毕业生知识库

近年来，由脑卒中和脊髓损伤等疾病造成的肢体瘫痪患者数量与日俱增。临床研究表明，运动康复能够在一定程度上帮助患者恢复肢体运动功能，重新获得日常生活能力。相对于传统的康复治疗方法，使用机器人辅助康复训练可减少康复训练中治疗师的人数和体力消耗，从而降低治疗成本；同时，可实现多模式个性化的康复训练，提高康复效果。本文在国家自然科学基金“共融机器人基础理论与关键技术研究”重大研究计划重点支持项目“人机共融的灵巧柔顺下肢康复机器人交互方法与应用”（91648208）、国家自然科学基金重点国际合作研究项目“康复机器人主动自适应控制策略与在线评价方法研究与应用”（61720106012）等项目的支持下，围绕下肢康复机器人的系统设计、人机交互控制以及康复训练策略展开研究，主要工作和创新点如下：
1. 针对不同的下肢康复训练需求，设计了足下垂康复踏车和全周期下肢康复机器人两款下肢康复训练样机。足下垂踏车融合了常规踏车训练与足下垂训练功能，足下垂机构是相对独立的模块，有效满足足下垂偏瘫患者的训练需求，弥补常规康复踏车的不足。全周期下肢康复机器人的下肢机构有三个主动关节，可为人体下肢提供单关节、多关节主被动训练，本文给出了机器人电控系统和上位机软件系统设计方案。
2. 提出了一种综合考虑关节耦合因素与粘滞力的摩擦力模型，分别建立了 下肢机构动力学模型和人体下肢动力学模型。在强约束和高度非线性的条件下， 采用粒子群算法和最小二乘估计对动力学参数进行精确辨识，从而获得精确的人机混合系统动力学模型，进而基于该模型估计出了康复训练过程中人体下肢主动力矩。
3. 针对康复训练过程中最优阻抗参数动态变化的问题，提出了一种基于阻抗参数模糊调节的自适应控制方法。将人机交互作用力、位置误差与速度误差作为变阻抗模糊调节系统的输入，并采用模糊推理实时调整阻尼系数与刚度系数，得到理想的阻抗控制参数，从而建立了一个自适应的主动柔顺人机交互训练环境。可避免康复训练过程中可能出现的下肢痉挛、抖动等异常的肌肉活动造成下肢与机构之间过度对抗而导致的二次伤害，确保了康复训练过程中的安全。并设计了间接自适应模糊控制器，对反映患者运动意图的设定轨迹进行稳定跟踪。
4. 提出了一种改进 Sage-Husa 自适应卡尔曼滤波方法，用于实时精确估计康复训练中的人机交互力。首先，建立人机混合系统的状态空间表达式。其次将交互力表达为时间的多项式函数，并作为系统状态引入到建立的状态空间方程，采用卡尔曼滤波方法对系统扩展状态进行在线估计，同时根据每一步估计结果实时调整交互力的动态函数阶次，实现交互力模型阶次的自适应。为了应对系统噪声协方差矩阵的不确定性和保持状态更新过程中测量噪声协方差矩阵的正定性，采用改进的 Sage-Husa 自适应卡尔曼滤波方法对系统噪声和测量噪声的协方差矩阵进行在线修正。在下肢康复机器人平台上的实验结果表明所提出的方法能够准确估计人机交互力的大小，且具有较好的实时性。
5. 提出了一种基于交互力精确估计和模糊变阻尼控制的下肢康复机器人速度自适应策略及相应的康复训练方法。首先将估计的人机交互力转化为关节末端沿参考轨迹前进方向的切向力。通过阻尼控制将此切向力转化为关节运动速度的调整量，动态调节训练速度。同时通过给定指向圆心的自适应法向速度，将运行轨迹逐渐拉向参考轨迹方向。并设计了一种基于模糊规则的阻尼参数调节器，可根据患者主动施加的交互力及偏离参考轨迹的范围来调节阻尼系数，从而实现根据患者肢体功能恢复情况调节训练强度。为了增加康复训练的趣味性，采用 Unity3D 软件设计了虚拟交互游戏，增强了患者和机器人之间的互动性和娱乐性。最后通过样机实验验证了所提方法的可行性。

In recent years, the number of limb paralysis patients caused by diseases such as stroke and spinal cord injury has increased. Clinical studies have shown that exercise rehabilitation can help patients recover limb motor function and regain their daily living ability. Compared with traditional rehabilitation methods, robot-assisted training can reduce the number of therapists and physical exertion in rehabilitation process, which lead to lower cost of treatment. At the same time, multi-mode personalized rehabilitation training can be realized to improve the rehabilitation effect. Supported by the National Natural Science Foundation of China (Grant 91648208) and the International Cooperation Program (Grant 61720106012), this paper focuses on the system design, human-robot interaction control and rehabilitation training strategies for a lower limb
rehabilitation robot. The main contributions and innovations of this research are as follows:
1. A foot drop rehabilitation treadmill and a full-cycle lower limb rehabilitation robot were developed for rehabilitation therapy of impaired lower limbs. The foot drop treadmill combines the functions of conventional treadmill and foot drop device. The foot drop mechanism is an independent module, which effectively meets the training needs of patients with foot drop and compensates for the deficiency of  conventional rehabilitation treadmill. The lower limb mechanism of the full-cycle lower limb rehabilitation robot has three active joints, which can provide single and multi-joint passive and active training for the lower limbs of the human body. The control architecture and software for the both robot system are designed.
2. Considering the joint coupling factors and viscous forces, a novel friction model is proposed. On this basis, the dynamic models of the lower limb mechanism and the human lower limb are established respectively. Under the condition of strong constraint and highly nonlinearity, the particle swarm optimization and least square method are used to identify the dynamic parameters, so as to obtain an accurate human-robot hybrid
system dynamic model. On this basis, the torque generated by the human lower limbs during the training process is estimated.
3. In order to obtain the optimal impedance parameters in the process of rehabilitation training, an adaptive control method based on fuzzy adjustment of impedance parameters is proposed. The human-robot interaction force, position and velocity errors are used as the input of the variable impedance fuzzy regulator. In order to establish an interactive interface with active compliance, the damping and stiffness coefficients are adjusted in real time by fuzzy reasoning to obtain an ideal impedance parameter. The proposed interaction control method can avoid the secondary injury caused by abnormal muscle activity such as spasm, which may lead to excessive confrontation between the lower limb and the mechanism, and ensure the safety during the rehabilitation training. An indirect adaptive fuzzy controller is designed to stably track the specific trajectory which reflects the patient’s motion intention.
4. An improved Sage-Husa adaptive Kalman filter method is proposed to accurately estimate the human-robot interaction force during the rehabilitation training in real time. Firstly, the state space equation of the human-robot hybrid system is established. Then, the interaction force is expressed as the polynomial function of time, and is introduced into the established state space equation. Moreover, the Kalman filter method is utilized to estimate the extended system state. At the same time, the dynamic function order of the interaction force is adjusted in real time according to the estimation result of each iteration step. In order to deal with the uncertainty of the system noise covariance matrix and maintain the positive definiteness of the measurement noise covariance matrix during the state update process, the improved Sage-Husa adaptive Kalman filter method is implemented to correct the covariance matrices of the system and measurement noise online. Finally, the experiment carried on the lower limb rehabilitation robot show that the proposed method can precisely estimate the human-robot interaction force and has good real-time performance.
5. A velocity adaptive strategy and corresponding training method based on accurate estimation of interaction force and fuzzy variable damping control are proposed for the lower limb rehabilitation robot. Firstly, the estimated human-robot interaction force is converted into a tangential force at the end of the robot along the forward direction of the reference trajectory. Then, the tangential force is translated into the adjustment
amount of the joint velocity by the damping control to dynamically regulate the training speed. At the same time, by giving the adaptive normal velocity, which points to the center of the circle, the robot will be gradually pulled back toward the direction of thereference trajectory. In order to adjust the training intensity according to the recovery of the limb function of the patient, a damping adjuster based on fuzzy rules is designed to regulate the damping coefficient in accordance with the interactive force and the deviation from the reference trajectory. In order to enhance the entertainment of rehabilitation training, the Unity3D software is adopted to design interactive virtual game. Finally, the feasibility of the proposed method is verified by prototype experiments.