Abstract:In order to enable high-precision accurate schemes to robustly capture shock waves and quickly obtain high fidelity simulation results of the flow field in supersonic inviscid compressible flow calculations, which usually contains complex structures such as discontinuities and small-scale vortices, a smoothness measurement algorithms based on combinations of sub-stencil derivatives were studied, and a new discontinuity detector was developed to balance accuracy and robustness, making troubled-cell recognition highly resolved for small-scale vortices. Hybrid WCNS(weighted compact nonlinear scheme) methods were studied, which could solve smooth and discontinuous regions in the flow field using linear and nonlinear weighting schemes respectively, to overcome the difficulty of a single nonlinear scheme in achieving design accuracy in smooth areas. Numerical results show that the hybrid WCNS scheme with new discontinuity detector performs well in simulating one-dimensional and two-dimensional Euler equations, and has higher computational efficiency compared to the original WCNS method that uses local characteristic decomposition in the entire flow field.

Abstract:Current multi-modal few-shot learning methods overlook the impact of inter-attribute differences on accurately recognizing sample categories. To address this problem, a multimodal cross-decoupling method was proposed which could decouple semantic features with different attributes and reconstruct the essential category features of samples, aiming to alleviate the impact of category attribute differences on category discrimination. Extensive experiments on two benchmark few-shot datasets MIT-States and C-GQA with large attribute discrepancy indicates that the proposed method outperforms the existing approaches, which fully verifies its effectiveness, indicating that the multimodal cross-decoupling few-shot learning method can improve the classification performance of identifying few test samples.

Abstract:To meet the computational requirements of safe airspace and the expected landing point in the recovery mission of the reentry capsule, a fast prediction algorithm of the flight pipeline based on the Koopman operator approach was proposed for the reentry capsule and the determination process for safe flight airspace of search and rescue helicopters was provided. The body-parachute dynamic model was constructed. A group of discrete state points was uniformly selected from the random state space by using the Halton sampling method, and the multiple possible trajectory was calculated. Based on the back pulling mechanism of Koopman operator, the initial probability density value was associated with the current state to obtain the flight pipeline and desired trajectory of the reentry capsule and its separation parts under uncertain conditions.The simulation results show that the fast prediction algorithm of the flight pipeline based on the Koopman operator approach is significantly better than the Monte Carlo method in terms of convergence speed and accuracy. After using the flight pipeline calculation results to plan the flight route of the rescue helicopter, the collision risk is reduced by 54% at most and the corresponding search time is reduced by 70%. The proposed algorithm has been successfully applied to the Chang′e-5 recovery mission.

Abstract:SEAD(suppression of enemy air defenses) is a typical application scenario of multi-UAV cooperation. Based on the characteristics of this scenario, the number of different types of UAV was also used as a decision variable in the task planning problem, fully characterizing the various constraints of the target, mission, and UAV, and establishing a heterogeneous UAV formation path problem model. A two-layer joint optimization method was designed to solve the model:the upper layer was designed with the task connection impact indicator to accurately assess the quantitative requirements of various types of UAVs and guide UAVs configuration adjustments; the lower layer improved the genetic algorithm, which can efficiently handle multiple coupling constraints and can accurately adjust the mission plan in conjunction with UAV quantity changes. The two layers coordinate with each other to obtain a UAV configuration and mission execution plan that meet the requirements. Simulation results show that the method can obtain a reasonable UAV configuration plan without traversing various UAV configurations, while obtaining an efficient and feasible mission execution plan.

Abstract:To explore the influence of BDS-3(BeiDou navigation satellite system phase Ⅲ) different frequency combination satellite DCB(differential code bias) on PPP(precise point positioning), the BDS-3 different frequency ionospheric-free combined satellite DCB correction model was derived and performed DCB correction experiments by seven different dual-frequency combinations based on the observation data of four MGEX stations during seven consecutive days. The results show that the DCB correction has a significant improvement of the PPP accuracy in the initial epoch, which helped to converge the filtering and improve the RMS of the single-day solution, but weakly in the final positioning accuracy. Although the positioning accuracy of the B2a/B3I and B2b/B3I combinations′ convergence speed are considerably thinner than other combinations, it still enhances after DCB correction. Otherwise, the positioning accuracy and convergence time of the other 5 combinations are equivalent after correction. The RMS of the static PPP single-day solution is about 5.50 cm, 2.50 cm, and 6.25 cm in the E, N, and U directions, which are approximately 20%～65% higher than before. The average convergence time, an increase of about 6%, is 38 minutes. And the kinematic PPP, an increase of about 20%, is 59 minutes. The final positioning accuracy is better than 5 cm in the horizontal component and 7 cm in the elevation component.

Abstract:The numerical experiments of three upwind schemes in contact discontinuity were carried out on the numerical dissipation under different flow field parameters, and the mechanism of numerical dissipation was analyzed. The numerical calculation results and theoretical analysis show that when the flux vector splitting scheme is used for contact discontinuity calculation, if the flow field is static or there exists a subsonic region in the flow field, the generation of density dissipation will induce numerical perturbation errors moving with characteristic velocity. These errors have no effect on the magnitude of numerical dissipation, but it will affect the distribution of velocity and pressure parameters in the flow field, thus changing the structure of the flow field. In two-dimensional flow fields, the mutual interference of the induced errors will produce numerous complex small-scale structures, which bring difficulties to the flow field structure identification. Meanwhile, in the flow field with linear distribution of density parameters, if the object reconstructed by the spatial discrete scheme is convective flux, using the flux vector splitting scheme to calculate the flow field will generate numerical errors, making it difficult to reach the second order of computational accuracy.

Abstract:In order to solve the problems of very few research literatures on the accuracy of different forecast wind intensity and the corresponding accuracy of the maximum aerodynamic load forecast value, the accuracy characteristics of different forecast wind intensity and the corresponding maximum aerodynamic load forecast value were respectively analyzed on the basis of the actual wind in a certain area, and the multiple linear regression method was used to establish a revised model. The results show that within 1～11 days of forecasting days, the greater the forecast wind intensity, the higher the forecast wind accuracy and the corresponding maximum aerodynamic load forecast value accuracy. Using the multiple linear regression method can improve the accuracy of the maximum aerodynamic load forecast, and the smaller the forecast wind intensity, the more obvious the improvement of the forecast accuracy of the maximum aerodynamic load. These findings have reference value in the flight assurance and safety decision-making of rocket launch.

Abstract:For high-definition observation of GEO(geosynchronous orbit) spacecraft, the optical satellite approaches the GEO spacecraft with continuous low-thrust and flies around the GEO spacecraft without thrust. On the deviation of Clohessy-Wiltshire (CW) equations, CW equations are improved by accommodating the non-spherical perturbation, and the quadratic terms of the nonlinearity in the differential gravitational acceleration. The secular growth of the nonlinear deviation and the most perturbation deviation have been accommodated in the ICW(improved CW) equations. On trajectory planning, we calculate the initial phase angle interval of the fly-around trajectory to ensure that the optical satellite can take pictures of the GEO spacecraft with favorable observation angles throughout the entire fly-around mission. Simulations are conducted based on CW equations and ICW equations, respectively. The simulation based on the CW equations fails, but the simulation based on the ICW equations succeeds. The total thrust required is only 4.67 m/s, which is highly feasible in engineering.

Abstract:In view of the conductive/insulating multilayer composite structure and micro/nano-size thickness of flexible thermal control film materials, a test was conducted on the charging and discharging characteristics of kapton-based secondary surface mirror film materials using 10~70 KeV electron irradiation. The key parameters such as surface charging potential and electrostatic discharge frequency were obtained. The transport law of irradiated electrons in multilayer thin film materials was studied by Monte Carlo method. The results show that there is no electrostatic discharge when the irradiated electron energy is 10 KeV due to the particularity of the kapton-based secondary surface mirror film structure. With the continuous increase of the electron energy, the surface charging potential amplitude of the film material and the electrostatic discharge frequency show a trend of first increasing and then decreasing. The space charging and discharging effect is most significant when the electron energy is 25 KeV.

Abstract:Traditional Yinyang K-means algorithm is computationally expensive when dealing with large-scale clustering problems. An efficient parallel acceleration implementation of Yinyang K-means algorithm was proposed on the basis of the architectural characteristics of typical many-core CPUs. This implementation was based on a new memory data layout, used vector units in many-core CPUs to accelerate distance calculation in Yinyang K-means, and targeted memory access optimization for NUMA(non-uniform memory access) characteristics. Compared with the open source multi-threaded version of Yinyang K-means algorithm, this implementation can achieve the speedup of up to 5.6 and 8.7 approximately on ARMv8 and x86 many-core CPUs, respectively. Experiments show that the optimization successfully accelerate Yinyang K-means algorithm in many-core CPUs.

Abstract:According to the characteristics of the heterogeneous multi-core DSP(digital signal processing) chip independently developed by National University of Defense Technology and the characteristics of the convolution algorithm, a high-performance multi-core parallel convolution implementation scheme for multi-core DSP architecture was proposed. A feature graph level multi-core parallel scheme is proposed for 1×1 convolution. For convolutions with kernels larger than 1, a window level multi-core parallel optimization design was proposed, and an element-wise vectorization based intra-core parallel optimization implementation was proposed. The experimental results show that the proposed parallel optimization method can reach a maximum single core computing efficiency of 64.95%. When the bandwidth is limited, the parallel expansion efficiency of multi-core can still reach 48.36% ~ 88.52%. Compared with E5-2640 CPU, the execution performance on the typical network ResNet50 achieves 5.39x performance acceleration.

Abstract:To address the significant computation and memory overhead of models based on attention mechanism, model compression techniques, such as collaborative optimization of quantization and pruning, were studied. A symmetric linear fixed point quantization method was proposed for four activation matrices of query, key, value and probability in the attention mechanism. Meanwhile, a probability matrix pruning method and a progressive pruning strategy were proposed to effectively reduce the pruning accuracy loss. Experimental results on different datasets show that for the typical attention-based model BERT, this optimization method can achieve 4 bit or 8 bit fixed point quantization and 0.93~0.98 sparsity with little or no accuracy loss, which greatly reduces the model computation and lays a strong foundation for accelerating the inference of quantized sparse models.

Abstract:With the increasing depth and the inconsistent length of processing sequences, the performance optimization of RNN(recurrent neural network) on different processors makes it difficult to researchers. An efficient RNN acceleration engine was implemented for the self-developed long vector processor FT-M7032. This engine proposed a row-first matrix vector multiplication algorithm and a data-aware multi-core parallel method to improve the computational efficiency of matrix vector multiplication. It proposed a two-level kernel fusion optimization method to reduce the overhead of temporary data transmission. Optimized handwritten assembly codes for multiple operators were integrated to further tap the performance potential of long vector processors. Experiments show that the RNN engine for long-vector processors is efficient, when compared with the multi-core ARM CPU and Intel Golden CPU, the RNN-like model long short term memory networks can achieve a performance acceleration of up to 62.68 times and 3.12 times, respectively.

Abstract:An efficient parallel computing method based on the characteristics of the high-performance heterogeneous accelerator and the training mode of MiniGo was proposed. The on-chip computing resources were reasonably planned to achieve pipelining parallel optimization between heterogeneous devices. The shared memory programming was designed according to the existence of shared storage segments between heterogeneous devices to reduce data transmission costs. According to the characteristics of multiple computing resources in a digital signal processing cluster, combined with the computing-memory access feature of the operators, different optimization strategies were designed. At the same time, this method provides an easy-use high-performance operator library for TensorFlow. The experimental results show that this method realizes the multi-core parallel computing of operators. The speedup of convolution was 24.69 compared with that was achieved on a single core. Compared with the cropped version of the 8-core FT2000+ CPU, the speedup of training and self-play execution on this method were 3.83 and 1.5, respectively.

Abstract:A GPU(graphic processing unit) based 5G software radio quasi cyclic LDPC (low-density parity check) code decoder was proposed. In order to save on chip and off chip bandwidth, code word shortening and punching techniques, two-stage quantization, and data packaging schemes were adopted to improve the utilization of data bandwidth. The experiment was based on the Nvidia RTX 2080Ti GPU platform to achieve parallel decoding of minimum and approximate decoding algorithms under high bit rates. By analyzing the optimal thread settings on the GPU, the decoding throughput of the 5/6 (2 080,1 760) LDPC algorithm is improved to 1.38 Gbit/s, and the decoding throughput performance is better than other GPU based LDPC decoders.

Abstract:It is of great significance to study GaN(gallium nitride) power devices and their radiation effects to meet the needs of space applications and promote the construction of a new generation spacecraft. Main structures and working principles of GaN power devices were introduced. Research progress of total ionizing dose effect and single event effect of GaN power devices in recent years was reviewed. Degradation and damage mechanisms of GaN devices caused by radiation effects were analyzed and discussed. Research results show that GaN power devices have strong resistance to total ionizing dose. However, GaN power devices are prone to leakage and single event burnout due to their weak resistance to single event burnout, and most of the burnout points occur at the drain side of the gate edge. Research on the radiation damage mechanism of GaN power devices lacks authoritative theories and needs to be further mastered to provide theoretical support for their space application. GaN power devices with planar structure are the current mainstream technical solutions. Monolithic integration, high frequency and miniaturization are the development directions of GaN power devices in the future.

Abstract:Aiming at the problem that it is difficult to discover neighbors without prior information of sector direction between nodes in the directional millimeter wave network, a neighbor discovery protocol base on blind rendezvous algorithm was proposed and the theoretical relationship among slot length, the number of sectors and the time of neighbors discovery was derived. Furthermore, in order to shorten the time of neighbor discovery base on blind rendezvous algorithm, a neighbor discovery protocol based on location prediction was proposed. Simulation results showed that the longest discovery time of neighbor discovery protocol base on blind rendezvous algorithm is less than that of ODND(oblivious directional neighbor discovery) protocol and average discovery time of neighbor discovery protocol base on blind rendezvous algorithm is less than ODND protocol when the number of sectors N is close to 2n(2n－1n,n>1,n∈Z). In addition, neighbor discovery protocol based on location prediction can effectively shorten neighbor discovery time.

Abstract:To reveal the injury mechanism of spherical fragments to human tissue, gelatin was used as a substitute for human tissue. Based on the theory of dynamic cavity expansion, the segmental motion theoretical model of spherical fragments penetrating gelatin was established by considering the velocity attenuation of spherical fragments in the stage of incomplete entering stage, and the motion law of spherical fragments penetrating gelatin was studied. The correctness of the model was verified by the experiments of steel balls and tungsten balls penetrating gelatin, and the optimal drag coefficients in the model were solved. The sources of errors in the theoretical calculation process were analyzed, and the expression of dimensionless penetration depth was derived. A sensitivity analysis of the influence of spherical fragment parameters (diameter, density and velocity) on the penetration depth was carried out by using the Sobol′ method. The results show that the motion model can well simulate the motion law of spherical fragments. The velocity attenuation of low-density spherical fragments in the incomplete entering stage cannot be ignored. The sensitivity of spherical fragment parameters to the effect of penetration depth is in descending order of velocity, density and diameter.

Abstract:Calculated weights by traditional PCM(pairwise comparison matrix) adjustment methods can′t retain the information of the original PCM to the maximum extent. Therefore, a method for PCM modulation based on the weights was proposed. The method constructed a completely consistent matrix according to the properties of weights, and the matrix was fitted by the AdaGrad method, so as to minimize the total change of the matrix relative to the original PCM. Corresponding weights was taken as the calculation result of the matrix. Numerical results show that this method can significantly improve the retention of original PCM information compared with other methods and does not require repeated consistency checking.

Abstract:For developing high-order scheme with low dissipation and low dispersion to accurately simulate the flow field with complex structures such as shock wave, a method of stencil reorganization was proposed based on the fifth-order finite difference WENO-Z scheme. A four-point central stencil recombined by three-point stencils is introduced in the calculation of WENO nonlinear weight in order to optimize the weight allocation of each template in the original format, and two improved WENO-Z schemes were proposed. The dispersion and dissipation properties of the schemes were compared and analyzed via the approximate dispersion relation analysis, which shows that the dissipation of the two improved schemes decreases at different extent. The numerical experiments show that the improved schemes have better shock capture property and higher resolution for small-scale flow structures.

Abstract:In order to improve the speed of on-chip cache and reduce the area and power consumption, an on-chip cache design method based on cooperative memory compilation and layout was proposed. This method estimated the timing margin of memory array based on its different spatial positions on chip, and then performed memory compilation through exhaustive combination of various configuration parameters such as splitting/ merging, size adjustment, threshold replacement and aspect ratio deformation. The best static random-access memory compilation configuration was selected according to the timing margin. This method was integrated with the existing physical design steps into a complete design flow. Experimental results show that this method can reduce the power consumption by about 9.9% and the critical path delay by 7.5%.

Abstract:It is the key for evaluating and prolonging the life of ammunition to acquire the temperature variation law of ammunition storage microenvironment efficiently and accurately. With the help of the neural network and the heat transfer theory, a model for predicting the microenvironment of ammunition storage was studied, and based on the model, a set of ammunition packaging box prediction software was developed to predict the temperature changes inside the box under different climatic environment. Relevant test was progressed to verify the software correctness and parameter optimization in Dunhuang and Mohe. Results show that with the help of the microenvironment prediction software of packing box developed on the basis of the neural networks algorithm, the microenvironment temperature control can be achieved by adjusting the storage location and storage initial temperature according to different packaging box materials.

Abstract:In order to enrich the optimization theory of hubless pump-jet thruster noise performance, the influence of blade number on the performance of hubless pump-jet thruster was studied, and the variation law of pressure pulsation in hubless pump-jet thruster with different number of blades was emphatically investigated. Based on the finite volume element method of computational fluid dynamics, the RANS equation was discretized, and the calculation model was closed by using RNG k-ε turbulence model, coupling the pressure-velocity by using the SIMPLEC. The unsteady numerical simulation of the flow field inside the hubless pump-jet thruster with different number of blades was carried out, and the calculated results were compared and analyzed in detail. The study results show that with the increase of blade number, the head, efficiency and thrust of hubless pump-jet thruster model increase gradually. The pressure distribution on the blade surface becomes more uniform. The pressure pulsations of the impeller rim and the middle part decreased obviously. The amplitude of pressure pulsations at the main frequencies of impeller inlet and outlet decreases gradually from rim to center.

Abstract:In order to comprehensively evaluate the survivability of launch vehicles in combat, a weapon equipment evaluation space was established by distinguishing parameters-performance-capability-effectiveness at four levels. In view of the survivability assessment of launch vehicles, graded five types of natural environment, three levels of threat environment and three kinds of combat flow status, the impact of external environment was taken as the necessary conditions of effectiveness evaluation, and an evaluation framework of three dimensions of object, content and level was constructed.The detection probability and damage probability were used to characterize the camouflage ability and anti-damage protection ability of launch vehicles, and the continuous transfer probability matrix of launch vehicles′ combat status was established. A comprehensive evaluation model of launch vehicle based on improved availability, dependability, capability model under typical combat flow and graded environment was proposed. Through simulation analysis, the survivability of various launch vehicles varies significantly in different environments and combat status, the model can dynamically evaluate the survivability of launch vehicles from multiple dimensions of performance space, environment space and combat flow space, and put forward optimization and quantitative suggestions for the performance design and battlefield application of launch vehicles.