘要：In this paper, the polyimide-coated fiber Bragg grating (PI-FBG) humidity sensor suitable for use in unsaturated soils was developed. Through a series of laboratory tests, the pore gas relative humidity in different sand-kaolin mixtures were measured and analyzed to verify the feasibility of the sensor in unsaturated soils. The results shows that the pore gas relative humidity in unsaturated soils first increases rapidly and then tends to be stable with the increase of water content. The relationship between them can be fitted by quadratic function when the pore gas relative humidity is less than 100%. In addition, the critical water content (w100%RH) of mixed soils increases linearly with the increase of kaolin content; when the water contents of mixed soils are same, higher kaolin content induces higher pore gas relative humidity.
查看更多>>摘要：Bearings are the core components of rotating machinery and are vulnerable to failure. Early fault prediction is a significant and challenging task for bearing due to the weakness of fault signal. To address this issue, a method based on enhanced long short-term memory and ensemble empirical mode decomposition energy moment entropy is proposed in this paper. First, the ensemble empirical mode decomposition is adopted to process the raw vibration signals. Intrinsic mode functions with sensitive features are selected based on the correlation coefficient and maximal information coefficient. The energy moment entropy is designed as the performance index to depict bearing performance degradation. Secondly, an enhanced long short-term memory is developed by virtue of quantum superposition principle to predict the early fault of bearing. Finally, the effectiveness and the superiority of the proposed method are validated on the bearing datasets in comparison with other existed methods.
查看更多>>摘要：Spectral coherence (SCoh) is one of the most dedicated methods for characterizing the second-order cyclostationarity of bearing faults. Under the assumption of second-order cyclostationarity, the SCoh of the bearing vibration signal is expected to have non-zero values along lines-determined by fault-related cyclic frequencies. However, the bearing vibration signal produced by a complex mechanical system always exhibits mixed cyclostationarity-a combination of several orders of cyclostationarity, rather than pure second-order cyclostationarity. The fault-unrelated cyclic frequencies exhibiting a series of non-zero elements on the bivariate plane are served as a tricky problem to interfere with the fault identification. To address this issue, an enhanced SCoh is proposed by combining two basic operators designed by using median filtering and autocorrelation, respectively. Analysis of simulated signals, vibration datasets obtained from artificial fault bearing experiments, and accelerated bearing degradation tests are conducted to verify and compare the performance of the improved SCoh.
查看更多>>摘要：Aiming at the problems of safety, management, environmental protection, virtual sensor technology of dredger slurry concentration based on a hybrid ensemble deep learning (HEDL) framework is proposed. The purpose of this paper is to use the dredging construction big data, through the method of artificial intelligence, to deeply excavate the hidden relationship between the slurry concentration and the construction monitoring parameters in the construction process to generate virtual sensors, and then overcome some limitations of physical sensors. Firstly, this method removes the time lag effect of the monitoring data of physical sensors and then selects variables potentially related to the mud concentration. It makes full use of the advantages of each model to build a HEDL dredger slurry concentration prediction and measurement model embedded with multiple intelligent algorithms. The base learner of the model includes Deep Belief Network (DBN), Muti-Layer Perception (MLP), Convolutional Neural Networks (CNN), Gated Recurrent Neural Networks (GRU), Long Short-Term Memory (LSTM), Support Vector Regression (SVR). Finally, taking the Tianjin Port Channel Deepening Project as an applied research case, the accuracy and applicability of slurry concentration virtual sensors are verified.
Koklu, MuratUnlersen, M. FahriOzkan, Ilker AliAslan, M. Fatih...
10页
查看更多>>摘要：The main product of grapevines is grapes that are consumed fresh or processed. In addition, grapevine leaves are harvested once a year as a by-product. The species of grapevine leaves are important in terms of price and taste. In this study, deep learning-based classification is conducted by using images of grapevine leaves. For this purpose, images of 500 vine leaves belonging to 5 species were taken with a special self-illuminating system. Later, this number was increased to 2500 with data augmentation methods. The classification was conducted with a state-of-art CNN model fine-tuned MobileNetv2. As the second approach, features were extracted from pre-trained MobileNetv2 ' s Logits layer and classification was made using various SVM kernels. As the third approach, 1000 features extracted from MobileNetv2 ' s Logits layer were selected by the Chi-Squares method and reduced to 250. Then, classification was made with various SVM kernels using the selected features. The most successful method was obtained by extracting features from the Logits layer and reducing the feature with the Chi-Squares method. The most successful SVM kernel was Cubic. The classification success of the system has been determined as 97.60%. It was observed that feature selection increased the classification success although the number of features used in classification decreased.
查看更多>>摘要：Robotic automated assembly in large-scale space is a challenging task both in aerospace engineering and automobile manufacturing industries, which involves several technical metrics, such as accuracy, efficiency as well as stability of the assembly system. In this work, we propose a high-accuracy pose measurement system to tackle this problem. Compared to existing measurement technologies, our system can adequately cover the assembly space and reach a high pose accuracy (<0.2 degrees on rotation and <0.2mm on translation), which meets the requirements of most robotic automated assembly tasks in large-scale space (usually in sub-millimeter). To achieve that, we design a multi-degree-of-freedom measurement platform, which allows the camera to move in the motion space so as to find the best viewpoints to respectively measure the robot end-effector and the assembly target. Another key issue is the localization of the robot end-effector in the assembly space. Traditional calibration methods cannot be applied directly, since the robot and the camera are movable. Hence, we design a practical mechanical tooling which is attached to the sixth axis of the robot. Cubes are uniformly distributed on the tooling, serving for the precise localization of the end-effector. We also propose a novel calibration method to solve the kinematics solution. Moreover, we design a camera-pointcloud collaborative method to effectively compute the accurate 3D coordinates in camera space for sub-pixel marker centers, which can make a significant promotion in measurement accuracy.
查看更多>>摘要：The region-based active contour models could present difficulties because of undesired initial contour, noise distribution and image weak edges. In order to overcome the above problems, this paper proposes a superpixelbased via a local similarity factor and saliency (SLSFS). Firstly, the initial contour is generated by combining super-pixel and fuzzy c-means clustering. Secondly, the difference between local space and local intensity is used to improve the segmentation accuracy under noise. Finally, the weak edge information is protected by improved saliency detection. In addition, a gradient similarity constraint is used to remove the redundant regions. SLSFS model can generate adaptive initial contour around the target, and protect the weak edge information of the target on the premise of ensuring certain noise robustness. Experiments show that the average dice of SLSFS is 14% higher than that of the optimal comparison model and 19% on gray images.
查看更多>>摘要：Human gait extraction is a pilot step for teaching humanoid robot. A simple use system is needed to extract the Euler angle time series (EATS) of human lower-body. This paper constructs a multi-sensor (Kinect and OptiTrack) system for evaluating EATS extracted by Kinect. The 95% Confidence Interval and 5% Trimmed Mean are used to compare the magnitude difference of EATS during the whole gait cycle. The Pearson correlation coefficient (PCC) is utilized to assess the EATS relative consistency between the two sensors. The result shows that all the ankle EATS have large magnitude difference. The knee yaw and the ankle pitch of Kinect v2 are negatively correlated with that of OptiTrack. Since that, we present an Optimized Kinect Convolution Neural Network (OKCNN) to optimize the EATS. The performance of EATS extraction by Kinect improves greatly. And we prove that OKCNN can omit the complex preprocessing steps.
查看更多>>摘要：A large-scale stepped shaft is an important component in precision machinery manufacturing. Aiming at the model engineering of large-scale stepped shafts and improving the surface profile measurement accuracy of large-scale stepped shafts, a cylindrical profile measurement model containing seven systematic errors is designed, in which the eccentricity error, tilt error, sensor probe radius error, probe offset error, probe support rod tilt error, and horizontal and vertical rail tilt error are considered. Further, a multi-systematic errors separation method is proposed based on the seven-systematic errors in the cylindrical profile measurement model. First, the autocollimator and image processing are used to accurately extract the verification parameters, and then the stepwise estimation method and the equalisation optimizer (EO) are used to obtain the sectional and spatial parameters. The measurement experiment is based on the large-scale stepped shaft profile measurement device, and uses the coaxiality indicator to evaluate the profile measurement accuracy of the large-scale stepped shaft, and compares our model with the dual-systematic errors model and the five-systematic errors model. The experimental results show that compared with the other two methods, the coaxiality measurement accuracy of standard stepped shaft No. 1 can be increased by 26.03% and 15.99%, respectively; the cylindricity measurement accuracy of standard stepped shaft No. 1 can be increased by 16.03% and 9.17%, respectively; the coaxiality measurement accuracy of the No. 2 standard stepped shaft can be increased by 29.86% and 16.29%, respectively; the cylindricity measurement accuracy of standard stepped shaft No. 2 can be increased by 17.32% and 10.46%, respectively -- this verifies the effectiveness of our measurement method. The measurement method used in this study provides an accurate theoretical basis for the implementation of high-precision model engineering of cylindrical parts for national metrology institutions and key laboratories.
查看更多>>摘要：Pneumonia is the primary cause of death in children under the age of 5 years. Faster and more accurate laboratory testing aids in the prescription of appropriate treatment for children suspected of having pneumonia, lowering mortality. In this work, we implement a deep neural network model to efficiently evaluate pediatric pneumonia from chest radio graph images. Our network uses a combination of convolutional and capsule layers to capture abstract details as well as low level hidden features from the radio graphic images, allowing the model to generate more generic predictions. Furthermore, we employ transfer learning approach to extract spatial features from the raw input radio graph images, allowing the model to save resources while enhancing performance. The capsule layer weights of the network are updated using the dynamic routing algorithm. The proposed model is evaluated using benchmark pneumonia dataset Kermany et al. 2018, and the outcomes of our experimental studies indicate that the capsules employed in the network enhance the learning of disease level features that are essential in diagnosing pneumonia. According to our comparison studies, the proposed model with Convolution base from InceptionV3 attached with Capsule layers at the end surpasses several existing models by achieving an accuracy of 94.84%. The proposed model is superior in terms of various performance measures such as accuracy and recall, and is well suited to real-time pediatric pneumonia diagnosis, substituting manual chest radiography examination.
NSTL
Elsevier