Most of the compound fault diagnosis methods regard the compound fault as a new single fault type, ignoring the interaction of internal single faults, and the fault analysis is vague in granularity and poor in interpretation. An improved Transformer-based compound fault decoupling diagnosis method was proposed for industrial environments with very little compound fault data. The diagnosis process included pre-processing, feature extraction and fault decoupling. With introducing the decoder of the Transformer, the cross-attention mechanism enables each single fault label to adaptively in the extracted feature layer focus on the discriminative feature region corresponding to the fault feature and predicts the output probability to achieve compound fault decoupling. Compound fault tests were designed to verify the effectiveness of the method compared with the advanced algorithms. The results showed that the proposed method had high diagnostic accuracy with a small number of single fault training samples and a very small number of compound fault training samples. The compound fault diagnosis accuracy reached 88.29% when the training set contained only 5 compound fault samples. Thus the new method has a significant advantage over other methods.

The dense small target detection algorithm LSA_YOLO based on YOLOv5s for UAVs with complex backgrounds and multiples of small targets with dense distribution was proposed for UAV images. A multi-scale feature extraction module LM-fem was constructed to enhance the feature extraction capability of the network. A new hybrid domain attention module S-ECA relying on multi-scale contextual information has been put forward and a algorithm focus on target information was established aiming to suppress the interference of complex backgrounds. The adaptive weight dynamic fusion structure AFF was designed to assign reasonable fusion weights to both shallow and deep features. The capability of algorithm in detecting dense small targets in complex backgrounds was improved given the application of S-ECA and AFF in the structure of PANet. The loss function Focal-EIOU was utilized instead of the loss function CIOU to accelerate model detection efficiency. Experimental results on the public dataset VisDrone2021 public dataset show that the average detection accuracy for all target classes improves from 51.5% for YOLOv5s to 57.6% for LSA_YOLO when the set input resolution is set to 1 504 × 1 504.

A multi-agent reinforcement learning algorithm based on priority experience replay and decomposed reward function was proposed in multi-agent pursuit and evasion games. Firstly, multi-agent twin delayed deep deterministic policygradient algorithm (MATD3) algorithm based on multi-agent deep deterministic policy gradient algorithm (MADDPG) and twin delayed deep deterministic policy gradient algorithm (TD3) was proposed. Secondly, the priority experience replay was proposed to determine the priority of experience and sample the experience with high reward, aiming at the problem that the reward function is almost sparse in the multi-agent pursuit and evasion problem. In addition, a decomposed reward function was designed to divide multi-agent rewards into individual rewards and joint rewards to maximize the global and local rewards. Finally, a simulation experiment was designed based on DEPER-MATD3. Comparison with other algorithms showed that DEPER-MATD3 algorithm solved the over-estimation problem, and the time consumption was improved compared with MATD3 algorithm. In the decomposed reward function environment, the global mean rewards of the pursuers were improved, and the pursuers had a greater probability of chasing the evader.

The improved YOLOv5 object detection algorithm was used to detect the facial region of the driver and a multi-feature fusion fatigue state detection method was established aiming at the problem that existing fatigue state detection method cannot be applied to drivers under the epidemic prevention and control. The image tag data including the situation of wearing a mask and the situation without wearing a mask were established according to the characteristics of bus driving. The detection accuracy of eyes, mouth and face regions was improved by increasing the feature sampling times of YOLOv5 model. The BiFPN network structure was used to retain multi-scale feature information, which makes the prediction network more sensitive to targets of different sizes and improves the detection ability of the overall model. A parameter compensation mechanism was proposed combined with face keypoint algorithm in order to improve the accuracy of blink and yawn frame number. A variety of fatigue parameters were fused and normalized to conduct fatigue classification. The results of the public dataset NTHU and the self-made dataset show that the proposed method can recognize the blink and yawn of drivers both with and without masks, and can accurately judge the fatigue state of drivers.

To reliably and accurately predict the axial bearing capacity of concrete-filled steel tube (CFST) columns, a prediction model of CFST column axial bearing capacity with ensemble machine learning was developed and explained. The quality of the CFST column database was evaluated using the Mahalanobis distance, the prediction model of CFST column axial bearing capacity was established by the extreme gradient boosting (XGBoost) algorithm, and the optimal hyperparameter combination of the model was found using the K-Fold cross-validation (K-Fold CV) and the tree-structured Parzen estimator (TPE) algorithms. The predicted values of the optimized XGBoost model were compared with the calculated values of the existing methods and the unoptimized XGBoost model using different evaluation metrics. The Shapley additive explanations (SHAP) approach was used to produce both global and local explanations for the predictions of XGBoost model. Results show that, after hyperparameter tuning, the XGBoost model’s performance surpasses performance of relevant standards and empirical formulas, and the SHAP approach can effectively explain the XGBoost model’s output.

A point cloud instance segmentation model with a k-nearest neighbors (KNN) module featuring attention mechanism and an improved associatively segmenting instances and semantics (ASIS) module was proposed to address the problems of discrete segmentation and insufficient feature utilization in traditional 3D convolution-based algorithms. The model took voxels as input and extracted point features through sparse convolution of 3D submanifolds. The KNN algorithm with attention mechanism was used for reorganizing the features in the semantic and instance feature space to alleviate the problem caused by the quantization error of extracted features. The reorganized semantic and instance features were correlated through the improved ASIS module to enhance the discrimination between point features. For semantic features and instance embedding, the softmax module and the meanshift algorithm were applied to obtain semantic and instance segmentation results respectively. The public S3DIS dataset was employed to validate the proposed model. The experimental results showed that the instance segmentation results of the proposed model achieved 53.1%, 57.1%, 65.2% and 52.8% in terms of mean coverage (mCoV), mean weighted coverage (mWCov), mean precision (mPrec) and mean recall (mRec) for the instance segmentation. The semantic segmentation achieved 61.7% and 88.1% respectively in terms of mean intersection over union (mIoU) and Over-all accuracy (Oacc) for the semantic segmentation. The ablation experiment verified the effectiveness of the proposed modules.

Most of existing structural image super-resolution reconstruction algorithms can only solve a specific single type of structural image super-resolution problem. A structural image super-resolution network based on improved Transformer (TransSRNet) was proposed. The network used the self-attention mechanism of Transformer mine a wide range of global information in spatial sequences. A spatial attention unit was built by using the hourglass block structure. The mapping relationship between the low-resolution space and the high-resolution space in the local area was concerned. The structured information in the image mapping process was extracted. The channel attention module was used to fuse the features of the self-attention module and the spatial attention module. The TransSRNet was evaluated on highly-structured CelebA, Helen, TCGA-ESCA and TCGA-COAD datasets. Results of evaluation showed that the TransSRNet model had a better overall performance compared with the super-resolution algorithms. With a upscale factor of 8, the PSNR of the face dataset and the medical image dataset could reach 28.726 and 26.392 dB respectively, and the SSIM could reach 0.844 and 0.881 respectively.

In order to build an aviation assembly knowledge graph composed of assembly process information, assembly technology knowledge, related industry standards and internal connections of the three, a named entity recognition technology framework based on continual learning was proposed. The characteristic of the proposed framework was that it maintained high recognition performance throughout the progressive learning process from zero corpus to large-scale corpus, without relying on manual feature setting. A comparative performance experiment of the proposed framework was carried out in practical industrial scenarios, the experiment proceeded from general assembly and component assembly, and the manipulations of the pull rod and cable installation were regard as a specific experimental case. Experimental results show that the proposed framework is significantly better in accuracy, recall, and F1 value than previous algorithms, while handling different-scale corpus environments. And the credible results for named entity recognition tasks can be provided consistently by the proposed framework in the aviation assembly domain.

An optimal design problem of long span steel-concrete composite floor system was investigated based on important parameters to aim at the economical and applicable conditions and optimizing orientations of long span steel-concrete composite floor. The objective function was set as economical equivalent steel consumption, and the variables contained eight parameters including dimensions of the steel section, intermediate distance between steel sections and thickness of concrete slab. The objective function was constrained to the plastic theory, standards and construction experience. The generalized reduced gradient method (GRG) was used to generate optimal sections with minimum economical equivalent steel consumption under different spans and live loads. According to the optimization results, the composite floor within a span of 60 m and a variable load of 6 kN/m² could efficiently facilitate the composite effect of composite structure and has good economic benefits. As for super-long span composite floor when the traditional I-section floor system is not economically suitable, the composite floor with corrugated web is recommended for improving the bearing efficiency of steel web. A new cable-supported composite floor system was proposed based on the cable-supported beam for its high efficiency in mechanism.

Existing knowledge representation methods only consider triplet itself or one kind of additional information, and do not make use of external information to semantic supplement knowledge representation. The convolutional neural network was used to extract feature information from text. The convolutional neural network based on attention mechanism was used to distinguish the feature reliability of different relationships, enhance the representation of entity relationship structure vector in the existing knowledge graph and obtain rich semantic information. A weighted hierarchical encoder which combined all the hierarchical type projection matrix of the entity with the relationship-specific type constraints, was used to construct the projection matrix of the hierarchical type, The link prediction and the triplet classification were performed on WN18, WN18RR, FB15K, FB15K-237 and YAGO3-10 datasets to analyze and verify the validity of the proposed model. The experiment showed that in the entity prediction experiment, the proposed model reduced the MeanRank(Filter) by 11.8% compared to the TransD model, and increased Hits@10 by 3.5%. In the triple classification experiment, the classification accuracy of the proposed model was 8.4% higher than the DKRL model and 8.5% higher than the TKRL model, which fully proved that the ability of knowledge representation could be improved by using external multi-source information.

A semantic segmentation network was designed for underwater images. A lightweight and efficient encoder-decoder architecture was used by considering the trade-off between speed and accuracy. Inverted bottleneck layer and pyramid pooling module were designed in the encoder part to efficiently extract features. Feature fusion module was constructed in the decoder part in order to fuse multi-level features, which improved the segmentation accuracy. Auxiliary edge loss function was used to train the network better aiming at the problem of fuzzy edges of underwater images, and the edges of segmentation were refined through the supervision of semantic boundaries. The experimental data on the underwater semantic segmentation dataset SUIM show that the network achieves 53.55% mean IoU with an inference speed of 258.94 frames per second on one NVIDIA GeForce GTX 1080 Ti card for the input image of pixel 320×256, which can achieve real-time processing speed while maintaining high accuracy.

A hierarchical federated learning framework based on wireless device-to-device (D2D) networks was proposed to solve the problem of large communication resource consumption and limited device computing resources faced by deploying federated learning in wireless networks. Different from the traditional architectures, the hierarchical aggregation was adopted for model training. The architecture performed the intra-cluster aggregation through D2D networks, and each cluster performed the decentralized training at the same time. A cluster head was selected from each cluster to upload the model to the server for global aggregation. The network traffic of the central node was reduced by combining the hierarchical federated learning and decentralized learning. The degree of the vertices in the D2D networks was used to measure the model convergence performance. The head selection and bandwidth allocation were jointly optimized by maximizing the total degree of selected cluster heads. An optimization algorithm based on dynamic programming was designed to obtain the optimal solutions. The simulation results show that compared with the baseline algorithm,the framework can not only effectively reduce the frequency of global aggregation and training time, but also improve the performance of the final model.

The main input characteristics of the existing wind power prediction models include meteorological data and power data. Aiming at the difficulties such as obtaining high-precision meteorological data, expressing the potential relationships between data, and slow convergence of the prediction models, the work proposed new ultra-short-term wind power prediction method which could extend the financial factors based on extreme gradient boosting regression tree algorithm (XGBoost). A prediction model which could derive financial factors based on wind power timing data was proposed. The application of XGBoost algorithm with high prediction accuracy and fast training speed in prediction could help the prediction model converge quickly. The relevant experiment was conducted for verification with the wind power data set of a wind farm in Inner Mongolia of China and that of Tennet in Germany. The results showed that taking R² score as an example, the proposed method improved by about 14.71% compared with the benchmark method. The total time for modeling and prediction of the proposed method did not exceed 500 ms.

A comprehensive evaluation and categorization of text-to-image generation tasks were conducted. Text-to-image generation tasks were classified into three major categories based on the principles of image generation: text-to-image generation based on the generative adversarial network architecture, text-to-image generation based on the autoregressive model architecture, and text-to-image generation based on the diffusion model architecture. Improvements in different aspects were categorized into six subcategories for text-to-image generation methods based on the generative adversarial network architecture: adoption of multi-level hierarchical architectures, application of attention mechanisms, utilization of siamese networks, incorporation of cycle-consistency methods, deep fusion of text features, and enhancement of unconditional models. The general evaluation indicators and datasets of existing text-to-image methods were summarized and discussed through the analysis of different methods.

The inverse kinematic step-by-step solution method based on Euler's angle was proposed to address the problem of incomplete or no analytical solution for the coaxial 3RRR spherical parallel mechanism (CSPM), which was the end-effector of the spherical wrist rehabilitation robot. The CSPM posture Euler angle can be decomposed into two sub-postures rotating around Z-axis and X, Y-axis based on the characteristics of the co-axial spherical parallel mechanism. The set of inverse kinematic solutions for the sub-postures rotating around X-axis and Y-axis was solved. The smaller value in the set of inverse kinematics solutions for each joint was selected and added to the angle of rotation around the Z-axis as the CSPM inverse kinematics solution. The correctness of the proposed method was verified by using CSPM forward kinematics. The actual attitude space of the wrist rehabilitation device was solved by using the proposed method with the constraints of no linkage collision point and no singularity configuration based on the real wrist motion range. The proposed inverse kinematics solution method was interconverted with unit quaternion in the actual posture space, and unit quaternion interpolation was applied to CSPM motion planning. The theoretical calculation results and experimental results were smooth trajectory curves, and the maximum value of both errors didn’t exceed 2.5°.

Aiming at the problem of passive adjustment of traditional docking methods without the support of field measured data, the optimization feedback control technology of fuselage docking process was studied based on the digital twin virtual reality combination technology. A digital twin system integrating the redundant control algorithm and the process optimization strategy was built. The process of measurement-optimization-feedback accuracy optimization based on field measured data was clarified. The digital twin model was accurately reconstructed based on the finite-state machine theory. The monitoring and precision prediction of docking process were realized. The secondary design of process parameters was completed according to the coaxiality evaluation index. The optimized process parameters were redistributed to the physical site to control the on-site docking process. The comparison of the docking position and attitude deviation showed that the optimal control method of the docking accuracy reduced the position deviation of the fuselage barrel by 60.03% and the attitude deviation by 53.94%.

Based on the microfluidic chip processing technology and using the microscopy-micromodel experimental system, the seepage experiment was performed by fabricating the quasi-two dimensional microfluidic chip model to imitate the internal skeleton and pore structure of porous media. The permeability of chip model was calculated by measuring and modifying the pressure drop of both end of the chip model. Computational fluid dynamics (CFD) method was adopted to make the numerical simulation of the seepage process compared with the results of experiment. Under the same condition, compared with the chip model with the square arrangement micro-pillar, the chip model with staggered micro-pillar showed that the tortuosity increased with an amplitude of 5.1%—7.9% microscopically, the flow resistance and pressure drop increased and the permeability decreased with an amplitude of 4.5%—7.4% macroscopically. The permeability of chip models was not only related to the internal pore structure and porosity, but also related to particle diameter and particle arrangement. When the porosity of model was 0.327—0.900, the permeability of the chip model obtained by the numerical simulation method was closed to the experimental results with the error of 9.78%—28.43%. Kozeny-Carman (KC) equation could not predict the experiment results correctly and the maximum error was 73.97%. A modified parallel plate duct flow equation was proposed to predict the permeability of quasi-two dimensional microfluidic chip model. The curve of predicted permeability was consistent well with the numerical and experimental data.nt well with the numerical and experimental data.

A character recognition method based on an improved convolutional recurrent neural network (CRNN) was proposed for the recognition of characters on the chip surface. The image was binarized by the threshold segmentation based on integral map operation, and the orientation correction of the text field image was completed using affine transformation to achieve the localization of text lines. Based on the original CRNN, the backbone network was replaced with MobileNet-V3 structure and the attention mechanism was added between the two layers of LSTM, while the center loss function was introduced. The improved CRNN was used to implement the text line character recognition and tested on 40 510 chip text line images. The multiple sub-models were obtained by fine-tuning the model training with small sample datasets to achieve integrated inference. The combined recognition accuracy used three models was stable at about 99.97%, and the total recognition time of a single chip image was less than 60 ms. The experimental results showed that the accuracy of the improved CRNN algorithm was improved by about 27.48% over the original CRNN, and the integrated inference of multiple models could achieve higher accuracy.

Considering the effect of terminal distribution service mode on service quality and distribution cost, a mixed integer programming model was established with the bi-objectives of distribution cost and customer satisfaction, and NSGA-Ⅱ was improved to solve the model. The validity of the model was verified by the solver GUROBI. The performance of improved NSGA-Ⅱ solution was proved to be stable by solving several cases. The improved NSGA-Ⅱ could obtain high-quality Pareto solution sets with only 1/10 of the computation time of GUROBI. Compared with the traditional NSGA-Ⅱ, the computation time only increased 23 s on average, while the quality of solutions improved 3.37% on average. The improved NSGA-Ⅱ was superior to the GUROBI solver and the traditional NSGA-Ⅱ. The sensitivity analysis shows that the comprehensive utilization of multiple distribution modes is better than the single distribution mode in balancing profit and customers’ satisfaction levels. The distribution cost can be reduced by reasonably increasing the number of parcel lockers and pick-up points and widening customers’ time windows.

A feature fusion emotion recognition method based on modality attention multi-path convolutional neural network was proposed, extracting the connection between the signals of each channel from the electroencephalogram (EEG) and functional near infrared spectroscopy (fNIRS) data induced by emotional video to improve the accuracy of emotion recognition. The EEG and fNIRS data were constructed as graph structure data, and the feature of each mode signal was extracted by multi-path graph convolution. The information of connection between different modal channels was fused by modality attention graph convolution. The modality attention mechanism can give different weights to different modal nodes, thus the graph convolution layer can more fully extract the connection relationship between different modal nodes. Experimental tests were carried out on four types of emotional data collected from 30 subjects. Compared with the results of EEG only and fNIRS only, the recognition accuracy of the proposed graph convolution fusion method was higher, which increased by 8.06% and 22.90% respectively. Compared with the current commonly used EEG and fNIRS fusion method, the average recognition accuracy of the proposed graph convolution fusion method was improved by 2.76%~7.36%. The recognition rate of graph convolution fusion method increased by 1.68% after adding modality attention.