A lattice self-reconfigurable modular soft robot based on the expansion-contraction motion rule is designed, which is composed of several soft modules, each of which is composed of a silica gel main body with positive hexahedron configuration and a master-slave docking surface. The internal bulged design makes it have a good expansion performance. The master-slave docking surface is composed of an iron disk and a suction disk type electromagnet connected with the silica gel main body by thread composition. Based on the relationship between the volume change of the soft module and the internal pressure, the expansion of the soft module is analyzed. The mapping relationship between the inflation pressure and the expansion of soft module is established. Besides, the inflation pressure required for the connection of adjacent two soft modules is obtained. Each soft module can expand 1.5 times under the working pressure of 30 kPa, and the docking and separation of two adjacent soft modules are realized by using the electromagnet connection and the expansion-contraction motion rules of soft modules. The self-reconfiguration of the modular soft robot can be realized by the sequential docking and separation of multiple adjacent modules. The feasibility of self-reconfiguration of soft robot is verified by the self-reconfiguration experiment.
Automatically extracting key data from annual reports is an important means of business assessments. Aimed at the characteristics of complex entities, strong contextual semantics, and small scale of key entities in the field of corporate annual reports, a BERT-BiGRU-Attention-CRF model was proposed to automatically identify and extract entities in the annual reports of enterprises. Based on the BiGRU-CRF model, the BERT pre-trained language model was used to enhance the generalization ability of the word vector model to capture long-range contextual information. Furthermore, the attention mechanism was used to fully mine the global and local features of the text. The experiment was performed on a self-constructed corporate annual report corpus, and the model was compared with multiple sets of models. The results show that the value of F1 (harmonic mean of precision and recall) of the BERT-BiGRU-Attention-CRF model is 93.69%. The model has a better performance than other traditional models in annual reports, and is expected to provide an automatic means for enterprise assessments.
When using the convolutional neural network (CNN) model to predict short-term traffic congestion, due to the convolution pooling operation of the model, part of the data for the information of the target position will be lost, resulting in the decline of the resolution of the output features and the decrease in the predictive ability of the model. To solve this problem, this paper proposes a dilated-dense neural network model. First, it uses dilated convolution to obtain the characteristics of a larger receptive field with fewer network parameters, and fully extracts complex and variable data spatio-temporal characteristics. Then, through down-sampling and equivalent mapping of dense network, it solves the problem of parameter degradation in the process of increasing layers of neural network. Finally, it uses the actual urban road average speed data blocks to verify the validity of the model. The results show that compared with the convolutional neural network model, the average absolute error of the network structure prediction is reduced by 3% to 23%.
Aimed at the problem of instability and deviation of multiple training model in limited samples, this paper proposes a method of distance metric learning based on the Gaussian mixture model, which can solve this problem more reasonably by dividing the dataset. Distance metric learning relies on the excellent feature extraction capabilities of deep neural networks to embed the original data into the new metric space. Then, based on the deep features, the Gaussian mixture model is used to cluster the analyzer and estimate the sample distribution in this new metric space. Finally, according to the characteristics of sample distribution, stratified sampling is used to reasonably divide the data. The research shows that the method proposed can better understand the characteristics of data distribution and obtain a more reasonable data division, thereby improving the accuracy and generalization of the model.
The nesting problem is how to nest objects of a specified shape in a two-dimensional space to obtain the maximum space utilization rate, which is of great significance in industrial production. The solution to the nesting problem requires frequent cross-checking of the objects to determine whether the nesting position is legal. The No Fit Polygon algorithm can be used to accelerate the procedure of cross-checking, but the algorithm cannot be used to calculate shapes containing curves, which limits its application. An algorithm based on orbit sliding can calculate No Fit Polygon between shapes which includes arc, but its effectiveness is not satisfactory. Aimed at this problem, and based on trajectory algorithm, the trajectory generation strategy and the profile algorithm are analyzed and improved. The improved algorithm can calculate the No Fit Polygon between shapes containing arc in a shorter time, solving both accuracy and effectiveness problems. Finally, the algorithm is tested in the real punch production process and the results of the test confirms the correctness and effectiveness of the algorithm.
In the dynamic and discrete ship block manufacturing process, lack of effective process resource organization and transparency in product processing leads to the problem of high cost and low efficiency for managers to acquire knowledge. A method for dynamic generation and updating of knowledge graph based on processing beat data flow is proposed. The definition of the processing beat data information model is defined by analyzing the processing flow and the station data characteristics of the ship blocks. The graph mapping steps, models, and fusion connection algorithms are proposed for static resources and processing beat data to realize the semantic association of station dynamic time series data and knowledge graphs. Based on the relationship between station process and product structure, the generation of workshop-level dynamic knowledge graph is realized. Taking the production process of a ship block as an example, the knowledge graph visualization prototype system is designed, developed, and verified. The results show that the proposed method is beneficial to the organization, acquisition, and reuse of knowledge in the process of ship block manufacturing.
Based on the problem that the oversampling method in the existing unbalanced classification problem cannot fully utilize the data probability density distribution, a method named latent posterior based generative adversarial network for oversampling (LGOS) was proposed. This method used variational auto-encoder to obtain the approximate posterior distribution of latent variable and generation network could effectively estimate the true probability distribution function of the data. The sampling in the latent space could overcome the randomness of generative adversarial network. The marginal distribution adaptive loss and the conditional distribution adaptive loss were introduced to improve the quality of generated data. Besides, the generated samples as source domain samples were put into the transfer learning framework, the classification algorithm of transfer learning for boosting with weight scaling (TrWSBoost) was proposed, and the weight scaling factor was introduced, which effectively solved the problem that the weight of source domain samples converge too fast and lead to insufficient learning. The experimental results show that the proposed method is superior to the existing oversampling method in the performance of common metrics.
In this paper, the static characteristics modeling and experimental study of pneumatic muscle fiber (PMF) are conducted. Considering the influence of end deformation, friction, and dead zone pressure on the static characteristics of PMF, a mathematical model of static characteristics of PMF is proposed. The static characteristics experimental platform is designed, and the static characteristics experiments of PMF and pneumatic muscle fiber bundles (PMFB) are performed, including the isometric experiment, the isotonic experiment, and the isobaric experiment. The static characteristics of PMF and PMFB with different specifications are compared and analyzed. Based on the isobaric characteristic curves obtained from the experiment, an experimental model of PMFB is proposed. Besides, the mathematical models of static characteristics of PMF and PMFB are identified by a large number of experimental data, which are in line with the actual situation. The study in this paper will lay the foundation for the precise control of micro bionic robot driven by PMF.
Aimed at the problems of navigation in distributed environment of a mobile robot, a regionalization vision navigation method based on deep reinforcement learning is proposed. First, considering the characteristics of distributed environment, the independent submodule learning control strategy is used in different regions and the regionalization model is built to switch and combine navigation control strategies. Then, in order to make the robot have a better goal-oriented behavior, reward prediction task is integrated into the submodule, and reward sequence is played back in combination with the experience pool. Finally, depth limitation is added to the primitive exploration strategy to prevent the traversal stagnation caused by collision. The results show that the application of reward prediction and depth obstacle avoidance is helpful to improve navigation performance. In the process of multi-area environment test, the regionalization model shows the advantages that the single model does not have in terms of training time and rewards, indicating that it can better deal with large-scale navigation. In addition, the experiment is conducted in the first-person 3D environment, and the state is partially observable, which is conducive to practical application.
Aimed at the problem that the traditional meta-path random walk in heterogeneous network representation cannot accurately describe the heterogeneous network structure and cannot capture the true distribution of network nodes well, a heterogeneous network representation method based on variational inference and meta-path decomposition is proposed, which is named HetVAE. First, combining with the idea of path similarity, a node selection strategy is designed to improve the random walk of the meta-path. Next, the variational theory is introduced to effectively sample the latent variables in the original distribution. After that, a personalized attention machanism is implemented, which weights the node vector representation of different sub-networks obtained by decomposition. Then, these node vectors are fused by the proposed model, so that the final node vector representation can have richer semantic information. Finally, several experiments on different network tasks are performed on the three real data sets of DBLP, AMiner, and Yelp. The effectiveness of the model is verified by these results. In node classification and node clustering tasks, compared with some state-of-the-art algorithms, the Micro-F1 and normalized mutual information (NMI) increase by 1.12% to 4.36% and 1.35% to 18% respectively. It is proved that HetVAE can effectively capture the heterogeneous network structure and learn the node vetcor representation that conforms more with the true distribution.
Aimed at the problem of group data anomaly detection with no data labels, a k-nearest neighbor (kNN) algorithm is proposed to detect group data anomalies in the unsupervised mode. In order to reduce false negatives and false positives caused by the mutual interference between abnormal and normal values, a reverse k-nearest neighbor (RkNN) method is proposed to filter the abnormal group data in reverse. First, the RkNN algorithm uses statistical distance as the similarity measure between different groups of data. Then, the anomaly scores of each group and the initial abnormality are obtained by using the kNN algorithm. Finally, the initial abnormality is filtered by using the RkNN method. The experiment results show that the algorithm proposed can not only effectively reduce the false negatives and false positives, but also has a high anomaly detection rate and good stability.
In view of the shortcomings of the traditional video anomaly detection model, a network structure combining the fully convolutional neural (FCN) network and the long short-term memory (LSTM)network is proposed. The network can perform pixel-level prediction and can accurately locate abnormal areas. The network first uses the convolutional neural network to extract image features of different depths in video frames. Then, different image features are input to memory network to analyze semantic information on time series. Image features and semantic information are fused through residual structure. At the same time, the skip structure is used to integrate the fusion features in multi-mode and upsampling is conducted to obtain a prediction image with the same size as the original video frame. The proposed model is tested on the ped 2 subset of University of California, San Diego (UCSD) anomaly detection dataset and University of Minnesota System(UMN)crowd activity dataset. And both two datasets achieve good results. On the UCSD dataset, the equal error rate is as low as 6.6%, the area under curve reaches 98.2%, and the F1 score reaches 94.96%. On the UMN dataset, the equal error rate is as low as 7.1%, the area under curve reaches 93.7%, and the F1 score reaches 94.46%.
In view of the lack of standardized test support for existing triboelectric research, a system suitable for the triboelectric performance test was independently developed. Through modular design and system integration, the three modules of loading, motion, and measurement control were designed in turn. The LabVIEW measurement control software was developed. By adding buffer springs and optimizing the loading structure, the stable loading under small load conditions was achieved. In addition, based on the insulation treatment of the upper and lower samples, the accurate measurement of micro currents was achieved. The reliability of the test system was verified by benchmarking test and friction electrification test, and the transferred charge of Cu-Al friction interface was analyzed. Moreover, the linear relationship between transferred charge and load was preliminarily determined. The results are beneficial to the standardization of tribo-electrical testing.
A method for constructing a novel offshore platform was proposed. Based on the cooperative control of multiple unmanned vessels, a self-assembling platform was realized, which could be reconfigured into different shapes according to requirements. A docking controller was designed to realize the docking of two modules. A connecting rod and electromagnetic forces were adopted to complete the docking and reduce the difficulty. Besides, a test site was constructed using the conditions of pool, and a model test was performed to verify the functions of the proposed self-assembling platform. The results show that the design can realize the docking of any unmanned vessel. Compared with the single platform, this self-assembling platform can perform more complex tasks, whose decentralized design makes it more flexible and reliable.
Traditional process monitoring methods ignore the time-series correlation between variables, and do not distinguish the dynamic relationship and static relationship between variables, resulting in poor monitoring effect. To solve these problems, a dynamic-static joint indicator monitoring method of batch process based on global slow feature analysis(GSFA)-global neighborhood preserving embedding (GNPE) is proposed in this paper, which can effectively extract dynamic global features and static global features. First, the dynamic and static characteristics of the process variables are evaluated. Variables with weak autocorrelation and cross-correlation are regarded as static variables, and the remaining variables are regarded as dynamic ones. Next, the GSFA and GNPE models are constructed for dynamic and static subspaces, respectively. Finally, the statistical information from each subspace is combined by using Bayesian inference to obtain the joint indicator of the mixed model to realize process monitoring. Finally, the proposed algorithm is applied to a numerical example and the penicillin fermentation simulation process for simulation verification. The results show that the proposed GSFA-GNPE algorithm has better fault detection effects than other algorithms.
Aimed at the problem of high dimension and nonlinearity of variable data in chemical process, a process fault detection algorithm based on neighborhood preserving embedding(NPE )-principal polynomial analysis (PPA) is proposed in this paper. The NPE algorithm is used to extract low dimensional submanifolds of high dimensional data, which overcomes the problem that the traditional linear dimensionality reduction algorithm cannot extract local structure information, so as to reduce the dimensions. The PPA method is used to describe data by a set of flexible principal polynomial components, which can effectively capture the inherent nonlinear structure of process data. The principal polynomial analysis is conducted in the reduced manifold space, and Hotelling’s T2 and square prediction error statistical models are established to determine the control limit for fault detection. Finally, compared with the traditional kernel principal component analysis and the PPA method, a group of nonlinear numerical examples and Tennessee Eastman chemical process data experiments are performed to verify the effectiveness and superiority of the NPE-PPA algorithm.
Combined with the current research status of the intelligent vehicle decision-making methods at home and abroad, this paper classifies and summarizes decision-making methods from four aspects: decision input and output, environment interaction, and algorithm types. Besides, it analyzes their advantages and disadvantages, and evaluates applicable scenarios. Moreover, it surveyes the common data sets and current evaluation standards which are used for decision-making researches. Furthermore it discusses the technical difficulties faced by current decision-making methods and future development trends.
The wireless powered intestinal robot transmits the intestinal images taken by the image acquisition system to the external upper computer for diagnosis. However, the image transmission process will be interfered by the circuit structure and external environment, leading to noise in the collected images. Therefore, an image denoising algorithm based on non-subsampled contourlet transform (NSCT) is proposed to reduce the noise of the images collected by intestinal robots. First, histogram equalization pretreatment is adopted to improve the brightness and contrast of intestinal noise images. Next, NSCT transformation is performed on intestinal noise images and a residual network model is constructed to reduce the noise of frequency domain information after transformation. Finally, the denoised image is reconstructed by NSCT inverse transformation. The results show that the proposed algorithm can effectively reduce the influence of intestinal noise in complex environments, and better maintain the visual effect of the image. Compared with other intelligent algorithm models, both subjective and objective noise reduction effects are improved, with peak signal to noise ratio (PSNR) improved by 1.35 to 3.45 dB and structural similarity index measure (SSIM) improved by 0.0083 to 0.0252.
The rolling restraint system between two space rigid bodies is a typical non-holonomic system. The incomplete characteristics can be used to simplify the mechanical structure and improve the reliability of the system. Aimed at the problems that the state variables of the pure rolling constraint non-holonomic system are difficult to control, the existing control methods are limited to specific models, and there is a lack of online control research, a solution method suitable for the online motion planning of the general rolling constraint system is established based on the rolling constraint first-order motion model. First, the offline motion planning is achieved by using the collocation method to obtain the reference trajectory. Then, the sequential action control (SAC) algorithm is used in real-time control combined with the rolling optimization framework to realize the online motion planning of the rolling system. The algorithm is applied to the real-time motion planning of the ball-plane rolling model and the rolling model between two spheres. The simulation results show that the method has a practical application value in broadening the control of the spherical robot and the operation of the dexterous manipulator.
Combining the flexible robot driven by intelligent materials with the octahedral variable geometry truss system, and based on the theory of octahedral variable geometry truss system, a single flexible unit is designed and driven by parallel shape memory alloy (SMA) spring. The kinematic model is established by utilizing the geometric method. The kinetic energy, the elastic potential energy, and the gravitational potential energy of the flexible manipulator are analysed. The general dynamics equation is established based on the Lagrangian dynamics. The driving force of a single flexible element of SMA spring is calculated by utilizing MATLAB. Adams simulation is performed, and the calculation results are in good agreement with the theoretical calculation results. Finally, a prototype of the flexible control arm is built and the rotation angles at different currents are measured. Modeling and simulation methods are useful references for other types of robots.
Existing decision-making methods for intelligent vehicles do not consider factors such as the right of way information, polite driving of the vehicle, and limited perception range of the vehicle, which may easily lead to safety hazards in merging scenarios. Aimed at these problems, a Stackelberg-game-based decision-making method is proposed. This method constructs a game model combining the right of way and conducts parametric modeling of the merging scenarios. Then, the cooperation factor is introduced to design the corresponding profit function. Finally, the vehicle decision-making solution framework is designed to achieve the maximum value of decision-making benefits in this scenario. The experimental results illustrate that the proposed method can effectively improve the accuracy of vehicle decision-making behavior prediction on the datasets and the decision-making robustness in a high traffic density environment.
Due to the high cost and time-consumption of artificial semantic tags, domain-based adaptive semantics segmentation is very necessary. For scenes with large gaps or pixels, it is easy to limit model training and reduce the accuracy of semantic segmentation. In this paper, a domain adaptive semantic segmentation network (DA-SSN) using the improved transformation network is proposed by eliminating the interference of large gap pictures and pixels through staged training and interpretable masks. First, in view of the problem of large domain gaps from some source graphs to target graphs and the difficulty in network model training, the training loss threshold is used to divide the source graph dataset with large gaps, and a phased transformation network training strategy is proposed. Based on the ensurance of the semantic alignment of small gap source images, the transformation quality of large gap source images is improved. In addition, in order to further reduce the gap between some pixels in the source image and the target image area, an interpretable mask is proposed. By predicting the gap between each pixel in the source image domain and the target image domain, the confidence is reduced, and the training loss of the corresponding pixel is ignored to eliminate the influence of large gap pixels on the semantic alignment of other pixels, so that model training only focuses on the domain gap of high-confidence pixels. The results show that the proposed algorithm has a higher segmentation accuracy than the original domain adaptive semantic segmentation network. Compared with the results of other popular algorithms, the proposed method obtains a higher quality semantic alignment, which shows the advantages of the proposed method with high accuracy.
Aimed at the mission planning for cleaning photovoltaic panels in large-area photovoltaic plants with mobile cleaning robots, a district planning strategy is hereby proposed. The photovoltaic plants, considering the position of wind gaps, the illumination time, and other environmental factors, adopt a hierarchical mission planning based on the cleaning priority, and use the Hamilton graph to turn the cleaning problem of photovoltaic panels into a travelling salesman problem (TSP). Considering the disadvantages of low efficiency and early convergence of the genetic algorithm, an improved genetic algorithm, which includes the hybrid selection operator combining the tournament selection with the roulette wheel selection and the crossover operator based on the segmentation rule is thus put forward. The improved genetic algorithm is applied to plan the cleaning order of robots to clean the photovoltaic panels. The experimental results show that in comparison with the adaptive genetic algorithm, the improved genetic algorithm has a higher efficiency and better results.
In order to improve the flexibility and response efficiency of warehouse dispatching, a cascaded improved differential evolution algorithm is proposed to construct the allocation of goods with the picking trolley running time, shelf stability, and inventory capacity as resource conditions. The maximum completion time for each item in the batch order assigned to the optimal location of the corresponding partition is the two-level target model that is re-batch-allocated for the conditional order. The Lagrangian interpolation algorithm is integrated into the improved algorithm of the standard differential evolution algorithm to solve the two-level target model, and the two-level solution process is cascaded to complete the cascaded differential evolution algorithm to solve the multi-order batch allocation problem. Based on the adaptive adjustment of differential evolution parameters, the improved differential evolution algorithm combines the local search ability of Lagrangian interpolation to optimize the differential evolution algorithm, and uses local and global switching factors to dynamically adjust the evolution direction and improve the convergence performance of the algorithm. The improved differential evolution algorithm is applied to solve the problem of multi-order batch allocation. The experimental results show that the improved algorithm optimization results are better than the particle swarm optimization algorithm, the genetic algorithm, and the standard differential evolution algorithm, which reduces the maximum completion time of each batch of orders and effectively balance the workload.
In order to realize the smooth control of vehicle by unmanned driving robot (UDR) in paths with different curvatures, a control strategy for UDR based on multi-objective fuzzy decision is proposed. First, the integrated dynamics models of the driving robot and vehicle are established. Then, a yaw rate generation method and a multi-objective fuzzy decision coordinated manipulation strategy are established. The yaw rate generation method generates the reference yaw rate according to the speed and path required by the test while the multi-objective fuzzy decision coordinated manipulation strategy generates sets of target speeds and target yaw rates according to the current speed. Finally, decisions are made on the scheme in the set under multiple constrains. The best scheme is chosen as the target speed and target yaw rate of the next moment. The test and simulation results demonstrate the effectiveness of the proposed strategy.
In view of feature redundancy in the convolutional neural network, the concept of orthogonal vectors is introduced into features. Then, a method for orthogonal features extraction of convolutional neural network is proposed from the perspective of enhancing the differences between features. By building the structure of parallel branches and designing the orthogonal loss function, the convolution kernels can extract the orthogonal features, enrich the feature diversity, eliminate the feature redundancy, and improve the results of classification. The experiment results on one-dimensional sample dataset show that compared with the traditional convolution neural network, the proposed method can supervise the convolution kernels with different sizes to mine more comprehensive information of orthogonal features, which improves the efficiency of convolutional neural network and lays the foundation for subsequent researches on pattern recognition and compact neural network.
In order to improve the life prediction effect of elevator brake in the real working environment, an unsupervised deep transfer learning (UDTL) method based on long short-term memory encoder-decoder (LSTM-ED) was proposed. The simulation data were used to analyze the health status of brake when it was working. First, the LSTM-ED and the fully connected network were initially trained through the source domain data. Then, the LSTM-ED was used as a feature extractor to map the simulated and actual data to the feature space, and the maximum mean discrepancy was adopted to achieve data alignment. Finally, the target domain data in the feature space was regressed through the fully connected network to predict the remaining useful life (RUL) of the real brake. In the training phase, a step-by-step training method was used to ensure the accuracy of a single module. The validity was verified by comparing the experimental simulation data with the real working data in the elevator tower. The results show that by introducing the transfer learning and step-by-step training methods, the proposed method can reduce the mean square error of RUL prediction to 0.0016, and can achieve accurate RUL prediction of elevator brakes in real working environment.
