With the development of aerospace technology, variable sweep wing aircraft have attracted attention for their excellent performance. Accurate analysis of their structural mechanics and multi-field coupling performance is a challenge. The research aims to analyse the structural mechanics and multi-field coupling performance of variable sweep wing aircraft. Using numerical simulation methods, the influence of four coupled fields of motion, vibration, heat, and aerodynamics on aircraft performance was comprehensively considered, with special attention paid to the two factors of contact friction coefficient and heat dissipation environment temperature. The results indicated that the overall stress distribution was relatively uniform, and no obvious stress concentration "hot-spots" were observed. In the absence of gravitational equilibrium, the contact forces of contact force numerical model (CFNM)1, CFNM2, and CFNM3 ultimately reached 3.5 x 1.63 N, 3.3 x 1.63 N/2.9 x 1.63 N. The maximum stress in a vibration environment was negatively correlated with the contact friction coefficient. When the contact friction coefficient increased from 0.8 to 1.2, the stress decreased from 191 MPa to 68 MPa. In addition, when the cooling environment temperature reached 50 degrees C the maximum stress in the structure exceeded 100, indicating that the structure no longer met the strength requirements. The results proposed a significant impact of contact friction coefficient and heat dissipation environment temperature on the structural performance of aircraft, and provided important information for improving flight safety and structural optimisation design.
With the aim of enhancing the accuracy of sensorless control system for permanent magnet synchronous motor (PMSM), this paper is on the grounds of integral fast super-twisting sliding mode variable structure model reference adaptive system observer (IFSTSM-MRASO). Firstly, that integral sliding mode surface is devised for the sake of guaranteeing the finite time convergence of the system error, and it is adopted with the intention of reducing the steady-state error of the system. Secondly, by designing a fast super- twisting algorithm with linear correction terms and combining it with the model reference adaptive system, the purpose of weakening the aforementioned chattering issue and improving the system accuracy are achieved. Ultimately, the efficiency of the presented tactic is demonstrated through comparing the consequences of simulation.
As an important automation equipment, welding robot has been widely used in the production process of various industries. In the welding process, due to the influence of welding parameters, technology and other factors, the shape, size, and quality of the weld would be different. Traditional methods usually use hand-designed feature extraction algorithms, which need to rely on the experience of domain experts, which limits the accurate extraction of key features, such as welds. The machine learning model is adopted to learn features from a large number of welding image data, which avoids the manually designed feature extraction algorithm and improves the ability of accurately extracting weld features. Developing a system that can automatically extract and identify key welding features is of great significance for improving the automation level and work efficiency of welding robots. In this paper, the data set containing weld defect images was selected for experiment, and the median filter was used to process the original images. Using the method of machine learning, a welding robot visual recognition system based on machine learning was designed. By comparing with rule-based visual recognition system, this paper tested its performance. According to the experimental results, when the training set size was 600, the accuracy rate of the rules-based visual recognition system was 0.88, while the accuracy rate of the machine learning system was 0.95. This shows that the system built in this paper has better performance. The system in this paper can improve welding quality and efficiency, realise automation and intelligent production, cope with complex environment and change, and promote the development of intelligent manufacturing. This would promote the innovation and progress of the welding process, and provide technical support and impetus for the upgrading and transformation of the manufacturing industry.
The rapid development of smart tourism has brought new opportunities and challenges to people's travel experience. However, due to numerous interest points and complex and diverse needs of tourists, traditional methods often cannot meet the needs of individual and multi-objective. To solve this problem, this paper proposes to establish an evaluation model which takes into account various factors, such as tourists' interest preference, time constraint, and budget constraint. By introducing an improved evolutionary algorithm, genetic operators are combined with local search strategies to generate personalised travel routes. Experimental results show that the improved multi-objective evolutionary algorithm performs well in point of interest recommendation and trip planning. Compared with traditional methods, WAEA algorithm can better meet the individual needs of tourists and deal with multi-objective optimisation problems.
This research aims to optimise the gait planning and joint mechanical structure of multi-legged robots in a specific work environment. The motion equation of a multi-legged robot based on D-H parameters was used, and the Lagrangian was used to derive the dynamic equation to optimise energy allocation. Recurrent neural networks (NNs) were used for gait planning research in the complex terrain. These results confirm that the optimum hardness of the torsion spring is 714 N*mm/deg, and the output torque is the smallest. The error between simulation results and theoretical results is 2%. The simulation results show that this optimisation method can save the torsion spring hardness and reduce the energy consumption by 53%. These results indicated that the recurrent NN-based motion control models had achieved significant results in joint mechanical structure optimisation and gait planning. This study provides strong theoretical support for the motion control of multi-legged robots in specific environments, and is expected to promote gait planning and optimisation in practical applications.
As the significant part of the system, the power system bears the function of power transmission and activation. Taking the ship's power system as an example, the starting and stopping of the entire ship require the power system to function. This research is aimed at the fault analysis of the ship power system. The programmable controller is used to code the algorithm parameters, and an improved algorithm model combining the multi-layer feedforward and radial basis function (RNF) neural networks is built. The new algorithm applies the least square method to calculate the weights of the hidden and the input layers, and adds the genetic algorithm to re-code the threshold of the multi-layer feedforward neural network. The research outcomes indicate that the accuracy of the output value comparison of the improved algorithm is 9% higher than that of the multi-layer feedforward neural network, and 6% higher than that of the radial basis neural network. The approximation error of the improved algorithm is 0.033 less than that of the multi-layer feedforward neural network, and the radial basis neural network is 0.019 less. The improved algorithm has higher accuracy and better precision in power systems' fault diagnosis, and the diagnostic findings are significantly better than the other two algorithms.
The urban sewage pipeline rupture and leakage can seriously affect the daily life of residents. The defect detection of drainage pipelines is an important support for the continuous functioning and later maintenance of pipelines. The internal environment of the wastewater drainage pipeline is poor. The existing defect detection methods have limited ability to extract image features. The accuracy of the detection methods cannot meet the expected requirements. Based on this, a pipeline defect detection robot is designed based on defect image analysis of waste drainage pipelines. For defect image recognition in robots, the Canny algorithm is improved to pre-process defect images. Then, the improved Mask-region-convolutional neural network (Mask-RCNN) model is constructed to extract the processed defect image features. According to the findings, the defect image recognition model constructed in the study converges only after 12 iterations. The accuracy of image recognition can be maintained at around 98.85%. The robot defect detection rate based on this image recognition model is 100%. This indicates that the proposed detection robot based on improved defect image feature recognition can better complete defect detection of drainage pipelines, providing support for drainage pipeline maintenance.
With the popularisation of artificial intelligence, more intelligent products have flooded into various industries. Among them, intelligent transportation and intelligent vehicles have become the main development directions of the automotive industry. Unlike traditional cars, intelligent cars have added driving decision-making functions. Deep learning (DL) neural network algorithms can make judgements on the driving decisions of intelligent vehicles. According to the characteristics of driving decision, an improved DL reinforcement algorithm is proposed based on the DL and reinforcement learning algorithm. In response to the changing environmental factors of driving decisions, a car simulation platform is used to simulate the driving decision changes training in real situations. The experimental results showed that the improved deep reinforcement learning algorithm had a pass rate of 98% in stable road testing. For complex road sections, the passing rate reached 74%. Compared with traditional algorithms, the improved algorithm saved 20% in training time. It has better stability, reliability, generalisation, and real-time performance. The research results indicate that this algorithm has long-term research significance and certain reference value for automotive driving decision-making.
Clean energy assumes paramount importance in the development of virtual synchronous generators (VSGs). In light of varying capacities and transmission line impedances, the application of VSG control strategies for parallel operation of multiple inverters can lead to power sharing issues among these inverters, posing significant risks to the power grid. A judicious power sharing strategy can enhance the green generation efficiency of multi-machine parallel VSGs. This paper proposes a dynamic virtual impedance control strategy to address the issue of reactive power sharing in parallel. Additionally, a secondary regulation controller for voltage and frequency is designed based on distributed cooperative control, ensuring stable voltage and frequency within the rated range. To validate the effectiveness of this proposed strategy, a MATLAB/Simulink-based simulation model is established, the simulation results have verified the correctness and effectiveness of the studied control strategy.
With the continuous increase in the number of cars, the A-pillar blind spot has also become a safety hazard on the road. To better address this issue, a vehicle A-pillar blind spot warning and detection model based on object detection algorithm and multi-source information fusion is proposed. Firstly, vehicle detection is achieved through object detection algorithms to extract vehicle location information. Then, the multi-source information fusion technology is adopted, including cameras, sensors, etc., to obtain information about the surrounding environment of the vehicle. Finally, the performance verification is conducted through simulation experiments. The results show that the proposed model performs better than comparison methods in terms of object detection rate, error rate, recognition accuracy, and recall rate for different objects in A-pillar blind spot in the construction dataset. This verifies the reliability and stability of the proposed model in detecting A-pillar blind spot. The study provides an effective solution to the safety issues caused by blind spots in car A-pillars.
To tackle issues like high costs, independent design, and poor user experience in electric vehicle (EV) charging and discharging technology, this paper proposes an optimisation method based on bidirectional power converters. This method involves analysing mathematical models of converters in various modes, deriving their power circuit transfer functions, and laying the groundwork for circuit theory analysis, control algorithms, and parameter design. Results indicate that in the bidirectional AC/DC circuit, output voltage stabilises at 350 V within 0.1 s. In constant current mode, it maintains a 20 A output while gradually increasing the voltage. In constant voltage mode, battery capacity exceeds 95% in 0.32 s, followed by a gradual decrease in current and voltage stabilisation at 310 V. This study not only proves the circuit's feasibility but also highlights its potential for extensive use in EV charging and smart grid applications.
For the optimisation of the machining process of the machine tool, the research from the machining process cutting parameters multi-objective optimisation (MOO) and considering the energy consumption (EC) of the workshop scheduling analysis, and then proposed a comprehensive consideration of the processing EC, productivity, and surface processing quality of the three tradeoff MOO method of machining parameters. In this study, grey correlation degree weight coefficient and orthogonal experiment were used to verify the optimisation method. Considering the influence of cutting parameters on machining time and EC, a dual-layer optimisation scheduling method of mixed flow shop was adopted. In the experimental results, the scheduling results with double-layer optimisation are lower in terms of total process time and total EC. When the binocular criteria are met, the total process time is 1,083.1 s and the total consumption is 11,845.3 kJ. Optimisation results the total process time is reduced by 205%, the total EC of the workshop is reduced by 12.8%, and the goal of energy-saving optimisation of the workshop processing is realised.
Faced with the problem of fault diagnosis for traction motors in multiple units, this study constructed a digital twin (DT) model to simulate the generation of traction motor fault data, and used the CWRU bearing dataset to obtain physical data. This study designs a training network on the ground of deep learning and domain adaptation model (DAM), and conducts experimental data processing. The outcomes showcase that the DT model can effectively simulate normal and faulty bearing entities. The variation of acceleration in the x-direction of the normal bearing has a regularity, showing a certain periodic small amplitude vibration, with a vibration range of +/- 10 m/s2. Compared to methods, such as deep domain confusion (DDC), the domain adaptive model method has higher accuracy. When the training frequency is 70, the domain adaptive model method has a maximum accuracy of 94.3%, which is 22.8% higher than the DDC method. The research method can effectively predict the traction motor faults of multiple unit trains.
In the field of control systems for permanent magnet synchronous generators (PMSGs), a novel sensorless control approach employing a sliding mode observer (SMO) based on a two-stage filter structure is introduced. This method aims to address the inaccuracies in rotor position detection and speed assessment, typically associated with the chattering phenomena in conventional SMOs. Initially, the proposed SMO is developed utilising the mathematical model of the generator. Subsequently, in lieu of a standard low-pass filter (LPF), a variable frequency complex coefficient filter (VFCCF) is designed. This filter is adept at mitigating harmonic distortions in the back electromotive force (EMF), thereby circumventing the phase delay issues inherent in traditional LPFs. Furthermore, a modified back-EMF observer (MFBEMFO) is employed to attenuate the ripple component in the back EMF signal, resulting in a more refined output. Finally, the effectiveness of the proposed control strategy is substantiated through simulations conducted on the MATLAB/Simulink platform, providing a comprehensive evaluation of its performance.
This paper introduces the basic structure of the gear transmission system in multiple-unit trains and established the corresponding kinetic equation model. Subsequently, the corresponding finite element model was constructed using the kinetic equations of the system. After verifying its validity, the paper tested the intrinsic frequency, critical axle speed, and transmission error of the transmission system under varying mesh stiffness. The first- order intrinsic frequency and axle critical speed showed significant increases when the mesh stiffness exceeded 104 N/m, but they remained essentially unchanged after reaching 109 N/m. The transmission error decreased as the mesh stiffness increased. In conclusion, the design of the gearing system in multiple-unit trains should focus on achieving a sufficiently large gear meshing stiffness. However, considering the cost of improving the meshing stiffness, an excessively high meshing stiffness may not necessarily yield better results. Based on the findings, this paper suggests that a meshing stiffness of 109 N/m exhibits the most favourable effect.
As robot technology continues to develop and innovate, it is widely applied in various fields of production and life. Path planning (PP) is the cornerstone of autonomous navigation of intelligent robots (IRs). However, the current particle swarm optimisation (PSO) and ant colony optimisation (ACO) algorithms still have problems, such as slow convergence rate (CR), complexity, and large amount of computation. Therefore, research will improve the hybrid ACO algorithm and PSO algorithm to obtain feasible robot PP. Then, ACO algorithm is used to obtain the optimal solution (OS), and the particle swarm ant colony fusion algorithm is obtained. Compared with PSO and ACO algorithms, the shortest path of the fusion algorithm is 43.78 m, which is closer to the optimal path. In an environment with an obstacle ratio of 0.4, the optimal performance index of the fusion algorithm is 8.84%, and the number of iterations during convergence is 24. Compared with genetic algorithm (GA) and sampling based PP algorithm, CR of this fusion algorithm is faster and the average value of the optimal path is smaller when the obstacle ratio is 0.7. In summary, the fusion algorithm proposed in this research is effective in IR path planning (IRPP). This can improve the path optimisation ability of IRs and provide a basis for exploring more effective global dynamic PP methods in the future.
For the low quality and slow efficiency of college physical education (PE) intelligent network teaching, a classification and evaluation method of college PE made large-scale open on-line courses (MOOC) mode is proposed. This method constructs a fuzzy ID3 decision tree model. On this basis, it uses triangular membership function and Kohonen feature mapping algorithm to discretise and fuzzify, and finally completes the evaluation of students' PE MOOC mode. The fuzzy ID3 exhibits the highest degree of accuracy in classification when the authenticity threshold is set at approximately 0.8. The classification accuracy indicates fuzzy ID3 in the four databases can obtain high accuracy. The classification accuracy of the training set of l-o database is 75.8%, 62.8%, 76.5%, and 95.0%. Fuzzy ID3 outperforms the minimum classifications uncertainty and yields better classification results for the l-o database with 18, 12, 16, and 10 classification rules, respectively. At an authenticity threshold of around 0.8, the fuzzy ID3 algorithm demonstrates the highest degree of accuracy for classification. The suggested MOOC model can extract a larger number of classification rules, which leads to better accuracy. In the future, the PE teaching model can be implanted in other colleges and universities.
A data transmission technology combining blockchain technology and interstellar file system is proposed to address the issue of lack of effective data confidentiality and transmission methods in logistics system data transmission. This technology converts all key operations into transactions on the blockchain, ensuring the recording and verification of the entire operation, and improving the transparency and data security of the logistics process. At the same time, collaborative filtering algorithms and light gradient boosting machine algorithms based on decision tree algorithms are used as recommendation systems to automatically match suitable drivers for order transportation, to optimise order allocation and loading work. The results show that the smart logistics (SL) system designed in the study has a high throughput and a low utilisation rate of system memory. When the concurrency is 800, the throughput of the registration and review function is 701.5/s. Through the integration of blockchain technology and interstellar file systems, as well as the use of collaborative filtering algorithms and decision tree algorithms in conjunction with gradient enhancement framework algorithms, logistics processes have been successfully optimised, improving data security and operational efficiency.
With the popularisation of autonomous connected vehicles, the growing demand for data exchange and storage urgently requires a decentralised, secure, and efficient storage mechanism. In view of this, this study proposes a hybrid big data storage system based on blockchain and Kademolia technology and applies it to the performance testing of the Internet of Vehicles (IoV) system. The results show that in the CompCars dataset, the proposed method iterated around the 20th time and achieved a fitness value of 99 times. In the comparison of fitting accuracy, at a time of 0.901 s, the fitting accuracy of the research method reached 98.92%. In the throughput results, the maximum throughput value of this method is 405 TPS after reaching a throughput of 450 transactions/s. The above results indicate that the proposed method has higher throughput and can accelerate transaction processing of storage systems with better performance, providing new technical support for optimising hybrid storage systems for autonomous networked vehicles.
With the growth of the economy and the progress of people's life quality, cars have become the primary choice for most people to travel. However, as travel becomes more convenient, traffic accidents are increasingly frequent. This study focuses on the active collision avoidance system (ACAS) of intelligent vehicles and the related strategies that have been analysed. First, utilising the principles of automotive safety control theory and an overall structural dynamics model of the vehicle, an active collision avoidance control system for the vehicle is developed. Next, convolutional neural networks (CNNs) are implemented to estimate the vehicle's safety. After collecting information from car sensors and radar, the system evaluates the car's driving status and actively takes measures to avoid collisions. The final safety distance serves as a determining factor for labelling a car as dangerous. Simulation experiments have verified the accuracy of the model. When the vehicle is stationary and at a speed of 60 km/h, and 60 m distant from the vehicle in front, the system can anticipate and control the distance between them within a safe range. Even in motion, this system can perform collision avoidance tasks and exhibit outstanding capabilities. The results indicate that the ACAS designed in this experiment based on CNNs has excellent performance in various complex road conditions.
