End-to-end control is one approach for self-driving. However, end-to-end self-driving methods based on reinforcement learning have troubles dealing with widely varying driving scenarios. The learning algorithm cannot easily determine the random decline velocity during training for complex scenarios. If the velocity is too fast, the algorithm will not obtain a reasonable policy for new scenarios, while if the velocity is too slow, the algorithm will not converge fast. A random policy and experience replay method based on the state distribution is developed here to improve the random decline velocity selection. The various random process parameters are selected based on the distance between the present state and the saved states. In addition, the replay probability of various scenes that occur less frequently is also increased. Simulations show that the algorithm has sufficient exploration ability in the later stage of the training when faced with scenes having very different situations from the situations in the early stages, which improves the lane keeping ability of the end-to-end self-driving approach based on the deep deterministic policy gradient (DDPG) in new situations.

Groups of intelligent electric vehicles are more difficult to control when experiencing communication delays. This paper presents a distributed adaptive robust control method for vehicle platoons. The platoon control model includes nonlinear terms, parameter uncertainties, communication delays and other factors. The influence of the nonlinear terms on the control system is eliminated by using an inverse model. Then, a distributed state feedback control structure is designed with a closed-loop model with communication delays, external disturbances and parameter uncertainties. The communication topology is then decoupled using property value decomposition. The existence condition of the adaptive controller is derived using the linear matrix inequality processing method. The matrix inequality is low dimensional and independent of the platoon length. The control system is shown to be stable with communication delays using Lyapunov stability theory. Simulations demonstrate the feasibility and effectiveness of this distributed adaptive robust control method when influenced by communication delays.

The safety reliability of steer-by-wire systems (SBW) is the main reason restricting their application; thus, their safety reliability are important. Some studies have proposed safe and reliable dual-redundancy steer-by-wire systems. This paper presents a current-balance dual-redundancy control method for such systems which balances the torques of the dual motors. The system can also cut the power to a faulty motor by reorganizing the control system which reduces the system order so that steering control is maintained after one motor failure. Vehicle tests show that this method provides steering control with balanced motors. The tests also show that the system modifies the control mode after a motor failure to maintain vehicle steering.

A lateral control method for intelligent vehicles is developed based on the optimal front-tire lateral force to improve the lateral stability and the path tracking accuracy of intelligent vehicles going around large curvature turns. A linear quadratic regulator (LQR) controller using feedforward and feedback control is used to determine the desired front-tire lateral force in real time to reduce the tracking error. The control input is then converted into the desired steering angle based on the brush tire model. This method properly retains the nonlinear characteristics of the vehicle and tire models. The LQR controller is verified via simulations on PreScan. The results show that this LQR controller not only reduces the path tracking error relative to the general LQR method, but also ensures lateral stability of the vehicle.

Most existing answer stance detection methods ignore the interactive dependence between question and answer. This paper describes an answer stance detection method based on a recurrent interactive attention (RIA) network. This method simulates the human interactions in question-answer reading comprehension using an interactive attention mechanism and iterations to simulate the interactive dependence between question and answer for detecting the answer stance. In addition, since the question text cannot explicitly express its stance, the question text is transformed into declarative sentences. Tests on a Chinese social media question-answer dataset show that this method outperforms existing answer stance detection methods due to the effective representation of the interactive dependence between the question and the answer.

Concurrent structures which are shared by the predicate-object phrase and the subject-predicate phrase in one sentence are common Chinese verb structures. However, their complexity makes such structures difficult to analyze. Therefore, recognition of concurrent structures is important for semantic analyses and downstream tasks. However, there are few existing concurrent corpora with no concurrent corpora for the Chinese AMR annotation system. This study summarizes a set of concurrent corpus annotation specifications and builds a concurrent corpus for Chinese AMR annotation systems which contains 4 760 concurrent sentences. The LA-BiLSTM-CRF model is then used to recognize concurrent structures with an F1 score of 86.06%. The recognition results are analyzed to determine needed improvements.

Multilingual neural machine translation is an effective method for translations of low-resource languages that have relatively small amounts of data available to train machine translations. Existing methods usually rely on shared vocabulary for multilingual translations between similar languages such as English, French, and German. However, the Burmese language is a typical low-resource language. The language structures of Chinese, English and Burmese are also quite different. A multilingual joint training method is presented here for a Chinese-English-Burmese neural machine translation that alleviates the problem of the limited amount of shared vocabulary between these languages. The rich Chinese-English parallel corpus and the poor Chinese-Burmese and English-Burmese corpora are jointly trained using the Transformer framework. The model maps the Chinese-Burmese, Chinese-English and English-Burmese vocabulary to the same semantic space on the encoding and decoding sides to reduce the differences between the Chinese, English and Burmese language structures. The influence of the shared vocabulary compensates for the lack of Chinese-Burmese and English-Burmese data by sharing the Chinese-English corpus training parameters. Tests show that in one-to-many and many-to-many translation scenarios, this method has significantly better BLEU scores over the baseline models for Chinese-English, English-Burmese, and Chinese-Burmese translations.

Many biomedical literature mining systems use the pre-training language model BioBert which provides state-of-the-art biomedical named entity recognition after pre-training. However, BioBert is too large scale and slow. This paper presents a faster biomedical named entity recognition model, FastBioNER, that is based on knowledge distillation. FastBioNER compresses the BioBert model using dynamic knowledge distillation. A dynamic weight function is used to simulate the real learning behavior to adjust the importance of the loss function of each part during training. Then, the trained BioBert is compressed into a small student model by dynamic knowledge distillation. The FastBioNER model was validated on three common data sets, NCBI disease, BC5CDR-chem and BC4CHEMD. The tests show that FastBioNER had the highest F1 values after BioBert at 88.63%, 92.82% and 92.60% for the three data sets while reducing the BioBert model size by 39.26% and the inference time by 46.17% at the cost of 1.10%, 0.86% and 0.15% smaller F1.

A spatial co-location pattern is a set of spatial features that are frequently observed together in space. Traditional co-location pattern mining uses a single distance threshold to define neighbor relationships while ignoring the impact of distance differences, but the minimum prevalence threshold is difficult to determine for inexperienced users. This paper presents a method for calculating the neighborhood membership degree based on fuzzy theory and d-grids. This method does not calculate the Euclidean distance and quickly finds the maximal cliques that satisfy the fuzzy neighborhood relationship by using the d-grid. The results was then combined with the Top-k algorithm to find the k most prevalent co-location patterns. Tests show that this method is more efficient and gives more detailed results. The recall rate shows that the k most prevalent patterns obtained by this method agree well with those obtained by the traditional co-location pattern mining algorithm, which shows the effectiveness of this fuzzy measurement and mining algorithm.

The development of the Internet has driven popularization of e-commerce and other applications. For these applications, the Internet needs various services such as temporary matching which matches various types of tasks with the servicer skills while minimizing the distance overhead between matching objects. This paper presents a spatial keywords task matching algorithm for these conditions. Given a task set and a servicer set with spatial locations and keywords, the sum of the distances between all the tasks and the matching servicers is minimized for the premise that all the tasks can be completed. The massive number of tasks and servicers and the wide range of keywords complicate efficient determinations of high quality matching results. This study uses a k-nearest neighbor incremental optimization strategy to improve the matching quality of the traditional matching algorithm. A grouping optimization strategy based on spatial partitioning is then used to improve the matching efficiencies for large datasets. These two strategies are then used to develop a keyword k-nearest neighbor incremental algorithm and a keyword-based grouping optimization algorithm. Tests on real datasets verify the effectiveness of these algorithms.

Automatic robotic drilling and riveting greatly improve aircraft assembly quality and flexibility while reducing costs, so automatic drilling and riveting are being widely used for aircraft assembly. However, almost all robotic drilling and riveting systems are based on six degrees of freedom serial robots. However, these robots have an inherent weakness due to their poor rigidity. Large pressing and drilling forces during drilling and riveting can lead to deformation and even flutter of the robot end actuator, which seriously affect the drilling and riveting accuracy and quality. A static stiffness model of a drilling and riveting robot is developed with joint stiffness measurements to predict robot joint stiffnesses. A stiffness evaluation index is then used to characterize the robotic arm stiffness in the working space. The stiffness index can be used to optimize the robot position and posture for a specific operation using a particle swarm algorithm to improve the stiffness. This improves the system stability and quality.

In five-axis flank milling, the continuous change of the tool posture complicates predictions of the instantaneous undeformed cutting thickness (IUCT). The milling force prediction accuracy is improved in this study using a micro-element milling force model for the an IUCT method for a flat-end milling cutter with the IUCT divided into two parts. The results show that the milling force model calculational efficiency is improved by 40% compared to a previous model. The model predictions were then verified against flank milling tests. These results show that the model is accurate and efficient.

The shapes of large aviation workpieces can deform due to temperature changes, which significantly impacts precise digital measurements of the product shape and the accuracy of component docking. This paper describes a method to optimize the transfer station parameters of a laser tracker to account for three-dimensional deformation of the workpiece. The center of a three-dimensional model of the workpiece was calculated using the ANSYS finite element thermal analysis to account for the thermal expansion of the workpiece. An objective function was then defined for optimizing the transfer station parameters that considered the three-dimensional thermal expansion of the workpiece. Numerous individuals in a population of models were randomly initialized to calculate the adaptive values for each individual. The particle swarm optimization (PSO) algorithm was then used to iteratively optimize the individuals in the population to obtain the optimal tracking transfer parameters. This method takes into account the three-dimensional thermal deformation of large workpieces caused by temperature changes to improve the tracker measurement accuracy.

This paper describes a classification method based on a gray-level co-occurrence matrix (GLCM) and the Adaboost algorithm for monitoring passivation of a continuously polishing machine disk. The texture image features of a continuously polishing machine disk are derived from a GLCM operator of a disk photograph. Four second-order statistics of the GLCM are input into a Adaboost classifier for training so that the classifier can then identify if the disk has been passivated. Tests show that the classification accuracy is best when the GLCM point-to-point distance is 11 and the GLCM gray level is 16. The image classification accuracy is 98.3%, which is 9.5% higher than that of the LBP algorithm and 2.08% higher than that of the PNN algorithm.

Pulsed current tungsten inert gas (TIG) shielded arc welding is widely used in high quality welding. The arc characteristics, including the variations of the arc shape, gray-scale and temperature with the pulsed current, are studied here to understand the dynamic physical characteristics of pulsed TIG welding arcs, especially the welding stability and the control of the weld seam quality. The pulse current and corresponding arc images were acquired synchronously during the welding process. Image processing using the Fowler-Milne method was used to characterize the arc shape, gray-scale and temperatures at specific positions within the pulsed TIG welding arc to analyze their periodic variations. The experimental results show that the arc characteristic parameters at specific positions within the pulsed arc are synchronized with the current variations to within 0.1 ms. The arc expands and the arc temperature quickly rises as the current increases along the front edge of the pulse and then the arc contracts and the arc temperature falls as the current decreases. The arc temperature 1 mm below the tungsten electrode rises at a rate of 11 613 K/ms as the current increases and decreases at a rate of 5 710 K/ms as the current decreases. The arc gray-scale changes less closer to the tungsten electrode before and after the current changes, but the arc temperature changes are greater.

Since signal transmissions over the sea surface are affected by the marine environment including seawater evaporation and tidal motion, existing terrestrial channel models are not suitable for maritime communications since they do not consider the effects of these environmental factors. This study evaluates the effect of the marine environment on wireless transmissions using channel measurement experiments in the Yellow Sea. The real-time hydrological and meteorological information in the area was also collected. The effect of seawater evaporation was modeled by an evaporation duct along the surface based on the sea state parameters with an analysis of the sensitivity of the duct height to various parameters. The effect of the tidal motion was modeled using a modified two-ray channel model that includes a tidal factor with the model verifying that the small-scale channel fading follows the Rice distribution. Comparison of the models with the measured data shows that this channel model better describes the channel characteristics over the Yellow Sea with a time-varying sea state environment than existing models.

The threat space of a satellite navigation signal distortion model describes the signal distortion range which causes a large differential error but is difficult to detect by receiver observations. The threat space may have disastrous consequences for civil aviation and other safety navigation services. A larger threat space increases the navigation services risk and the distortion detection requirements. Thus, satellite navigation systems need appropriate methods to minimize the threat space. This paper presents a method that reduces the range bias detection threshold to reduce the threat space. The effectiveness of this method is evaluated using the BeiDou Navigation Satellite System (BDS) B1C and B2a signals as examples. The results show that the threat space can be reduced by more than 40% compared to a space for a 20 m threshold when the ranging bias threshold is 5 m. This method should be implemented by using satellite onboard receivers to reduce the threat space.

Signal occlusion is one of the main difficulties in the application of the global navigation satellite system (GNSS) to the positioning of spinning objects. Due to the directional antenna, the conventional tracking algorithm based on the assumption of slow amplitude variation is invalid. This paper proposes an anti-rotation GNSS tracking algorithm based on the joint processing of double antennas, which can adjust the parameters of the blocked channel according to the visual one to ensure that the loop will not lose lock when the signal is reproduced. Through the joint processing of measurement from the channels of double antennas, the differential demodulation of data bits is realized. Tests use a turntable to create a spinning environment. This method can complete the synchronization of the spinning signal. Moreover, the tracking performance, data demodulation ability, and positioning accuracy are better than that of the scalar tracking algorithm of a single antenna with discontinuous reception, greatly improving the robustness.