查看更多>>摘要：Lightning is a strong discharge phenomenon that occurs in nature and poses a great threat to people's property and life safety. The generation of lightning originates from the continuous accumulation of electric charges in clouds, and the atmospheric electric field instrument, as a measurement device reflecting the most fundamental cause of lightning generation, is used to detect the occurrence of lightning, which has been very widely used due to its low price and easy installation. However, its detection results are directionless and the detection range is limited. Therefore, this paper proposed a method for spatio-temporal localization of lightning based on deep Long Short-Term Memory (LSTM) neural network and interpolation method. The time series data of electric field detected by 30 atmospheric electric field instruments was fed into deep LSTM network for training, and the prediction results were classified into five categories according to the time period of lightning occurrence by softmax function. Furthermore, data from the networked stations were interpolated using ordinary Kriging (OK) to obtain the electric potential distribution in Guangzhou city, which was used to infer the approximate area where lightning may occur. The above two algorithms passed the accuracy test respectively. Finally, two case studies were done based on LSTM-OK. The results show that it can obtain satisfactory prediction performance.
查看更多>>摘要：Tool wear condition monitoring (TCM) is an important part of machining automation. In recent years, deep learning (DL) based TCM methods have been widely researched. However, almost DL-based methods need sufficient learning samples to obtain good accuracy, which is hard for TCM in terms of cost and time. In order to enhance the recognition accuracy of DL-based TCM under small samples, this paper proposed a new improved multi- scale edge-labeling graph neural network (MEGNN). Each channel signal of a cutting force sensor is expanded to multi- dimensional data through phase space reconstruction. Then, these multi- dimensional data are encoded into a gray recurrence plot (RP), and aggregated into a color RP, which is input to MEGNN to extract features for establishing a fully connected graph. Finally, the tool wear condition is estimated through the updated edge labels using a weighted voting method. Applications of the proposed MEGNN- based method to PHM 2010 milling TCM dataset and our experiments demonstrate it outperforms three DL-based methods (CNN, AlexNet, ResNet) under small samples.
查看更多>>摘要：Highly precise measurement of depth in micro meter is very important in many fields. Many methods measuring depth for spherical objects have been proposed. However, few of them can measure the depth for plane objects, such as mu LED, which has advantages in brightness, power consumption and response speed. The depth measurement is one of the key technology for the transfer printing of mu LED. In this paper, we proposed a method measuring depth for mu LED array using single camera. The proposed method was calibrated by a precision stage. The uncertainties, such as camera distortion, camera tilt, curve fitting and surface roughness of mu LED were estimated. The results showed that the relative standard uncertainty in depth position was about 2.2%. This showed that the proposed system was very stable and repeatable.
查看更多>>摘要：Plant conditions can be monitored by direct-leaf measurements. For this purpose, several researchers have developed applications based on smartphone cameras, but the variable quality of smartphone cameras, even those of the same brand, produces non-standardized results. The present study describes a new and reliable technique that measures the chlorophyll content of maize leaves for plant monitoring on any smartphone. To check the accuracy of the data generated by the developed handheld optical sensing system, the obtained data were compared against those of an established chlorophyll-monitoring meter (a SPAD-502 monitor). The required SPAD, RGB, and global positioning system data were collected from maize fields (similar to 2 ha) at the research farm of the Indonesian Coffee and Cocoa Research Institute. The collected data were analyzed using conventional statistics/regression analysis and a deep neural network (DNN). The inverse-distance weighted outputs were interpolated to generate zoning maps. In a conventional statistics/regression analysis, the chlorophyll levels were significantly correlated with the SPAD values (R-2 = 0.82-0.84, root mean squared error [RMSE] = 2.95-3.05). However, after applying the DNN with 12 extracted input features, four hidden layers, and 637 parameters, the chlorophyll content estimation was significantly improved (R-2 and RMSE = 0.89 and 2.6, respectively), and the zoning map generated by the developed system was nearly aligned with the SPAD zoning map. The findings confirm that this technique is applicable to all types of smartphones regardless of their camera properties and provides the light-aided intensity necessary for direct-leaves measurement. The locations of the collected leaves were simultaneously monitored in real-time to generate the mapping results.
查看更多>>摘要：With the increase of the scale and complexity of mechanical equipment, traditional intelligent fault diagnosis (IFD) based on shallow machine learning methods is unable to meet the demand of coupling faults. In the past decades, the vigorous development of deep learning (DL) brings new opportunities for IFD, especially the representation learning based on Autoencoder (AE) theory has been widely applied. To provide a more comprehensive reference, the theoretical foundations of multi-type AEs and the training method of stacked autoencoder (SAE) are briefly introduced. Then the application advances of AE are reviewed from optimization and combination aspects, which are aiming at improving the representation learning ability. To provide ways for the application of AE-based methods, two typical study cases for ideal and complex engineering systems are illustrated respectively. Finally, the challenges and prospects of AE-based representation learning are reported from four aspects, which give a guidance for the future research direction.
查看更多>>摘要：In this paper, an improved convolutional neural network (CNN) named Feature Selection CNN (FS-CNN) and Synthetic Minority Over-sampling Technique (SMOTE) named Multiply Stochastic SMOTE (MS-SMOTE) are proposed for imbalanced ECG multi-factor classification. FS-CNN includes convolutional expression layer, multi residual block, and decision layer. The primary purpose of FS-CNN is to find the most influential features during automatic model optimization dynamically. MS-SMOTE combines the advantage of SMOTE and borderline SMOTE to balance the number of different categories. By dynamically selecting SMOTE and borderline SMOTE based on the proportion of data, the synthetic data can have a better category property. Comparing to other wide used algorithms, our method is better than most other algorithms in many indicators. And in simulation 2, our method can deal with a large number of ECG multi-factor data classifications. Also, two self-controlled experiments are designed to examine how different parameters affect the result of the above problem.
查看更多>>摘要：Accurate recognition of coal and gangue is the core technology to realize intelligence in fully mechanized top coal caving mining. The influence of emissivity of the thermal imager on the efficiency of the coal/gangue recognition is considered in this paper. The temperature of coal and gangue are measured by the infrared thermal imager with different emissivity. The results show that: compared with high emissivity, low emissivity conditon can enlarge the measured temperature difference between coal and gangue, which will improve the recognition efficiency greatly. The maximum temperature difference reaches 4.5 degrees C-4.8 degrees C in liquid intervention experiment with low emissivity (epsilon = 0.1-0.3). Setting the emissivity of infrared thermal imager at a low level may be a new approach to widen the temperature difference between coal and gangue, which will improve the recognition efficiency greatly. When epsilon = 0.3, the maximum temperature difference ratio reaches 9.0, which shows the best potential for improving coal/gangue recognition efficiency with infrared thermal imager at low emissivity.
查看更多>>摘要：This research examined the potential of a pushbroom near infrared hyperspectral imaging (NIR-HSI) system (900-1600 nm) for ripening stage (unripe, ripe, and overripe) classification based on the days after anthesis (DAA) and dry matter (DM) prediction of durian pulp. The performance of five supervised machine learning classifiers was compared including support vector machines (SVM), random forest (RF), linear discriminant analysis (LDA) partial least squares-discriminant analysis (PLS-DA), and k-nearest neighbors (kNN) for the ripening stage classification and a partial least squares regression (PLSR) model was developed for the DM prediction. The classification and regression models were developed and compared using the full and selected wavelengths by genetic algorithms (GA) and principal component analysis (PCA). For classification, LDA showed the best result with a test accuracy of 100% for both full wavelength and selected 135 wavelengths by GA. A total of 11 wavelengths selected from PCA achieved a test accuracy of 93.6% by LDA. The PLSR models predicted the DM with the coefficient of determination of prediction (Rp2) greater than 0.80 and a root mean square error of prediction (RMSEP) less than 1.6%. The results show that NIR-HSI has the potential to identify ripeness correctly, predict the DM and visualize the spatial distribution of durian pulp. This approach can be implemented in the packaging firms to solve the problems related to uneven ripeness and to inspect the quality of durian based on DM content.
查看更多>>摘要：When detecting an underground target with a portable transient electromagnetic system, the system excites and records digitized data on a 5 x 5 or 7 x 7 grid above the target, and with the positioning system to accurately record the sensor positions, which makes the detection more complex and less efficient. To simplify the detection process, a novel three-position measurement method without positioning system is proposed to estimate the target characteristics. First, a new forward model based on the relative position is established, and the relative position is estimated by differential evolution algorithm at each measurement. Then, the proposed method is optimized to quickly and precisely characterize the target. Results show that the horizontal and depth errors are less than 5 and 9 cm, respectively. The proposed method takes about 28.4 s to characterize the target, only 11% of the traditional one, which substantially improves the detection efficiency and flexibility of the portable sensor.
查看更多>>摘要：Evaluation of fabric shape retention is an indispensable part of fabric performance evaluation. The existing evaluation methods for investigating the wrinkle recovery, drape, stiffness of fabrics are very mature, but a comprehensive and intuitive evaluation of fabric shape retention is not achieved. Based on the previously proposed fabric shape retention testing device and image processing method, this paper proposes a method to fit fabric crease contour lines to Gaussian curves to optimize the fabric crease contour lines and describe the fabric crease shape more intuitively. The indexes, which are the vertex angle (VA), the vertex height (VH), the maximum curvature (MC), and the enclosed area (EA), can be extracted from the Gaussian curves of the stable state to characterize different shape retention properties of the fabric. By comparing the test results of the standard method, it was found that VA and MC can form a linear regression relationship with the crease recovery angle to describe the crease recovery property of the fabric. VH, MC and EA have linear regression relationships with static drape coefficient or bending length to describe the drape and stiffness of the fabric. Compared with previous studies, this article enables to inspect the fabric crease recovery process dynamically, depict the crease recovery shape visually and evaluate the fabric shape retention performance comprehensively and accurately. This paper provides an experimental basis for investigating the evaluation method of fabric shape retention performance.
NSTL
Elsevier