【作者】 高强；

【导师】 钱林方；

【作者基本信息】 南京理工大学 ， 机械设计及理论， 2008， 博士

【摘要】 目前,我国陆军的远程火力和西方发达国家相比还有一定差距,因此,研制新型远程多管火箭炮,具有非常重要的意义。火箭炮的射击精度和反应速度依赖于火箭炮位置伺服系统的性能,所以在新型火箭炮的研制过程中,高性能的位置伺服系统的研究尤为重要。本文以某新型远程多管火箭炮为工程背景,研究了该火箭炮泵控缸电液位置伺服系统的模型辨识与控制策略。论文的主要工作包括以下几个方面:(1)分析了火箭炮泵控缸电液位置伺服系统的结构和工作原理,推导了电液位置伺服系统的传递函数,利用MATLAB中的SimMechanics和SimHydraulics工具箱搭建了系统的仿真模型,并分析了该系统的非线性和时变性因素,为下一步的控制研究和试验分析奠定基础。(2)研究了离线训练与在线微调相结合的系统辨识策略。离线辨识时,采用基于遗传优化的BP神经网络辨识方法:首先利用遗传算法优化神经网络的权值和阈值,得到优化初值,再由BP算法按负梯度方向寻优,进一步优化神经网络。该方法较好地解决了BP神经网络易陷入局部最小的问题,并且离线训练后的权值参数为合理值,从而使在线微调避免了振荡现象的发生;在线辨识时,采用附加动量项和自适应学习率相结合的快速BP算法,加速了网络的收敛速度,使其能很好的运用于在线辨识的研究中。(3)研究了泵控缸电液位置伺服系统的神经网络间接模型参考自适应控制方案。由于神经网络控制器反向传播需要已知被控对象的数学模型,而对于本文所研究的具有非线性和时变性的系统,神经网络控制器的学习修正就很难进行。为了解决该问题,采用带有神经网络在线辨识器的神经网络间接模型参考自适应控制方案,利用神经网络在线辨识器实时地为神经网络控制器提供梯度信息,使得神经网络控制器的学习修正能够正确的进行。(4)研究了泵控缸电液位置伺服系统的自适应模糊滑模变结构控制方法。该方法利用自适应模糊系统来逼近等效控制,从而解决了由于外界干扰与参数不确定性的存在使得等效控制律无法直接获得的问题。为了解决滑模变结构控制存在的抖振问题,采用了两种方法:一是利用抖振参数及切换函数的绝对值作为输入变量,设计模糊系统动态调节边界层厚度;一是以切换函数及其变化率作为输入变量,设计模糊系统动态调节控制增益。(5)设计了硬件电路和控制软件,并在半实物仿真试验台上进行了模拟试验研究,试验研究验证了本文理论与仿真研究的正确性,为系统的进一步样机制作提供了理论指导。更多还原

【Abstract】 Currently, the technology of long-distance firepower in our army still lags behind that of these developed countries. Therefore, it holds great significance to work on the new long-range multi-barrel rocket launcher .The performance of the position servo system of rocket launcher is a bottleneck , which restrict the firing accuracy and reaction rates of rocket launcher .In order to develop the new rocket launcher , the study of the new position servo system becomes important . Based on the development of a new long-range multi-barrel rocket launcher, present work mainly focuses on its identification and control strategy of the electro-hydraulic position servo system of pump-controlled cylinder (EPSSPC).The study reasons transfer function for EPSS based on the detailed analysis of the structure and working principle of EPSSPC. Using SimMchanics and SimHydraulics toolbox in MATLAB, a new simulation model of this system is proposed. The nonlinear and time-variation factors existing in the present system are analyzed, which paves the way for the controlling study and experimental analysis in the next stage.This study also introduces system identification scheme of "off-line training first, on-line minor adjustment following". When conducting off-line identification, this paper first uses the genetic algorithm to optimize the values of weights and thresholds, and then obtains an optimized initial value. Then the BP algorithm is applied to optimize at a negative gradient direction to find out an optimal values of weights and thresholds. The method of BP neural network based on genetic algorithm characterized by higher identification accuracy is employed, which offers a better solution to the problem of being prone to fall into local minimum in BP neural network . The method also avoids the oscillation in using on-line minor adjustment by keeping weight after off-line training a reasonable value. When conducting on-line identification, the fast BP algorithm containing the accessional momentum and adaptive learning rate is adopted, which accelerates the convergence of BP algorithm, makes it perform better in the on-line identification study.Furthermore, this work advances the neural network model reference adaptive control (NNMRAC) method of EPSSPC. The mathematic model of the known process plant is needed in the back propagation of the neural network controller, therefore it is very difficult to proceed the learning and modifying of neural network controller in the present system with nonlinear and time-variations features. To solve this problem, the NNMRAC scheme with the on-line neural network identifier is presented. The on-line neural network identifier offers real-time gradient information for the neural network controller, which guarantees the proper learning and modification of the controller.Moreover, this study proposes the adaptive fuzzy sliding mode control scheme of EPSSPC, in which the difficulty in directly obtaining the equivalent controller due to external disturbance and parameters uncertainties is solved by using adaptive fuzzy system to approach equivalent controller. Two means are adopted to solve the chattering reduction problem commonly found in the sliding mode variable structure control: (1) the thickness of the boundary layer is tuned online by a fuzzy system which is designed based on the chattering variable and the absolute value of the switching function; (2) the control gain is tuned online by a fuzzy system which is designed based on the switching function and its variation.Finally, the hardware circuit and controller-software are designed and the simulation experiments are conducted on the semi-physical simulation test bench, the result of which proves the accuracy of the theory and the simulation in this work, offering a reference theoretically to develop the prototype of the system in the future.更多还原