中国科学院自动化研究所机构知识库

       微纳作业平台作为实现极端小尺度运动定位的通用平台，在纳米研究、生物医药、精密设备、先进制造等领域发挥着重要作用。本文从致动器位移融合测量、位移跟踪控制和载荷二维姿态控制等方面展开微纳作业平台位姿感知与智能控制方法研究，论文的主要研究成果包括：

       一、对压电致动器位移的感知展开研究。针对传统纳米级位移传感器与微纳作业平台兼容不佳的问题，提出了一种基于自感知方法的位移融合测量方法。首先，提出了一种带漏电漂移补偿的自感知位移测量方法，并引入面向应变片测量的时间-数字转换方法作为补充，设计了基于二者的测量电路，实现了低功耗、低空间占用、低装配精度要求的位移测量。其次，提出了一种基于卡尔曼滤波的异频异步数据融合算法将自感知与时间-数字转换方法进行融合估计。在此过程中，基于广义误差对卡尔曼滤波进行推导，得到一种符合所提方法误差特征的参数设置方案，并在此基础上设计了自适应参数调节机制以提高动态目标位移估计的准确性。最后，通过位移测量实验验证了所提融合测量方法的有效性。

       二、针对压电致动器位移跟踪控制问题展开研究，提出了一种神经网络模型预测控制方法。首先，针对压电致动器非线性特性和模型预测的计算负担问题，提出了一种并行输出整个预测序列的神经网络预测模型。其次，基于该模型设计了模型预测控制器，并针对优化初值问题设计了神经网络前馈控制器。再次，基于模型预测和扩展卡尔曼滤波提出了一种处理反馈延迟的位移估计方法，以提高对动态轨迹的跟踪性能。最后，通过位移轨迹跟踪实验验证了所提控制方法的有效性。

       三、针对载荷多自由度姿态控制问题，提出了一种数据驱动的多自由度姿态角位移控制方法。首先，设计了一种以线性位移致动的模块化二维姿态角微调节平台。其次，提出了一种基于神经网络逆运动学模型和直接逆控制的二自由度姿态控制器。再次，针对低速率的全局姿态反馈，为控制系统的反馈环节设计了一种姿态融合估计方法及相应的反馈控制器。最后，通过二维姿态角位移控制实验验证了所提方法在机构存在一定装配误差的情况下仍具有较好的控制性能。

    Micro/nano manipulation platform is a general platform with the ability to perform motions in extremely small scale, playing an important role in the field of nano research, biomedicine, precise equipments and advanced manufacture. This thesis researches on the sensing and intelligent control of position and attitude for micro/nano manipulation platform, focusing on fused measurement and tracking control of displacement of nano actuator, also on the multidimensional attitude control of the load. The technical contributions of this thesis are as follows:

    Firstly, the nano-scale displacement sensing method for piezoelectric actuator is researched. As traditional nano-scale displacement sensors are poorly compatible with actual micro/nano manipulation platform, a fused displacement measuring method based on self-sensing is proposed. To achieve measurement with low-dissipation, low space occupation and low requirement for assembly, a self-sensing method with compensation to the leakage drift is proposed, and a time-digit-conversion (TDC) method for strain-gauge measurement is introduced as supplement, a measuring circuit is designed to fulfill the two methods.Then a Kalman-filter-based fusion is proposed to fuse self-sensing and TDC to give estimation of displacement. For this purpose, the Kalman filter is derived based on generalized errors to make the settings of parameters fit the characteristics of proposed measurements, based on which a self-adaptive adjustment mechanism for parameters is developed to improve accuracy of estimation for dynamic objectives. The effectiveness of proposed fused measuring method is verified by displacement measuring experiments.

    Secondly, a neural-network-based model predictive control method is proposed for nano-scale displacement tracking. To relieve the problem of nonlinear characteristics of piezoelectric actuator and the computational burden of model prediction, a neural network model is proposed with the ability to parallel output the entire series of displacement predictions. Then a model predictive controller is designed based on the proposed model, and a neural feedforward controller is designed to solve initial value problem for optimization. Furthermore, to relieve the effect of feedback delay, a displacement estimation method based on model prediction and extended Kalman filter is also proposed to improve tracking performance for dynamic trajectories. The effectiveness of proposed controller is verified by displacement tracking experiments.

    Lastly, aiming at attitude control problem of load, a data-driving multidimensional angular shift control method is proposed. A modularized 2-degree-of-freedom (2-DOF) attitude adjuster actuated by linear displacement with minimal structure is designed. A 2-DOF attitude controller is proposed based on neural network inverse kinematic model and direct inverse control. The control system is completed with a feedback controller based on fused estimation to relieve the problem of low-rate global feedback of attitude. Through 2-DOF angular control experiments, the proposed controller is verified of its effectiveness in situation where assembly errors of mechanism are non-ignorable.