Search

For the issue of limited filtering accuracy of interactive multiple model particle filter algorithm caused by the resampling particles don't contain the latest observation information, we made improvements on interactive multiple model particle filter algorithm in this paper based on mixed kalman particle filter algorithm. Interactive multiple model particle filter algorithm is proposed. In addition, the composed methods influence to tracking accuracy are discussed. In the new algorithm the system state estimation is generated with unscented kalman filter (UKF) first and then use the extended kalman filter (EKF) to get the proposal distribution of the particles, taking advantage of the measure information to update the particles' state. We compare and analyze the target tracking performance of the proposed algorithm of IMM-MKPF in this paper, IMM-UPF and IMM-EPF through the simulation experiment. The results show that the tracking accuracy of the proposed algorithm is superior to other two algorithms. Thus, the new method in this paper is effective. The method is of important to improve tracking accuracy further for maneuvering target tracking under the non-linear and non-Gaussian circumstances.

With the widespread using of GPU hardware facilities, more and more distributed machine learning applications have begun to use CPU-GPU hybrid cluster resources to improve the efficiency of algorithms. However, the existing distributed machine learning scheduling framework either only considers task scheduling on CPU resources or only considers task scheduling on GPU resources. Even considering the difference between CPU and GPU resources, it is difficult to improve the resource usage of the entire system. In other words, the key challenge in using CPU-GPU clusters for distributed machine learning jobs is how to efficiently schedule tasks in the job. In the full paper, we propose a CPU-GPU hybrid cluster schedule framework in detail. First, according to the different characteristics of the computing power of the CPU and the computing power of the GPU, the data is divided into data fragments of different sizes to adapt to CPU and GPU computing resources. Second, the paper introduces the task scheduling method under the CPU-GPU hybrid. Finally, the proposed method is verified at the end of the paper. After our verification for K-Means, using the CPU-GPU hybrid computing framework can increase the performance of K-Means by about 1.5 times. As the number of GPUs increases, the performance of K-Means can be significantly improved.

Aiming at the distance measurement of moving sound sources in shallow seas, this paper proposes a method of histogram filtering to realize underwater distance estimation of moving sound sources in shallow seas. The algorithm used the transmission loss, target motion parameter in the sound propagation and receival signal as prior knowledge to updated the state vector of the sound source, so as to realize the distance estimation of the shallow sea sound source, and this paper used SwellEx-96 database for experimental verification. The experimental results shown that: the depth estimating error of moving sound source is small, and when the detected horizontal distance is in the range of 10 km, the maximum range error of the horizontal distance is ±10 m, meanwhile the accuracy of ranging can be improved by improving the prior knowledge of the target motion parameters, which verifies that the histogram filtering algorithm can achieve better ranging for underwater moving targets.

Compared with traditional gear transmission, the multi-branch split-torsion gear transmission system has the advantages of large transmission power, small size and high reliability, so it is more and more used in high-speed heavy load occasions such as ships and aircraft. Since the transmission system of multi-branch split torsional gears belongs to over-constrained configuration, it is necessary to meet strict tooth matching condition in the design process in order to realize the correct synchronous meshing of each branch, which is of great significance to ensure its uniform installation and motion synchronization.Aiming at the coaxial six-branch twisted herringbone gear transmission system, this paper establishes a calculation method for the proper meshing conditions of each branch on the basis of considering the movement synchronization of each branch and preventing geometric interference.In addition, the calculation of gear allocation was carried out for a ship's power transmission system, and a parameter scheme that satisfies the requirements of transmission ratio, concentricity and synchronous meshing was obtained.The correctness of the calculation method of tooth matching in this paper is verified by three-dimensional modeling. This method has universal application value to the tooth matching design of other coaxial multi-branch gear transmissions.

Most of the solutions to existing test selection problems are based on single-objective optimization algorithms and multi-signal models, which maybe lead to some problems such as rough index calculation and large solution set limitations. To solve these problems, a test optimization selection method based on NSGA-3 algorithm and Bayesian network model is proposed. Firstly, the paper describes the improved Bayesian network model, expounds the method of model establishment, and introduces the model's learning ability and processing ability on uncertain information. According to the constraints and objective functions established by the design requirements, NSGA-3 is used to calculate the test optimization selection scheme based on the improved Bayesian network model. Taking a certain component of the missile airborne radar as an example, the fault detection rate and isolation rate are selected as constraints, and the false alarm rate, misdiagnosis rate, test cost, and test quantity are the optimization goals. The method of this paper is used for test optimization selection. It has been verified that this method can effectively solve the problem of multi-objective test selection, and has guiding significance for testability design.

This paper selects SST k-w turbulence model and VOF wave to construct a numerical calculation model of moving body planning on a flat free surface based on STAR-CCM+ numerical simulation software. The construction model is checked through foreign classic literature, and the numerical simulation results are in good agreement with the experimental results. The hydrodynamic numerical errors are less than 5%, which is within the engineering error range. The model can be used for the numerical simulation of the planning cylinder. In this paper, it is used to simulate the planing process of cylinder with different speeds and different submerged depths, and the flow field characteristics and hydrodynamic characteristics in the planing process are obtained. The results show that waves appear at the tail and the tail liquid splashes to form a water splash during the planing process of the cylinder on a flat surface. The higher the speed of the planning cylinder is, and the deeper the submersion depth, the more pronounced waves at the tail. When the cylinder has a Fr number Cv≥8, the hydrodynamic force of the cylinder is almost unchanged, and it is not affected by the speed. But when Cv=3, the hydrodynamic characteristic coefficient is higher. The drag coefficient is 20% higher than that in the high-speed (Cv≥8) planing process. The lift coefficient is 3 times of high-speed planing lift coefficient. It is related to the surface pressure and frictional force distribution of cylinder during the low-speed planing. There is a linear relationship between the drag coefficient and the submerged depth during the cylinder planing at different submerged depths.

The fault detection function of GLT method is approximately proportional to the square of fault value, the value of the fault detection function is lower than the detection threshold when the soft fault is small, which leads to the low sensitivity of soft fault detection and the high rate of missing detection. The IM-SPRT method rapidly increases the value of fault detection function above the threshold of fault detection by accumulating small soft fault values, so it has high sensitivity for soft fault detection, however, the end time of the fault cannot be judged timely, resulting in a high false alarm for fault detection. In view of the above problems, this paper proposes a high-sensitivity fault detection method for redundant systems based on GLT/IM-SPRT. This method uses the IM-SPRT method to detect weak and soft faults and sets the IM-SPRT method fault detection function value to zero by the GLT fault detection result, and isolates the fault with GLT. On this basis, a framework of high-sensitivity fault detection and isolation methods for redundant ADS and INS systems is designed. The simulation results of soft and hard fault detection and isolation of three-redundancy ADS subsystem level and INS sensor level show that the method in this paper can improve the sensitivity of fault detection and reduce the rate of missed detection.

Path planning is the key technology for UAV to achieve autonomous flight. Considering the shortcomings of path planning based on the conventional potential field method, this paper proposes an improved optimization algorithm based on the artificial potential field method and extends it to three-dimensional space to better achieve flight constrained 3D online path planning for UAVs. The algorithm is improved and optimized aiming at the three problems of goal nonreachable with obstacle nearby (GNWON), easy to fall into local minimum, and path oscillation in traditional artificial potential field method. First, an improved potential field function with relative distance is used to solve the GNWON, and an optimized repulsive potential field calculation method based on different obstacles or threat models is proposed to optimize the planned path. Secondly, in order to make the drone jump out of the local minimum trap, a method of setting heuristic sub-target points is proposed. For local path oscillation, a method using memory sum force was proposed to improve the oscillation. The simulation results show that the improved optimization algorithm in this paper effectively makes up for the shortcomings of the traditional artificial potential field method, and the designed 3D online path planning algorithm for the UAV is practical and feasible.

In the case of nonlinear systems with random bias, the Optimal Two-Stage Unscented Kalman Filter (OTSUKF) can obtain the optimal estimation of system state and bias. But it requires random bias to be accurately modeled, while it is always very difficult in actual situation because the aircraft is a typical nonlinear system. In this paper, the faults of the Inertial Measurement Unit (IMU) are treated as a random bias, and the random walk model is used to describe the fault. The accuracy of the random walk model depends on the degree of matching between the covariance of the random walk model and the actual situation. For the IMU fault diagnosis method based on OTSUKF, the covariance of the random walk model is assigned with a constant matrix, and the value of the matrix is initialized empirically. It is very difficult to select a matching matrix in practical applications. For this problem, in this paper, the covariance matrix of the random walk model is adaptively adjusted online based on the innovation covariance matching technique, and an adaptive Two-Stage Unscented Kalman Filter (ATSUKF) is proposed to solve the fault diagnosis problem of the IMU. The simulation experiment compares the IMU fault diagnosis performance of OTSUKF and ATSUKF, and verifies the effectiveness of the proposed adaptive method.

Aiming at the requirement of attitude information module with high precision, small size and low power consumption for the control of miniature UAV, a practical attitude estimation algorithm based on the micro-electro-mechanical sensor is proposed in this paper, which realizes the accurate estimation of the attitude of the UAV under the condition of low acceleration. A low-cost MEMS gyroscope, accelerometer, and magnetometer are used in the system. The Euler angle is obtained by the state observer method based on Direction Cosine Matrix (DCM) which can be got by fusing the sensor data. Firstly, based on the basic idea of TRIAD algorithm, a method to determine the attitude rotation matrix by accelerometer and magnetometric measurement is proposed. Compared with the traditional method, this method does not have to calculate the inverse of the matrix. Secondly, a state observer is intended to estimate the attitude of the system. The state observer doesn't have to observe the bias of the gyroscope, but still ensures the convergence of the Euler angle. Finally, the simulation based on the actual sampling data of the MEMS sensor shows that the output of the state observer designed in this paper still has high accuracy and good dynamic characteristics under the condition of gyroscope noise and bias.