【作者】 张洪波；

【导师】 黄心汉；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 硬盘驱动器作为当今信息时代不可缺少的存储设备,在人们日常生活中正扮演着越来越重要的角色,同时它也成为信息时代科学技术飞速发展的助推器。然而,随着信息量的日益增长,人们对硬盘驱动器存储容量的要求越来越高。但另一方面由于传统硬盘驱动器的低带宽、低定位精度,导致磁头很难准确地定位在目标磁道中心位置,从而限制了存储容量的持续增加。因此,如何设计合理的伺服系统结构和先进的算法以提高磁头定位精度日益成为硬盘存储技术研究的一大挑战,本文的研究工作就是在这样的背景下展开的。就伺服系统结构而言,传统的硬盘驱动器仅采用音圈电机作为单一的致动器驱动磁头实现数据的读写。采用这种单级磁头定位结构,由于音圈电机存在机械谐振和高频不确定性,从而限制了伺服带宽的提高,进一步制约了磁道密度的增加。为了满足磁头定位速度、精度和频率响应快的要求,本文开发出两级致动器结构,即采用粗—精两级伺服定位系统以获得高伺服带宽和定位精度。在这一结构配置中,音圈电机作为第一级致动器控制读/写磁头进行粗定位,主要运动于低频段;微致动器作为第二级精定位致动器控制磁头在数据磁道上实现精确的定位,此时该级致动器工作于高频段。就先进的算法而言,首先考虑到硬盘驱动器伺服信号存在着噪声,噪声的存在降低了磁头定位的精度,进而影响磁盘密度的提高,严重情况下还会发生误读写,因此事先需要经过滤波处理。由于噪声源和有用信号的频谱相互间有重叠,因而采用一般的数字滤波器达不到滤波的效果。本文在这样的需求下提出了一种新颖的加权递推最小二乘法。为便于展开后续的硬盘两级磁头定位研究工作,先简要介绍了单级磁头定位情况,并在此基础上首次将分数阶PI~λD~μ控制器引入到硬盘伺服控制领域。分数阶PI~λD~μ控制器作为常规整数阶PID控制器在微积分阶次上的推广,除了通常的比例、积分和微分系数可调整外,其积分阶次和微分阶次也是可调整的,这使得分数阶PI~λD~μ控制器的设计更灵活,用在硬盘驱动器磁头定位上,鲁棒性更强,控制效果更佳。在硬盘两级磁头定位研究工作中,为了实现磁头更快捷、更精确的定位,同时还避免致动器的饱和,从而取得更满意的控制效果,本文基于现代控制理论,提出了一种崭新的控制方法——复合非线性反馈控制结合跟踪微分器。前者的设计思想为开始寻道时系统的阻尼比设置较小,从而保证系统有较快的响应速度。随着磁头逐渐靠近目标磁道而动态地增加系统的阻尼比进而减小超调,这一思想使得系统在获得较快上升时间的同时还保证较低的超调。后者采用前馈的方式将其引入到伺服系统中,主要用来减小输入给致动器的控制电压,从而避免致动器的饱和。此外,本文还通过构建一个恰当的Lyapunov函数,验证了在这一控制方案作用下闭环系统的渐近稳定性。考虑到信号电缆产生的常值干扰及其它不确定干扰会使系统存在稳态误差,本文提出了一种H_∞几乎干扰解耦控制器,控制器参数可以自由调整,这尤其适用于系统有不确定性干扰的情形。本文根据硬盘驱动器模型详细地设计了H_∞鲁棒控制器,并分别在时域和频域上验证了该控制器的有效性。在磁道跟踪模式下,鉴于硬盘驱动器不可避免地存在着由主轴马达产生的重复性扰动信号,扰动信号的存在使磁头很难定位在期望的磁道上。因此,本文提出一种自适应前馈扰动补偿器。这种自适应补偿器需要求解三个参变量,即相位超前参数、自适应增益和前馈项系数。为了得到最优的参数值,首先将该补偿器等效为一个线性时不变的模型,然后利用环路成形法、奈奎斯特准则和根轨迹分析法定量地获取了这些最优参量值。仿真结果表明该自适应前馈补偿器几乎完全消除了重复性扰动信号,从而保证磁头在磁道跟踪模式下良好的跟踪效果。随着硬盘磁道密度越来越大,尺寸越来越小,导致建立准确的对象模型变得越来越困难。基于此,本文提出一种不依赖于精确被控对象模型的控制方法,即采用激活函数可调的神经网络训练分数阶PIλDμ控制器的比例、积分和微分系数。这种控制方法不仅具有对被控对象模型的弱依赖性,而且具有控制性能上的高精度、强鲁棒性。最后,对全文的研究工作做了总结,并对今后要进一步开展的研究工作进行了展望。更多还原

【Abstract】 Hard disk drive (HDD), acted as requisite storage equipment in current information age, plays a more and more vital role in people’s daily life, and it becomes a roll booster in rapid development of science and technology. However, with the increase of information capacity, we put forward a severe request for HDD data storage capacity. Unfortunately, due to the low bandwidth, low positioning accuracy in conventional HDD, magnetic head is hard to be positioned onto the destination track center, thus it limits the continuing increase in storage capacity. Therefore, how to design reasonable servo system structure and advanced algorithms so as to improve head positioning accuracy becomes a challenge in HDD storage technique research, this dissertation is mainly developed based on the above background.As far as servo system structure is concerned, conventional HDD only applies with voice coil motor (VCM) as a single actuator to drive the magnetic head to realize data reading and writing. Applying this single-stage head positioning structure, due to mechanical resonances and high frequency uncertainties existed in VCM, it is hard to improve the servo bandwidth, and further increase track density. In order to satisfy the requirements of the head positioning speed, accuracy and rapid frequency response, this dissertation develops dual-stage actuator structure, that is, coarse-fine dual-stage servo positioning system. This system helps to obtain high servo bandwidth and positioning accuracy. In this configuration, VCM, acted as the first actuator, is to control the read/write (R/W) head to realize coarse positioning, and its movement is mainly limited in low frequency range. While micro-actuator (MA), acted as the secondary actuator, is to control the head to realize accurate positioning in data track, and its movement is mainly limited in high frequency range.As far as advanced algorithms are concerned, first of all, there are noises existing in HDD servo signal. The existence of noises reduces the head positioning precision, affects the improvement of disk density, and to be more serious results in track misregistration. Therefore, we need to apply filtering technique in advance. Since there are overlaps between frequency spectrum of noise sources and that of useful signals, it is hard to obtain good filtering effect by using general digital filter. This dissertation puts forward a novel weighted recursive least square (WRLS) algorithm under this requirement.In order to expand dual-stage head positioning research work, this dissertation simply presents the situation of single-stage head positioning, and then introduces fractional order PI~λD~μcontroller into HDD servo control field. This controller is the generalization of conventional integer order PID controller. Besides the usual proportional、integral and derivative parameters, the orders of integrator and differentiator can also be tunable, which make the design of fractional order PI~λD~μcontroller be more flexible. In order to verify the superior performance of the fractional order PI~λD~μcontroller, simulation experiment, using hard disk drive servo control as an example, is done. Simulation results show that this controller can make the whole system be more robust and excellent.To realize more rapid and accurate head positioning, and simultaneously avoid actuator saturation, this dissertation proposes a novel control scheme----composite nonlinear feedback control combined with tracking differentiator which is based on modern control theory. The design philosophy of the former is that the damping ratio of the overall system is set very little in the seeking stage such that the system has rapid response. When the magnetic head gradually approaches the destination track, the control scheme dynamically increases the system damping ratio so as to reduce the overshoot caused by the initial little damping ratio. This idea can not only obtain fast rising time but also ensure less overshoot. The latter is injected into the servo system using a feedforward way, and the tracking differentiator is mainly used to reduce the control voltage to the actuators and hence avoid their saturation. Furthermore, this dissertation verifies the stability of the closed-loop system under this control scheme by constructing an appropriate Lyapunov function.Considering constant disturbances caused by data flex cable and other uncertain runouts which make the servo system be steady-state error, this dissertation presents an H_∞almost disturbance decoupling controller. The parameters of this controller can be arbitrarily adjusted. This is especially suitable for those occasions which have uncertain disturbances. According to the model of the HDD, this dissertation designs this robust controller in detail, and verifies the effectiveness of this controller in time domain and frequency domain.In track following mode, there inevitably exists repeatable runout (RRO) which is caused by spindle motor, and this RRO makes the head be hard to be positioned onto the desired track center. Therefore, this dissertation presents an adaptive feedforward compensation (AFC) controller. This adaptive controller needs to solve three types of parameters, that is, phase advance parameters, adaptive gains and feedthrough coefficients. In order to obtain the optimal parameters, this time-varying compensator should be converted into a linear time invariant (LTI) model, and then quantitively acquire these optimal parameters by using loop-shaping approach, Nyquist criteria and root locus analysis. Simulation results show that this adaptive feedforward compensator nearly eliminates the repeatable runout signal, and hence ensures the servo system achieve satisfactory tracking effect in the track following mode.With the improvement of HDD track density and the reduction of HDD dimensions, the accurate model construction becomes more and more difficult. This dissertation proposes an advanced control method which is independent of plant model, to be more specific, this method is tunable activation function-multilayer forward neural network (TAF-MFNN), this neural network can be used to train proportional, integral and derivative coefficients of the fractional order PI~λD~μcontroller. This control strategy can not only have weak dependence on plant model, but also have good characteristics such as high precision and robustness.Finally, a summary has been done for all discussions in the dissertation. The research works in further study are presented.更多还原