In order to solve the problem of refined tracking for formation targets, a gray refined tracking algorithm based on distance vector is proposed.Firstly, the distance between the points and the tracks of the individuals in the formation is extracted to build the distance vector.Secondly, multiple combinations of point/track mapping are established based on the distance vector.Finally, the optimal combination among these mapping combinations is chosen as the refined correlation result by utilizing the gray theory, and then the state update of the tracks of each target inside the formation is completed by using the Kalman filter.The simulation results show that, compared with the Structure Branch Multiple Hypothesis Tracking (SB-MHT) algorithm which has superior performance in the traditional muti-target tracking field, this algorithm has better performance in tracking accuracy, real-time performance and the effective tracking rate. Moreover, this algorithm has excellent engineering applicability.

Armament System-of-Systems (SoS) planning is an important part of the armament SoS construction, which has important guiding significance for armament demonstration work.Firstly, qualitative analysis is made and formulation description is presented for armament SoS planning problem.The modeling method of SoS based on super network is proposed, and the quantitative calculation model of capability index is constructed.NSGA-II algorithm paradigm is adopted.The decision makers preference representation method is analyzed, and integrated into the optimization process.Thus the original optimization problem is transformed into a three-object optimization problem with the goals of super volume, unknownness and decision maker satisfaction.The extensional measure is used to redesign the dominant and evolutionary strategies.Finally, an example of air-based combat SoS planning is cited for verifying the efficiency and rationality of the algorithm, which can supplies as a support for decision-making in armament development and top-level planning.

To identify whether the products failure mechanism in accelerating tests is changing, a testing method of failure mechanism consistency of Weibull life distribution based on the principle of acceleration factor constancy is studied.Weibull distribution is converted to I-type extreme value distribution.A method for statistical testing of the scale parameters in extreme value distribution is studied by using Ansari-Bradley method.By using Arrhenius acceleration model, t statistic is constructed based on the least squares method and characteristics of regression coefficients.Statistical testing of the acceleration factor and distribution parameters is implemented, and the stress under which the failure mechanism will change is distinguished.Simulation results show the effectiveness of the proposed method.

In order to improve the tracking performance of the SAMF algorithm in complex scenarios, an object tracking algorithm combining SAMF with visual saliency is proposed.Based on the SAMF correlation filter tracking framework, the reliability of the SAMF tracking results is evaluated by designing a confidence discrimination strategy.When the tracking results are identified as low confidence, the saliency detection algorithm is used to correct them, so as to realize the targets relocation to address the tracking drift caused by occlusion and other factors.Experiment results show that the proposed algorithm greatly improves the overall performance and tracking robustness of the SAMF algorithm when dealing with various interference factors, while maintaining good real-time performance.

For non-Gaussian stochastic distribution systems with faults and disturbances, a novel robust fault-tolerant control scheme is proposed based on Iterative Learning Observer (ILO).The linear B-spline neural network is used to establish the relationship between output Probability Density Function (PDF) and dynamic weights.An iterative learning observer is designed to accurately estimate the faults with less computational load.By using the fault estimation information, a fault-tolerant controller is designed, so that the system can still track the expected weight vector after the fault occurs.Simulation results demonstrate the effectiveness of the proposed approach.The iterative learning observer can rapidly reconstruct the faults after a short transition period, and the fault-tolerant controller based on PI tracking has good tracking effects on both the constant and time-varying weight vectors.

The Dynamic Output Feedback Control (DOFC) for Networked Control Systems (NCSs) based on the event-triggered mechanism is studied.By comprehensively considering the problems of induced network delays and timing disorders in NCSs, a closed-loop model containing the event-triggered mechanism and the DOFC is constructed.Compared with the traditional static control, DOFC can ignore the problem of limited pole configuration and increase the adaptability of the system.Simulation results have verified the effectiveness of the proposed event-triggered mechanism and the DOFC.

In the background of the anti-sea striking launched by coordinated anti-ship missiles and supported by information from the satellite, the distribution area of the target is categorized based on probability density models.The joint acquisition probability and the overlapping acquisition probability are taken as evaluation indexes.The design process of a coordinated striking method based on satellite information is constructed, and then the striking method with coordinated missiles is given based on probability density.Through simulation analysis, it is proved that compared with the striking method based on uniform distribution, the coordinated striking method based on probability density models is more reasonable and can provide a reference for decision.

In view of cooperative task allocation of Autonomous Underwater Vehicle (AUV), a Multi-Objective Discrete Particle Swarm Optimization and Simulated Annealing (MODPSO-SA) algorithm based on Pareto is proposed.In order to make the particle swarm optimization avoid from falling into local minimum, a new multi-objective local search strategy is proposed by integrating the improved simulated annealing.The simulation results show that the MODPSO-SA algorithm can obtain multiple Pareto solution sets, and solve the multi-AUV task allocation problem effectively.

To solve the problem of finite-time constraints when the missile intercepts a maneuvering target in the terminal guidance stage, the mathematical model of the target interception system is established considering the dynamic delay of the autopilot.To solve the problem of poor guidance stability caused by the singularity of the traditional terminal sliding mode guidance law, a new non-singular terminal sliding mode guidance law based on the improved Extended State Observer (ESO) is proposed.This method ensures that the Line-of-Sight (LOS) angular rate converges to zero in a finite time and converges quickly at the equilibrium point.In order to avoid curve chatter caused by the traditional ESO, an improved ESO is adopted to accurately estimate external disturbances and compensate for the systems state.A model of the surface-to-air missile is designed for simulation verification.The simulation results show that this method not only shortens the systems convergence, but also improves the guidance precision of the surface-to-air missile.

In addition to the influence of the vibration of ground vehicles in subgrade environment, the amphibious weapons will also be affected by the swinging of its carrier in marine environment, which may result in the deviation of the launch angle of the amphibious weapon launcher.By using the self-adaptation and self-learning ability of wavelet neural network, an internal model control method based on self-built wavelet neural network is proposed for the amphibious weapon servo system.The excitation strength and attenuation degree of the wavelet basis function are used to add neuron nodes, or to trim or delete neuron nodes to optimize the hidden layer structure, and then the LM algorithm is used to improve the learning rate.The self-built wavelet neural network is used to identify the forward and inverse models of the internal model control system to improve the control technology.The final results show that the method can effectively improve the systems anti-jamming capability, launching accuracy and adjustment speed.

In air combat, the target often performs maneuver in advance based on the current battlefield situation, so that the traditional tactical attack zone of air-to-air missiles cannot fully and accurately reflect the current attack/defense situation.This paper proposes a simulation method of the tactical attack zone by predicting target maneuver.The mathematical three-degree-of-freedom model of the missile, the model with limited extrapolation for target maneuver prediction and the model of the guidance law are established respectively to simulate the attack area.The results show that, the early correction of the attack area through the predicted target maneuver mode enables the pilot to grasp the changes of battlefield situation of the two sides, make the right decision, and complete the OODA loop as soon as possible, which also provides effective auxiliary information for the subsequent decision.

In order to obtain the law of Doppler spectrum distribution in the area radiated by the antennas mainlobe when the radar altimeter moves relative to the ground, a wide-band Doppler echo model is established, and the beam irradiation area is divided into several rectangular scattering grids by using the grid mapping method based on the equal Doppler band and the equal distance ring.The simulation of echo Doppler power spectrum is conducted by using the method of echo power superposition of each rectangular grid in the equal Doppler band.It overcomes the shortcomings of complicated calculation based on the polar-coordinate integration method, which is not conducive to the subsequent engineering realization.The simulation results are consistent with the theoretical values, which shows that the model is reasonable.Finally, the law of Doppler spectrum distribution of the radar altimeter at different altitudes and speeds is obtained, which provides a theoretical support for the subsequent echo simulation of the radar altimeter.

The multi-task priority allocation method of a real-time embedded uni-processor system is studied.Firstly, the Deadline Monotonic (DM) priority allocation method is selected and implemented for the multiple tasks, which are classified for running the acquisition and incentive software of the avionic relay system.Then, by using the preemptive priority-scheduling algorithm of VxWorks, the schedulability of the task set is theoretically analyzed and calculated.The general process of the scheduling design for the VxWorks-based data acquisition and incentive software is also given.Finally, a data-forwarding experiment was performed on the avionic relay system based on the scheduling algorithm.The experimental results show that the maximum delay time for the avionic relay system to forward data does not exceed 3.5 ms, which can realize the stable and correct forwarding of 4 types of data.Therefore, the avionic-relay task scheduling design based on the preemptive scheduling algorithm of DM priority allocation can meet the real-time and reliability requirements.

For continuous zoom lens, the optimal design of pressure angle of cam curve is an important factor influencing the function of the lens.Based on the cam driving structure and mechanics theory, an analysis is made to the influence of the cam curves pressure angle on the transmission torque.With consideration of the machining error and pressure angle model, the constraint conditions and optimizing principle are proposed.By using the smooth characteristic of Gauss function, the optimization method of non-uniform widening of the cam curve is realized along vertical direction of the optical axis, and the maximum pressure angle of the cam curve is restrained.The method has been applied for optimization of a cam curve of a zoom lens obtained by optical design, for which the range of pressure angle is optimized from 0°~39.01° to 15°~31.16° under constraints, and the maximum pressure angle is reduced by 20.12%.These results indicate that the method can be effectively applied to the design of cam zoom system for improving the performance of the system.

To solve the problems of fuzzy boundary, fracture and target loss in semantic segmentation of long-distance targets in complex environment, a semantic segmentation model using boundary information based on DeepLabV3+ network is proposed.The improved Darknet-53 network is used to replace the original DeepLabV3+ feature extraction network to speed up the models operation, and a feature fusion module is designed as a low-level feature to recover the detailed information in the decoding stage.In order to further optimize the targets boundary, by using the principle of feature sharing, a boundary extraction module is designed to predict the targets boundary by learning multi-scale information through the feature sharing layer of the main network, so as to optimize the segmented image and improve the prediction accuracy of the model at the boundary.The experimental results show that the proposed semantic segmentation model can effectively alleviate the problems of fuzzy boundary, fracture and target loss in the semantic segmentation of long-distance targets.

Camera calibration is very important in machine vision technology.In view of the cumbersome operation of the traditional 3D object calibration method and the influence of the initial weight and threshold value on the BP neural network calibration, this paper proposes a mind evolution neural network calibration method based on the optical axis convergence model.The BP neural network algorithm can approach nonlinear function, and mind evolution algorithm has strong global optimization ability.Therefore, the problems of BP neural network such as easily falling into local minimum and randomization of initial weight and threshold can be effectively solved.Experiments show that, compared with the classical Zhang Zhengyou calibration method and BP neural network method,the mind evolution neural network calibration method can achieve better accuracy of binocular calibration.

The problem of global robust output regulation for uncertain nonlinear systems with linear external systems is studied when the control direction is unknown.It is assumed that the system and high-frequency gain symbols are unknown, and the unknown parameters can be arbitrarily large, which bring a challenge to the design of the controller.In order to solve this problem, firstly, coordinate transformation is conducted to calculate the new state equation.Then, by using Nussbaum dynamic gain technology, the adaptive control method and the internal model principle, the robust output regulation problem is transformed into the robust stability problem, and a dynamic output feedback controller is designed.Finally, Lyapunovs derivative is made negative definite by using the designed dynamic output feedback controller.In this way, the problem of global robust output regulation of the nonlinear output feedback system is solved theoretically.The effectiveness of the proposed adaptive controller is illustrated by an example.

To solve the problem of autonomous evasion of the carrier aircraft facing incoming enemy missiles, a deep reinforcement learning method based on the improved DDPG algorithm is adopted for training and learning.In addition to considering the evasion performance in the reward function, rewarding models are established respectively for the cost of the aircrafts altitude maintenance and speed maintenance, as well as the relative altitude change and the approaching speed change of the incoming missile.Finally, training simulation tests and analysis were conducted based on the aircraft model.Through simulation, it can be seen that the training results can effectively realize the evasion decision of the incoming missile, and the designed reward function and input parameters can also play a correct role, and the results have certain generalization ability.

The air-water operation system composed by a drone and an unmanned boat can greatly improve the ability and efficiency to complete operations.Therefore, the control of the rotor drones autonomous landing on moving unmanned boats is studied.The position tracking of the unmanned boat is realized by using the unmanned boats movement information and the drones GPS information obtained by the drone through communication, as well as the calculation result of the recognition and positioning of the landing marks.According to the obtained attitude information of the unmanned boat, the drone keeps the attitude tracking of the unmanned boat and finally autonomously lands on the moving unmanned boat.The air-water system composed of the quadrotor drone and the unmanned boat is built to conduct experiment.Experiment results indicate that the system can reliably complete the tasks of each stage in the process of autonomous landing of the drone, which verifies the feasibility of the system.

In order to collect the data generated by the aerocraft during the wind tunnel test in real time and control the test flow, a scheme based on Windows 7+RTX64 is proposed to design and develop the software system.Firstly, a flow for testing the real-time performance of RTX64 is designed and the test result is compared with that of Windows under the same conditions, verifying that RTX64 can significantly enhance the real-time performance of the system.Secondly, the design of Inter-Process Communication (IPC) and test flow of the software system for aerocrafts virtual flight is described in detail, and inter-process data interaction and synchronization is realized by using the shared memory and the event.Finally, the test results and analysis are given.The results show that the sensor data from the gyroscope, the encoder and the balance can be collected in real time and the time sequence of the virtual flight test of the aerocraft can be controlled correctly during the 4 ms period, proving the feasibility of the design.

The 3D measurement technology using structural light has wide application due to its advantages of no contact and high precision.However, local brightness saturation will occur on the captured images for the highly-reflective surface, which will lead to information distortion or loss and inaccurate 3D reconstruction.In order to realize the measurement of the highly-reflective surface, an improved 3D measurement method based on phase detection of projected grating is proposed.The method conducts binaryzation on the projected encoded phase-shifting image, fuses the images captured under different exposure time, and reconstructs the fringe image sequence.The method can improve the robustness of encoded images and avoids the inconsistency of the cameras dynamic range with the brightness range of the objects surface reflection.The experimental results show that the proposed method can overcome the defect of loss of measurement information caused by local saturation of the image in the existing methods, and realize the high-precision and complete measurement of 3D topography of the highly-reflective curved surface by using structural light.

To solve the problems of external disturbances and nonlinear characteristics in the positioning control of the AC servo system of antiaircraft guns, an improved single-neuron Active Disturbance Rejection Controller based on Self-Recurrent Wavelet Neural Network (SRWNN-ADRC) is designed.The Single Neuron Adaptive Controller (SNAC) takes the nonlinear gain in the nonlinear error feedback control law as its weight coefficient.SRWNN is taken as an identifier, through which the gradient information of the controlled object is identified online and supplied to SNAC.Through the self-learning function of SNAC, the parameters in ADRC can be adjusted online.Simulation results show that this control strategy enables the system to have better steady-state performance, stronger anti-interference ability, and an optimized dynamic quality.

The bonds elastic modulus and Poissons ratio have important effects on optical-mechanical simulation of the reflective mirror, but these parameters are influenced by too many factors to accurately obtain.An experiment is conducted to obtain the test mode of the reflective mirror, a finite element simulation is conducted according to the bonds parameters in the Literature［1］ to get FEM mode, and then a correlation analysis is conducted to find that the test mode has a poor correlation with FEM mode.Subsequently, the bonds elastic modulus and Poissons ratio are taken as variables, and an SQP iterative optimization is implemented to obtain these parameters.After model modification, the correlation between the modified FEM Mode and the test mode is significantly improved.Finally, the surface figure and stress of the reflective mirror under two working conditions of assembly stress and low-temperature environment after model modification is compared with that before the modification.The calculation results show that, after model modification, the change of surface figure resulted from assembly stress will have influence on optical imaging, and the problem of stress under low temperature cant be ignored.

Micro-displacement driving and control with nano-scale precision are usually achieved by using the piezoelectric ceramic driver to drive the flexible mechanism.However, the travel distance of the piezoelectric ceramic driver is so small that a flexible amplification mechanism is needed to amplify its displacement.The nonlinear characteristics of the piezoelectric ceramic driver such as creeping and hysteresis greatly affect the motion accuracy of the output displacement after being amplified by the amplification mechanism.To solve the above two problems, a new type of flexible amplification mechanism is designed by replacing the four bridge arms of the traditional bridge mechanism with the Scott-Russell mechanism.At the same time, a compound control strategy which combines feedforward control based on the dynamic inverse model with PID feedback control is adopted to improve the control bandwidth and motion accuracy of the amplification mechanism.The experimental results show that, compared with that of the traditional PID feedback control, the RMS tracking error of the compound control strategy is reduced by 33% at 1 Hz input frequency and by 73% at 10 Hz.The compound control strategy which combines feedforward control based on dynamic inverse model with PID feedback control greatly improves the tracking accuracy of the amplification mechanism.