Due to the reasons as wear of materials and change of external environment during the operation of vital systems including engineering facilities and military equipment,the performance of the system is gradually decreasing or even fails,which will result in economic losses and loss of life. Therefore,in order to ensure the normal operation of the system,the Remaining Useful Life (RUL) prediction technology has attracted the attention of researchers. In the era of big data,the obtained monitoring data has the characteristics of high dimensions and strong coupling,so it is difficult to model the data by using the traditional RUL prediction method.The deep learning method can accurately establish the mapping relationship between the monitoring data and the degradation states or the life tag. This paper elaborates the research status of four typical deep learning models in the field of RUL prediction in detail,summarizes the advantages and disadvantages of these models,and discusses the development direction of deep learning based RUL prediction of complex degradation systems.

Traditional convolutional neural networks have a large demand of computation and memory,which makes the development of embedded devices for intelligent applications become a challenge.In order to deploy highly complex deep learning applications into the low-cost embedded platforms with limited performance,a small embedded system for image classification is designed.Based on the Structured Sparsity Learning (SSL) algorithm,a sparse convolutional neural network model is constructed under the framework of Caffe and deployed on IndustriPi minimization system.Test results show that the accuracy of85.5% and operating speed of more than 8 frames per second are achieved.Compared with classical models,the sparse model can reduce computational amount and memory occupancy to a great extent,and increase the embedded device operating speed.

Both attack angle and sideslip angle are important flight state parameters of aircraft.However,it is difficult for the atmospheric data system to accurately measure the airflow angles under the conditions of adverse weather,high attack angle or maneuvering flight.A method for estimation of aircraft airflow angle is studied based on flight data.Taking the external disturbances,the different data sampling frequencies,and the unknown statistical characteristics of flight data into account,the aircraft system state equation and measurement equation are established.The adaptive Kalman filtering algorithm based on the maximum likelihood criterion is integrated with the non-equal interval theory.Taking the turning and climbing flight of aircraft as an example,external disturbance is applied to estimate the attack angle and sideslip angle.Experimental results show that the algorithm has better estimation precision and higher robustness against external disturbance than the extended Kalman filtering algorithm and the unscented Kalman filtering algorithm.

To realize the recognition of intermittent sampling repeater jamming,an algorithm for fusing multi-domain recognition results is proposed.Firstly,the internal differences between four kinds of signals,namely,the echo of real targets and the interference of intermittent sampling (including ISDJ,ISRJ and ISLJ) are analyzed in time domain,frequency domain and fractional domain,and a Singular Spectrum Analysis (SSA) is conducted for each domain.The results of SSA can be taken as a basis for distinguishing signal differences.Secondly,three base classifiers of time domain,frequency domain and fractional domain are constructed,and the kurtosis and skewness of the singular spectrum are extracted and taken as the characteristic factors of the base classifiers for training and verification.Finally,posterior probability and local credibility are combined to construct the Basic Probability Assignment (BPA) function,and a DS fusion decision is made for the recognition results of each base classifiers.The simulation results show that the fusion method can balance the recognition results of each base classifiers.When the SNR is 0 dB and the JSR is greater than 5 dB,the average recognition rate is over 95%.

When setting the weights of the indexes of threat assessment in air combat,the objectivity of threat assessment data can′t be fully reflected,and the subjective weighting method will affect the results of weight determination.To solve the problems,a weighting method combining the AHP method with the CRITIC method is proposed.The method uses the AHP method to synthesize the experts′ judgment of index weights first,and then determines the relationship between the objective data according to the CRITIC method,so as to realize the combination weighting of threat assessment indexes.Then,the VIKOR method is used to assess the threats in air combat through the simulation of an example.Finally,taking into account the subjective preferences of different decision makers,the method sets different attribute values of the VIKOR method to obtain different results of target threat assessment for comparative analysis.The simulation results show that this method has certain practical values.

In the current testability evaluation methods,rich priori information cannot be fully utilized under the condition of small samples,which will result in a low-confidence conclusion.To solve the problem,a new and integrated testability evaluation method based on Bayes theory is proposed.Firstly,five kinds of priori information are analyzed,and the parameters of corresponding priori Beta distribution are obtained.Then,the combined weight is calculated by adopting the theory of K-L information distance,so as to obtain the mixed priori distribution.Finally,the mixed distribution is taken as the final priori distribution,so as to conduct Bayes assessment by using field test data,and an example application is given.

Error separation and conversion of the missile weapon′s guidance tools are important research contents of hit accuracy evaluation.Firstly,by adopting the working principle of Strapdown Inertial Navigation System (SINS),a real number genetic algorithm is proposed to realize error separation and conversion,and specific implementation steps are given.Then,in order to solve the problems of slow optimization and easily falling into local optimum during error separation,the corresponding strategy is used to improve the genetic algorithm,the mechanism for adjustment of cross probability and mutation probability is set up,and the elite retention strategy is used to effectively improve the accuracy of error separation and ballistic conversion.Finally,a simulated trajectory is used to verify the method.The results show that,compared with the traditional genetic algorithm,the improved genetic algorithm has faster convergence and higher separation accuracy,and the velocity difference between telemeasuring data and measurement data converted to the full trajectory is closer to the true value.

In order to meet the requirement of real-time estimation of the frequency of sinusoidal signals for the receiver of the reconnaissance radar,an improved Rife algorithm based on the approximate kernel and the approximate modulus method is proposed.This algorithm uses a 4-point approximate kernel Fast Fourier Transform (FFT) instead of full-precision FFT to eliminate multiplication operations,then uses an approximate modulus method to avoid multiplication operation and root extraction operation,which can reduce the calculation amount to a great extent.The simulation results show that the algorithm maintains the characteristics of high-precision frequency measurement in the full frequency band,and the Root Mean Square Error (RMSE) is close to the Cramer-Rao Lower Bound (CRLB).Even in the case that the Signal to Noise Ratio (SNR) is -7 dB,the mean value of RMSE of frequency estimation is 53.75 kHz;in the case that the SNR is 10 dB,the mean value of RMSE of frequency estimation is only 7.58 kHz.According to the analysis,the calculation amount of complex multiplication operation of the improved Rife algorithm is about 44% of that of the I-Rife algorithm.The calculation amount is greatly reduced,which can meet the requirements of real-time processing of the hardware.

This paper focuses on mission planning of cooperative reconnaissance for multiple UAVs in complex environment with obstacles.Firstly,several models of sensor detection,duration time of effective reconnaissance and complex polygonal obstacles are formulated respectively for different reconnaissance missions.Secondly,a two-stage framework for mission planning of multi-UAV cooperative reconnaissance in complex environment with polygonal obstacles is established by comprehensively considering the constraints of reconnaissance duration time and obstacle avoidance.The two-stage solution effectively reduces the complexity of solving the cooperative reconnaissance problem.Finally,considering the UAV kinematic constraint,a heuristic path planning algorithm based on the greedy strategy is designed,which can obtain a flying path that is as short as possible while avoiding the polygonal obstacles at the same time.The simulation results demonstrate the effectiveness of the proposed algorithm.

In order to solve the problems of low saliency value of the ships whose color are similar to sea surface color,and background interference such as coastlines and islands,a remote sensing ship detection algorithm based on ship saliency in candidate regions is proposed.The improved Frequency Tuning (FT) saliency detection and edge detection of Hessian matrix are used to obtain two types of saliency maps,which are then fused by using Pulse Coupled Neural Network (PCNN) to obtain the comprehensive saliency map,so as to improve the saliency value of the ships whose color are similar to the color of sea surface background,thus extracting effective ship slices of candidate regions.Then,the migration VGG16 network is used to extract the features of the dataset and train SoftMax classifier,so as to identify the slices of the candidate region and separate the possible background interference in the candidate region,thus realizing ship target detection.The experimental results show that the proposed algorithm has good accuracy.

To solve the problem of few correct matching points and low registration accuracy in the registration process of optical remote-sensing images and SAR images,an optical and SAR image registration algorithm combining Perona-Malik (P-M) filtering with the improved Local Self Similarity (LSS) is proposed.Firstly,the anisotropic diffusion equation is used to filter the speckle noise in the SAR image.Secondly,as to the significant gray difference between images,the extracted features of local area are used to obtain the self-similar values of each direction,the feature direction of the main direction is determined,and a stable LSS feature descriptor is formed to register the image by using the feature gradient position direction.Experimental results show that the proposed algorithm is robust to the imaging differences between optical images and SAR images,while effectively preserving the edge information of the image.Compared with that of similar registration algorithms,the registration accuracy of the proposed algorithm is improved by nearly 20%.

Classification and recognition of aerial targets by using the radar is a key technology in the field of signal processing of the radar.At present,most target recognition methods by using the radar have poor generality and high requirements on the radar.To solve this problem,this paper proposes a method of aerial target recognition based on the processing of modulation spectrum graphs by using the Convolutional Neural Network (CNN),after visualizing the periodic information of modulation spectrum of the target echo.The modulation spectrum graph of the target automatically learns the target features through the layer-by-layer transformation of the CNN,and finally the aerial targets are classified and recognized by using the classifier.This method avoids the traditional way of manually formulating the characteristics of the modulation spectrum,and realizes an end-to-end recognition of aerial targets.The experimental results show that the CNN model based on the modulation spectrum graph has higher accuracy in aerial target recognition,and the model has better robustness and generalization ability.

In the current airborne sensor management,the airspace planning problem in information-guided radar search is determined by information errors and empirical formula.The searched airspace is large,which makes the information guidance meaningless.According to the basic principles of airspace planning in radar search,the assumption of targets′ Gaussian distribution and the derivation of target interception probability are adopted,and an information-guided airspace search model is established.The model aims at minimizing the airspace to be searched,and can meet the requirements on target interception probability while satisfying radar detection constraints.The model realizes the planning of the airspace to be searched,which can meet different requirements on interception probability.The Particle Swarm Optimization (PSO) algorithm with several constraints is introduced to solve the model.The simulation results have proved the effectiveness of the model and a comparison is made with the traditional method,the genetic algorithm and the artificial fish swarm algorithm.

Compared with visual sensors,LIDAR has the advantages of strong anti-interference ability and high measurement accuracy.But it is difficult to make accurate measurements on the point cloud data which is sparse when multi-line LIDAR is far away from small obstacles.To solve the problem,YOLO (You Only Look Once) and HSV spatial color histogram matching are combined to perform long-range target detection.In the process of robot motion,when the amount of laser data in the detection area meets the requirements,the LIDAR data at this time is clustered,the feature points are extracted,and the key parameters are calculated based on the sensor calibration results,so as to complete the measurement of obstacles.A 16-line Velodyne LIDAR and an industrial IDS camera are used for verification of the method.The results show that this method can increase the data volume of LIDAR by up to 7.83 times.Even in motion scenes,it is guaranteed that the maximum width error of small obstacles is less than 2.4%,and the distance-measuring error is less than 0.15%.

When there is maneuvering acceleration,low-accuracy Attitude and Heading Reference System (AHRS) cannot measure the gravity accurately,which will cause large errors in horizontal attitude,if we still use the output of the accelerator to calculate the attitude.In order to weaken the influence of maneuvering acceleration and improve the estimation accuracy of horizontal attitude when there is maneuvering acceleration,an adaptive Kalman-filter horizontal-attitude estimation algorithm is proposed.A pseudo-measurement equation is established by taking the attitude quaternion and gyro drift as state variables,taking the quaternion attitude-update differential equation as the state equation,and taking the output of accelerator as measurement variables.Based on the residual error of accelerator output update,the variance matrix of measurement noise is estimated in real time and dynamically adjusted.This method solves the problem of horizontal attitude estimation of AHRS under the influence of maneuvering acceleration,and improves the estimation accuracy of horizontal attitude.Driving tests have verified the effectiveness of the proposed algorithm.

In order to improve the disturbance compensation ability of the photoelectric stabilized platform subjected to uncertainties and disturbances,an improved Internal Model Control (IMC) strategy based on Active Disturbance Rejection Control (ADRC) is proposed.The strategy takes the azimuth axis velocity loop simplified by using the current loop as the control object,and uses the IMC to give the desired trajectory of the system output.By introducing a robust control law based on the ADRC into the IMC scheme,the strategy can reduce the effects of model errors and external disturbances on the system,and enhance the system′s anti-disturbance ability.The design procedures of the proposed controller and the structure of the closed-loop control system are given.The high performance of the proposed strategy is validated by comparative simulations.The simulation results show that the proposed strategy has strong robustness to system parameter perturbation,and has higher stabilization precision compared with that of the standard IMC and PI.

It is difficult to determine the dividing points in the current AMESim distributed simulation,and irregular division will further affect the real-time performance of simulation.To solve the problem,this paper takes the electromechanical systems of civil aircraft in collaborative simulation as the research object,and proposes a method for the task division of collaborative simulation of electromechanical systems based on the network model and the combination weighting method.Firstly,the distributed platform based on Data Distribution Service (DDS) serves as a basis of communication,so as to ensure the synchronization of distributed simulation.Secondly,the topological attributes of the network model are introduced into the collaborative simulation of electromechanical systems of civil aircraft,and the simulation is analyzed hierarchically.The measured values of the simulation attributes are calculated by using the entropy weight method,so as to determine the edge weights.Simulation results show that the simulation task division method can reduce the calculation time of the model by 57.2% without simplifying the model.

With the development of the avionic system,ARINC 818 has become a new generation aviation video transmission bus protocol standard.Aiming at this standard,an airborne high-speed video switching system was designed.This paper introduces the aerial video transmission bus protocol standard of ARINC 818,expounds the design principle of airborne high-speed video switching system based on the standard,and introduces the design scheme of each system component in detail.A hardware test platform was built up for testing and verification.The development of the system adopted integrated and modular design.With Field Programmable Gate Array (FPGA) and high-speed crossover protocol chip as the core,a high-speed,real-time video switching output display with 8-channel ARINC 818 video input and 24-channel ARINC 818 video output,a resolution of 1920×1080@60 Hz and a link rate of 4.25 Gibit/s was designed.The design scheme has been successfully applied to multiple types of aircraft,and has important practical application value.

There are a series of complex nonlinear problems such as friction torque,inertia moment and load disturbances in the AC servo system of a vehicle-mounted stable platform driven by a Permanent Magnet Synchronous Motor (PMSM).In view of the strong anti-interference ability of Active Disturbance Rejection Control (ADRC) and the strong self-learning ability of BP Neural Network (BPNN),an Active Disturbance Rejection Controller improved by BPNN (BPNN-ADRC) is designed.In order to simplify the process of parameter tuning of ADRC,which is time-consuming and laborious,the BPNN is used to tune the important parameters of the ADRC controller online.In order to overcome the defects of slow convergence of the BPNN and its tendency of falling into local optimum,the genetic algorithm is introduced to perform online optimization of the weights of the initially connected nodes,so as to further improve the control accuracy of the system.The results of simulation experiment show that the control strategy can effectively improve the anti-interference ability of the system,which is a feasible scheme for improving the control performance of the servo system of the vehicle-mounted stable platform.

To solve the problem of lateral oscillation of small UAVs in high-speed flight,this paper analyzes all kinds of factors causing oscillation from the perspective of the principle of the flight control system, focuses on such factors as flight control parameters and system delay,and conducts simulation verification to determine the cause of the oscillation.By adopting the measures of adding an adjusting coefficient to flight control parameters and reducing the system delay in high-speed flight,the effectiveness of the improvements is verified by simulation and actual flight,and the problem of lateral oscillation is solved.

This paper proposes a high-precision passive geolocation method aiming at ground targets.This method uses a rotating ellipsoid model parallel to the surface of the earth and the posture angles of the line-of-sight of the photoelectric detector system to deduce the analytical algorithm of the high-precision passive geolocation of ground targets.In order to improve the precision of measurement and positioning,based on the motion of ground targets in the earth coordinate system,a current statistical model of the acceleration is established and taken as the state equation,the solution of passive geolocation of ground targets is taken as the measurement equation,and the method of standard Kalman filtering is used to further improve the target-positioning accuracy.The method calculates the effects of the earth′s curvature and avoids the secondary multiplication of the errors.This algorithm is simple with high precision.Compared with that of the traditional geolocation method,which conducts positioning calculation based on the exact value of laser ranging after estimating the relative distance of the target first,the positioning accuracy of the proposed method is promoted greatly.

The widely-used Constant Power Loads (CPL) have a crucial impact on the stability of the aircraft power system.The characteristics of CPL are analyzed,and an equivalent circuit model of the High-Voltage DC (HVDC) power system of More Electric Aircraft (MEA) is established.Based on the small-signal analysis method,the effects of CPL variation on the stability of the system are studied.It is concluded that an increasing CPL will reduce the system′s stability,and the capacitive loads in parallel can slow this trend.In order to protect the system,a reference voltage controller is designed,which can reduce the current in the load line and compensate for the load voltage when the system is under heavy load.Based on the Simulink platform,a time-domain simulation model is established.Simulation results have proved the correctness of stability analysis and the effectiveness of the designed controller.