RBF neural networks hysteresis modelling for piezoceramic actuator using hybrid model

An radial basis function (RBF) neural networks rate-dependent hysteresis hybrid model for piezoceramic actuator is proposed. The piezoceramic actuator cannot be described by neural networks like the back propagation (BP) static neural networks because of its multi-valued hysteresis non-linearity. The proposed hybrid hysteresis model consists of hysteresis-like non-linearity in series with a dynamic RBF neural networks used for implementing non-linear transformations of the phase lag and non-linear magnitude. The hysteresis-like non-linearity model, which is composed of the previous output of piezoceramic actuator and input signal, differs from the hysteresis behaviour of piezoceramic actuator in only ways of their phase and magnitude, and it is used to describe the non-smooth behaviour of piezoceramic actuator. The results of both simulation and experiment show that the new modelling approach is very effective and of higher precision under a decayed input signal with the varying frequency.     

A rate-dependent Prandtl-Ishlinskii model for piezoelectric actuators using the dynamic envelope function based play operator

In this paper, a novel rate-dependent Prandtl-Ishlinskii (P-I) model is proposed to characterize the rate-dependent hysteresis nonlinearity of piezoelectric actuators. The new model is based on a modified rate-dependent play operator, in which a dynamic envelope function is introduced to replace the input function of the classical play operator. Moreover, a dynamic density function is utilized in the proposed P-I model. The parameters of the proposed model are identified by a modified particle swarm optimization algorithm. Finally, experiments are conducted on a piezo-actuated nanopositioning stage to validate the proposed P-I model under the sinusoidal inputs. The experimental results show that the developed rate-dependent P-I model precisely characterize the rate-dependent hysteresis loops up to 1000 Hz.     

Modeling of piezoelectric actuator for compensation and controller design

This work proposes a novel method for describing the hysteretic non-linearity of a piezoelectric actuator. The hysteresis behavior of piezoelectric actuators, including the minor loop trajectory and the residual displacement near zero input, are modeled by a set of hysteresis operators, including a gain and an input-dependent lag, as well as the parameter scheduling method. A hysteresis model, using the identified parameters, and containing only the dominant hysteresis operator, is presented herein. Based upon a simplified hysteresis model, tracking is controlled to reduce the non-linear effects in the characteristics of the piezoelectric actuator. A proportional-integral (PI) controller, with inverse model feed-forward, suppresses the tracking error to within ±1% full span range (FSR) of the actuator, noticeably improving the tracking performance of the piezoelectric actuator.     

压电执行器动态迟滞建模与LQG最优控制器设计

为提高空间天文望远镜稳像系统中压电快摆镜(Fast Steering Mirror,FSM)的动态性能,对压电执行器(Piezoelectric Actuator,PZT)动态迟滞补偿和控制进行研究。鉴于基于广义Play算子Prandtl-Ishlinskii(PI)模型的求逆复杂性和迟滞曲线的非对称性,构造一种基于广义Stop算子PI逆模型来补偿压电执行器迟滞非线性。采用Hammerstein模型对压电执行器动态迟滞特性进行建模,以广义PI模型和自回归遍历模型(Auto-regressive Exogenous Model,ARX)分别表征Hammerstein迟滞模型中的静态非线性和率相关性,并针对迟滞率相关模型不确定性问题,提出一种前馈补偿和线性二次型Gauss最优控制算法(Linear Quadratic Gaussian,LQG)相结合的复合控制策略。利用自适应差分进化算法(Adaptive Differential Evolution algorithm,ADE)辨识和整定模型及控制器参数。实验结果表明:该动态迟滞模型能够有效描述1～100Hz频率范围内压电执行器迟滞曲线,拟合均方根误差为0.077 1μm(@1 Hz)～0.512 3μm(@100Hz),相对误差为0.31%(@1Hz)～2.09%(@100Hz);实时跟踪幅值为24.5μm的变频目标位移,LQG控制算法的跟踪精度相比于直接前馈控制和PID控制分别提高48.6%和27.02%。     

压电微定位台的率相关动态迟滞建模及参数辨识

针对压电微定位台固有的率相关迟滞非线性严重限制其微定位精度的问题,研究了基于Backlash-Like的Hammerstein率相关迟滞非线性模型及其建模方法。以改进的Backlash-Like分段辨识模型描述压电微定位台的静态非线性特性,结合ARX(Auto Regressive eXogenous)模型,建立描述压电微定位台的率相关动态迟滞模型。同时,针对传统的粒子群算法(Particle Swarm Optimization,PSO)进行模型参数辨识时易陷入局部最优的问题,提出一种具有交叉变异策略的改进型粒子群算法进行模型的参数辨识。实验结果表明:与传统的Backlash-Like模型相比,改进的Backlash-Like分段辨识模型在输入电压为60V,频率为2Hz的信号时,模型辨识的最大误差由0.68μm下降到了0.104μm,最大相对误差由2.69%下降为0.35%。当压电微定位台输入电压为60V,频率分别为30Hz,60Hz和90Hz的单频信号时,Hammerstein率相关迟滞模型较Backlash-Like分段辨识模型,均方根误差由0.393 1～0.700 6μm下降至0.054 1～0.190 4μm,相对误差由1.721%～3.087%下降至0.236%～0.831%。验证了基于改进Backlash-Like的Hammerstein率相关迟滞模型较传统的Backlash-Like静态迟滞模型能精确地描述压电微定位台的率相关动态迟滞特性,具有较好的频率泛化能力,提高了压电微定位平台的定位精度。     

菱形微位移压电作动器输入输出杂交建模

针对某定位装置,研究了一种新型菱形微位移压电作动器,该压电作动器由压电堆、菱形位移放大机构以及柔性铰链组成。菱形微位移压电作动器的核心驱动部件为压电堆,由于压电材料的迟滞特性,菱形压电作动器具有非线性迟滞特性。为了消除迟滞对压电作动器在后续控制中的影响,发展了一种Preisach杂交建模的方法,该方法在传统Preisach模型的基础上,有效结合了Preisach离散模型和支持向量机(support vector machine,简称SVM),建立了微位移压电作动器输入输出杂交模型。试验结果表明,SVM有效解决了因1阶滞回曲线数量不足而导致Preisach模型精度低的问题,同时与传统Preisach模型相比,杂交建模能更准确地描述迟滞特性,具有更高的精度。     

Second-order sliding mode tracking control for the piezoelectric actuator with hysteretic nonlinearity

For the piezoelectric actuator model with hysteretic nonlinearity, a second-order sliding mode tracking controller is proposed. First, a second-order nonlinear dynamic model is introduced for the piezoelectric actuator. Next, a second order sliding mode control law with dual-phase sliding movement is designed. Then, the system stability is proved with a theorem. From theoretical analysis, the feedback control system is stable in the sense that all signals involved are bounded. The simulation results show the validity of the proposed method for this kind of nonlinear dynamic model of the piezoelectric actuator.     

压电驱动器迟滞非线性的分数阶建模及实验验证

存在于压电陶瓷全工作范围内的迟滞非线性特性,往往会导致压电陶瓷执行器的系统精度下降、振荡,甚至造成系统的不稳定。针对周期性的正弦输入信号,提出一种基于分数阶算子的迟滞建模方法。首先,在分析压电特性和分数阶算子特性的基础上,采用结构简单参数少的分数阶算子来描述压电陶瓷的迟滞特性;然后,搭建了基于dSpace的压电驱动微位移定位实验平台;最后,将基于分数阶算子的迟滞建模方法应用于压电驱动微位移定位平台中,对压电陶瓷的迟滞非线性特性进行辨识。实验结果表明采用基于分数阶的迟滞模型(FOM)比传统的Prandtl-Ishlinskii模型(PIM)及其改进的增强型Prandtl-Ishlinskii模型(EPIM)更有优势;在低频段,FOM模型比PIM模型和EPIM模型精度略有提高,但是在高频段,FOM模型比PIM模型和EPIM模型精度则提高显著。在输入频率为100HZ的情况下,所提出的FOM模型较PIM模型的均方根误差(RMSE)值精度提高69.84%,较EPIM模型的RMSE值精度提高68.88%。     

Adaptive identification of hysteresis and creep in piezoelectric stack actuators

The adaptive identification of the non-linear hysteresis and creep effects in a piezoelectric actuator is proposed in this paper. Model uncertainties related to the hysteresis and creep effects, most prominently in the high frequency zone (to 100 Hz), large operating amplitude and/long operating time, can make a piezoelectric actuator-driven micro-positioning system unstable in the closed loop. Furthermore, these uncertainties may lead to inaccurate open-loop control and frequently cause harmonic distortion when a piezoelectric actuator is driven with a sinusoidal input voltage signal. In order to solve the above issues, it is important to determine an accurate non-linear dynamic model of a piezoelectric actuator. An unscented Kalman filter-based adaptive identification algorithm is presented, which accurately determines the non-linear dynamics of a piezoelectric stack type actuator such that the non-linear hysteresis and creep effects can be accurately predicted. Since hysteresis and creep are dominant in open loop, the actuator is driven in an open-loop mode in this investigation.     

PBP驱动器率相关迟滞特性研究及其线性化控制

后屈曲预压缩压电双晶片(Post-buckling pre-compression,PBP)驱动器作为一种大行程压电舵机驱动器,存在着严重的率相关迟滞现象。为了使PBP驱动器能够作为具有较高控制精度的微小型飞行器舵机驱动器,利用基于Bouc-Wen模型的Hammerstein率相关迟滞模型对其进行参数识别,并通过试验验证了该模型能够较好地预测PBP驱动器的率相关迟滞特性;在此基础上为PBP驱动器设计一种具有在线自适应能力的前馈单神经元PID复合线性化控制器,在多种单复合频率信号作用下对其控制回路进行位移跟踪半实物仿真试验,并与基于径向基函数(Radial basis function,RBF)神经网络PID控制器进行对比,结果表明前馈单神经元PID控制器具有更快的响应速度和更高的控制精度。     

压电陶瓷驱动微位移机构迟滞非线性改善方法研究

压电陶瓷驱动器在精密微位移系统中有着广泛的应用,但其固有的迟滞非线性严重影响系统的运动精度,使控制困难.通过数学建模方法建立压电陶瓷的迟滞曲线逆模型,并用此模型对压电陶瓷进行位移补偿,可以有效减小压电陶瓷迟滞非线性对系统精确控制的影响.提出一种用两个压电陶瓷协同驱动微动平台运动的微位移机构,通过基于迟滞逆模型的开环控制方法,对压电陶瓷位移进行误差补偿,可以实现微动平台运动与控制信号较好的线性对应关系.     

工业机器人谐波减速器迟滞特性的神经网络建模

谐波减速器中柔性环节与传动的非线性摩擦,导致谐波传动出现了不可避免地影响传动精度的复杂迟滞特性,为了描述谐波减速器的迟滞特性,本文构建了一个结构简洁的神经网络迟滞混合模型。该模型由类迟滞特性预处理环节和动态RBF神经网络两部分组成:对输入信号进行类迟滞预处理,处理后的信号与输入信号之间具有类迟滞特性;充分利用动态RBF神经网络实现类迟滞到谐波减速器迟滞特性的高精度映射。根据本文搭建的实验平台,在不同实验条件下获得的数据进行建模验证,在不同频率输入信号、不同负载,实现相同建模精度下,神经网络迟滞混合模型的验证精度为0.449 6(MSE),远高于经典RBF神经网络模型的3.032 1(MSE)精度,证明了所构造的神经网络迟滞混合模型的有效性和适应性。     

Observer-based sliding mode control with adaptive perturbation estimation for micropositioning actuators

Control of piezoelectric actuators is under the effects of hysteresis that could affect actuators micropositioning accuracy. In this paper a modified Prandtl-Ishlinskii (PI) operator and its inverse is utilized for both identification and real time compensation of the hysteresis effect. As a result, the actuator dynamic model would be transformed to the second order linear dynamic model. Considering the parametric uncertainties, PI estimation error and probably unmodeled dynamics, a variable structure controller coupled with adaptive perturbation estimation is proposed for trajectory tracking of the piezoelectric position. Considering the very noisy output of the actuator, a high-gain observer would estimate full states from the only measurable position trajectory. The stability of the controller in the presence of the estimated state is demonstrated with the Lyapunov criterion. Comparing to the widely used proportional-integral controller, the experimental results depicts that the proposed approach is greatly achieved in precisely tracking of multiple frequency trajectories.     

Preisach迟滞逆模型的神经网络分类排序

为了补偿影响压电陶瓷执行器纳米定位系统精度的迟滞非线性,提高系统的控制精度,开展了基于压电陶瓷执行器的迟滞非线性逆模型的研究。兼顾到迟滞的擦除特性和建模的精确度,提出了一种Preisach逆模型分类排序法的神经网络实现方法,用神经网络取代了传统的反查值方法,以避免插值误差。建立三层BP神经网络,运用实测数据进行训练,确定各层权值;然后,结合排序得到的电压和位移极值信息,通过神经网络方法拟合出较精确的输入电压值。运用若干组实验数据检验了此逆模型的有效性,结果表明,该神经网络的实现方法将逆模型的平均误差降低到了1.5V以下,最大误差绝对值降低到了2.7V以下。与反查值方法相比,神经网络实现方法有效提高了压电陶瓷执行器纳米定位系统的迟滞逆模型的精度。     

压电陶瓷执行器的神经网络实时自适应逆控制

目的：为了提高压电陶瓷执行器执行精度,提出消除压电陶瓷的非线性、非光滑的迟滞特性的方法。 方法：提出了基于内积的压电陶瓷动态神经网络非线性、非光滑的迟滞逆模型,采用反馈误差学习方法,避免了求取压电陶瓷的Jacobian信息,快速地在线得到压电陶瓷的逆模型,并结合PID反馈控制,在dSPACE系统平台上,实现压电陶瓷的神经网络自适应逆控制,为了提高实时性,程序采用效率高、速度快的C-MEX S Function编程。结果：实验结果表明：神经网络自适应逆控制的控制精度为：0.13μm,而PID控制精度为：0.32μm 。结论：所提出方法有效地消除了迟滞的影响,控制精度高。     

基于改进PI模型的压电陶瓷迟滞特性补偿控制

压电陶瓷执行器的迟滞特性会降低星间激光通信精瞄系统的定位精度,对信标光的捕获以及链路的稳定性造成影响。针对这一问题,通过分析压电陶瓷执行器迟滞特性产生机理,提出一种基于PLAY迟滞算子的改进(Prandtl-Ishlinskii,PI)数学模型及其辨识方法,并利用该模型对压电陶瓷执行器迟滞特性进行前馈线性化逆补偿。并通过实验来验证数学模型和线性化的有效性,实验结果表明,通过对系统输入不同频率下等幅和减幅正弦控制信号来验证前馈逆补偿性能时,改进的模型最大拟合误差均在1%之内,通过前馈模型逆补偿的控制方法可使压电陶瓷驱动的线性度误差由5%减小到1%以内,并且改进PI模型在计算复杂度上由O(n)简化为O(1)。     

压电陶瓷迟滞非线性前馈补偿器

为了有效补偿压电陶瓷的迟滞非线性,提出了基于STOP算子的改进PI模型以改善传统基于PLAY算子的PI模型解析求逆的复杂过程以及通过插值算法求逆的大量耗时。介绍了传统的基于PLAY算子和基于STOP算子的PI模型,然后基于STOP算子的叠加形式建立了以预期位移为输入,以控制电压为输出的PI模型,并将这一模型直接作为前馈控制器补偿压电陶瓷的迟滞效应。为了更好地平衡全局寻优与局部寻优能力,对粒子群优化算法进行了改进,利用其辨识出各算子的权值。最后,利用实验的方法验证了改进的PI模型对迟滞非线性的补偿效果。进行了两组实验测试,结果显示:无论对于规律变化还是随机变化的输入,提出的改进PI模型都可以很好补偿迟滞非线性,跟踪误差可控制在1%以内。因此,基于STOP算子的改进PI模型在压电陶瓷控制领域中具有很好的实用价值。     

纯电力加载下压电陶瓷内环迟滞特性的实验研究

以典型压电陶瓷堆叠执行器为研究对象,对纯电力加载作用下压电陶瓷的内环迟滞行为进行了实验研究和分析。实验采用Sawyer-Tower方法获得了执行器的极化强度,利用高精度电容式位移传感器测量了执行器的输出位移,结合外围实验设备,通过计算得到了堆叠执行器在输入电压频繁折返过程中的电压-极化和电压-应变曲线。从压电陶瓷内部晶体的电畴转向角度出发,对其行为特性的产生机理进行了分析,给出了压电陶瓷堆叠执行器的建议工作区域。     

基于灰色理论的压电陶瓷迟滞特性的神经网络建模研究

针对不光滑非单值对应的非线性压电陶瓷的迟滞特性,提出了采用灰色理论中的累加生成(AGO)方法,将压电陶瓷的非负多值对应的输入与输出特性的数据转换为单值对应的新数据,再通过神经网络建立压电陶瓷的迟滞特性模型。仿真实验结果表明,用该方法建立迟滞特性对象的模型是有效的。     

Sliding mode control of the inchworm displacement with hysteresis compensation

This paper proposes a new modeling scheme to describe the hysteresis and the preload characteristics of piezoelectric stack actuators in the inchworm. From the analysis of piezoelectric stack actuator behavior, the hysteresis can be described by the functions of a maximum input voltage and the preload. The dynamic characteristics are also identified by the frequency domain modeling technique based on the experimental data. The hysteresis is compensated by the inverse hysteresis model for precise control of inchworm displacement. Since the dynamic stiffness of an inchworm is generally low compared to its driving condition, the mechanical vibration may degrade accuracy of the inchworm. Therefore, the SMC (Sliding Mode Control) and the Kalman filter are developed for the motion control of the inchworm. The feasibility of the proposed modeling scheme and the control algorithm is tested and verified experimentally.