【作者】 王少威；

【导师】 周理兵； 万山明；

【作者基本信息】 华中科技大学 ， 电机与电器， 2012， 博士

【摘要】 目前永磁交流伺服系统常见控制策略有矢量控制,直接转矩控制,无位置传感器控制等,也有很多新控制策略层出不穷,如智能控制、滑模控制等。由于其广泛地应用于航空航天、数控机床、机器人、工业电气自动化等行业,人们对其伺服性能要求越来越高,因此许多科技工作者还在孜孜不倦地研究伺服系统的一些关键环节和细节处理。本文主要针对矢量控制系统速度环进行一些探讨,在速度检测、速度控制器设计、转动惯量辨识等影响速度性能的环节做一些新尝试和研究。首先,针对M法中根据位置编码器计算输出速度之缺陷,提出单维卡尔曼滤波器计算速度,并在此基础上计算加速度。分析M法测速特点后,建立系统噪声方差自适应的单维卡尔曼滤波器估算电机转速,由于系统误差主要与转速及其变化程度有关系,所以构建一个含转速及其变化程度的多项式关系式近似表示自适应噪声方差。以卡尔曼滤波器输出速度为参考速度,运用软件锁相环估算加速度,避免微分运算就可计算加速度。然后,为调节基于PI控制的速度环PI系数需要辨识出转动惯量,提出用蚁群矢量移动算法同时辨识永磁交流伺服系统负载运行时的转动惯量和负载转矩,以利于速度环PI参数整定和转矩补偿。运用采样得到的d轴电流和速度序列数据,基于最小方差原理,建立蚂蚁信息素散发模型,使得蚂蚁位置与实际转动惯量和负载转矩越接近,蚂蚁散发的信息素越大。根据蚁群总信息素分布情况,计算蚁群的理想分布期望,与实际蚁群分布比较后,启发蚁群矢量移动,并朝最优方向聚集,收敛点为辨识最优解。另外,分析电机模型后,建立电机输出量的误差函数,使之含有各种待估参数,引入单层神经网络,运用梯度方法动态更新权值,再通过权值估算电机参数。接着,提出最小拍法设计速度控制器。研究用最小拍无纹波数字化方法设计永磁交流伺服系统速度控制器。把电机传动过程模型的连续域传递函数转换成数字域Z形式传递函数,根据速度指令的Z变换输入形式,事先设定预期误差传递函数,使得在有限拍内就能输出稳定的速度信号,然后依据最小拍无纹波零极点配置原则,反推出数字控制器的形式,并使速度控制器输出的电流指令信号在有限拍内就能稳定。最后,描述柔性连接引起的速度谐振问题,利用IIR和FIR方法设计陷波器,比较两种方法,并做理论仿真。本文提出的一些思路和方法有一定效果,以后研究中需要坚持不懈地完善这些方法,使之能够满足工程需求且应用于实际产品中。更多还原

【Abstract】 The permanent magnet synchronous motor (PMSM) AC servo system has employed many perfect control strategies, such as vector control, direct torque control, sensorless control, and even many novel control strategies including intelligent control, sliding mode control,etc., have been applied into the system. However, because of its widely use in aerospace, machine tools, robotics, industrial automation, electrical industry and so on, it needs more and more high performance,and many scientists are still diligently studying some key aspects and detail technologies about the servo system. In this paper,some research about velociy control based on the vector control system such as identification of the moment of inertia, the design of speed controller,etc., which affect the servo sytem’s performance, have been done through other new methods.Firstly, the angular velocity and acceleration through first-order and second-order differential calculation of the position pulses from photoelectric encoder have serious errors. Single-dimension Kalman filter with adaptive system noise variance is proposed to estimate velocity, since the noise variance is associated with velocity and its variations. Combining with virtual error model method presented by Jazwinski, a polynomial expression multiplied by exponential is established to describe the adaptive noise variance approximately. Regarding the velocity out of Kalman filter as the reference input, the software PLL-type regulator is utilized to estimate acceleration, whose integral is another estimated velocity. Regulating the optimal proportional gain and integral gain, the acceleration can be estimated without differential computation when the PLL’s estimated velocity can strictly follow the Kalman filter’s estimated velocity. The step velocity and sinusoidal velocity signals are used to verify the above algorithm. The results demonstrate that the simplified single-dimension Kalman algorithm becomes succinct, and the filtered speed becomes smooth without jump and obvious ripple phenomenon, and the acceleration has fewer spikes than that of the differential method.Secondly, the ant colony vector moving algorithm is proposed to identify the load torque and moment of inertia for load-running PMSM servo system while run-time loading, in order to adjust the PI parameters and compensate the torque. Vector moving is decomposed into horizontal and vertical directions, one for moment of inertia, the other for load torque, and every ant position means one possible solution. With the sampled sequences of d-axis current and speed data, based on the minimum variance principle, the pheromone expression model is established, making the closer distance between the ant position and actual load and inertia, the more pheromone. The expecting distribution for ant colony is calculated after the total pheromone statistics. The ant colony is inspired to move towards the optimal direction with the convergence point for the identified results. The normalized distribution for ant colony is selected to improve convergence performance when the dynamic ant colony distribution is converted to the normalized range. Additonally, after analyzing the motor model, the output quantities error function which contains the parameters to be estimated, is established. The single-layer neural network using gradient method to dynamically update the weights is proposed to achieve estimation. Learning rate affects estimation accuracy and convergence rate.Then, the deadbeat ripple-free method is proposed for the design of PMSM servo system’s digital speed controller. The continuous domain transfer function of PMSM drive model is converted into the form of discrete Z-domain transfer function. According to the speed command input in the Z-transform form, the error transfer function is pre-determined, making the speed can be output stable within a limited sample periods. Then on the basis of deadbeat ripple-free zero-pole configuration principle, the Z-form digital speed controller is derived to keep the current command from the controller become stable in several periods.Finally, to resolve the velocity resonace in the spring connection, the notch filter is designed by FIR and IIR method.The proposed ideas and methods have been verified by simulations and experiments, and the results shows that they are effective.In the follow-up study, these methods required to be constantly further improved so that they can be applied to the practical products.更多还原