【作者】 张翠芳；

【导师】 邹继斌；

【作者基本信息】 哈尔滨工业大学 ， 电机与电器， 2008， 硕士

【摘要】 感应同步器测角系统是一种采用电磁感应原理的角度测量设备,高精度的感应同步器测角系统输出的位置和速度信号可以提高伺服系统的控制精度。感应同步器的测量精度主要取决于感应同步器的精度、信号放大与转换模块的精度。为了提高测角系统的测量精度,要对测量误差进行补偿。硬件补偿存在一定的局限性,采用软件补偿成为提高测角系统的精度的主要手段。人工神经网络尤其是基于误差反向传播算法的多层前馈网络,广泛应用于非线性建模、函数逼近、模式分类等。本文主要对鉴幅型感应同步器测角系统及基于BP(Back Propagation)网络的误差补偿方法进行了深入研究。首先,进行了感应同步器测角系统的设计,包括硬件电路设计,软件设计。硬件电路中模拟部分实现对感应同步器输出信号放大、滤波和相敏解调等处理,数字部分主要实现模数转换、数据采集等功能。软件部分主要是进行数据处理与转换。文中详细介绍了各部分电路的设计原理及性能分析,尽量减少硬件电路带来的测量误差。然后,将BP神经网络的理论用于感应同步器的误差补偿。分析了BP神经网络学习算法的优缺点,针对一般BP算法收敛速度慢,易陷入局部极小值的缺陷,采用对标准BP算法改进的措施,以加快收敛速度。将改进的算法的BP神经网络用于测角系统误差补偿。以实测的0°~360°之间720点零位误差数据为基础,分析了测角系统零位误差特征,以此数据为样本训练并建立BP神经网络模型。仿真结果表明,这种改进方案不仅能够提高BP算法在训练过程中的收敛速度,而且训练后的BP神经网络具有较强的自适应和自学习能力,实践结果表明,基于BP神经网络的测角系统的零位误差补偿效果明显。更多还原

【Abstract】 Digital read-out inductosyn transducer is a kind of angular measurement system using elector-magnetic theories. The control precision of servo can be improved by high precision angular measuring system of inductosyn. The precision of inductosyn angular measuring system are decided by the precision of inductosyn itself, and also the precision of signal amplification and conversion module. In order to improve the measurement precision, it needs compensate the measure error. Because hardware compensation has some limit, software compensation has become the main method to improve measurement precision. Artificial neural network especially the multiple-layer feed forward network based on back propagation can approximate random continuous function with random precision. It is widely used in nonlinear model building, function approximation and model classification, etc. After introducing the development, status quo, basic theory of neural network, this thesis mainly studies the structures and algorithms of Back Propagation neural network and its application to error compensating of amplitude discrimination mode of inductosyn.Firstly, the design of error measuring system of the inductosyn includes the design of hardware circuits, software and the technique of anti-jamming. In the hardware circuits, the analog circuits consist of signal amplifying circuit, the active filter and the phase demodulator, the digital circuits consist of the analog-digital converter and the circuit of data collection. The theory of design of the hardware circuits and the analysis of the circuit performance are introduced in detail, which can decrease the measurement error brought by hardware circuit as much as possible.And then, it applies the trained BP neural network using improved algorithm in the angular measurement system’s error compensation. To deal with the defects of the steepest descent in slowly converging and easily immerge in partial minimum frequently, this paper improves BP algorithm in some extent to accelerate the rapidity of convergence and achieve optimization. And it also applies the trained BP neural network using improved algorithm in the angular measurement system’s error compensation. Based on the 720 points of pitch error in the range of 0°~360°gained by test, it analyzes the pitch error characteristic of angular measurement system, and the data also be used as sample to be trained and build model. The simulation results show that this improved algorithm not only increases the convergence speed in the training process, which has rather strong capabilities of adaptive and self-study. The validity of this error model on the BP network has been verified by simulation.更多还原