【作者】 胡涛；

【导师】 郑淑涛；

【作者基本信息】 哈尔滨工业大学 ， 机械工程（专业学位）， 2025， 硕士

【摘要】 随着人机交互技术的快速发展,背部外骨骼机器人在康复训练与力量增强领域的应用日益广泛,尤其在辅助人体提升重物方面表现突出。背部外骨骼主要通过机械储能或主动助力方式,结合传动机构提供弯腰起身时的助力,从而有效减轻腰部肌肉负荷。适配人体脊柱结构与实现人体运动意图识别,是背部外骨骼机器人技术的核心挑战。本文针对上述关键问题,从人体脊柱运动特性出发,提出一种基于连续体机构的背部外骨骼结构,并开展系统性的建模与实验研究,主要研究内容包括以下三个方面:针对背部承载机构自由度受限的问题,本文深入分析人体脊柱各部分的差异化运动特性,提出一种基于连续体机构的背部外骨骼结构设计方案,建立完整的机构运动学与动力学模型,并进一步构建人机耦合模型,从理论上分析背部助力对减轻肌肉负荷的作用机制。实验结果表明,该连续体机构静态模型具备良好的准确性,且背部结构不会显著干扰人体自然运动,验证结构设计的合理性。针对传统运动识别方法精度低、分类模式单一的问题,本文重点研究基于表面肌电信号的运动意图识别方法,深入分析肌电信号特点,挖掘肌电信号与运动模式之间的潜在关联,强调基于肌电信号实现人机交互及多模式识别的必要性。在此基础上,提出一种基于卷积神经网络的关系网络识别框架,结合模板样本与输入样本的深度距离度量,实现运动意图的精准识别,并建立多模式搬运肌电数据集。实验结果显示,本文提出的关系模型在公开数据集与自建数据集上的识别准确率均显著优于对比方法,表现出良好的识别精度与泛化能力。为进一步验证背部外骨骼系统的综合性能,本文搭建软硬件协同的背部外骨骼实验平台,设计基于虚拟阻抗模型的双模式辅助控制框架。通过单一场景与多场景搬运任务实验,分析不同任务下肌电信号的变化规律,验证背部外骨骼系统助力的有效性及关系模型的在线识别准确性。实验结果表明,该背部外骨骼系统能够有效降低相关肌肉激活程度,显著缓解背部负荷。更多还原

【Abstract】 With the rapid development of human-robot interaction technology,back exoskele-ton robots have been increasingly applied in rehabilitation training and strength augmen-tation,particularly in assisting humans in lifting heavy objects.These exoskeletons pri-marily utilize mechanical energy storage or active actuation mechanisms combined with transmission systems to provide assistance during trunk flexion and extension,thereby re-ducing lumbar muscle load.The core challenges lie in adapting to human spinal structure and achieving accurate recognition of human motion intent.To address these key issues,this study proposes a back exoskeleton design based on a continuum mechanism inspired by spinal biomechanics and conducts systematic modeling and experimental research.The main research contents include the following three aspects:To address the limited degrees of freedom in conventional back support mechanisms,this study analyzes the differential motion characteristics of spinal segments and proposes a back exoskeleton structure based on a continuum mechanism.A complete kinematic and dynamic model of the mechanism is established,along with a human-exoskeleton coupling model to theoretically analyze the mechanism of muscle load reduction through back assistance.Experimental results demonstrate the accuracy of the static model and confirm that the back structure does not significantly interfere with natural human motion,validating the rationality of the design.To overcome the low accuracy and single classification modes of traditional motion recognition methods,this study focuses on motion intent recognition based on surface electromyography(EMG)signal.By analyzing EMG signal characteristics and exploring their latent correlations with motion patterns,the necessity of human-robot interaction and multi-mode recognition using EMG signals is emphasized.A relational network recogni-tion framework based on convolutional neural networks is proposed,which combines deep distance metrics between template samples and input samples to achieve precise motion intent recognition.A multi-mode load-handling EMG dataset is constructed.Experimen-tal results show that the proposed relational model significantly outperforms comparative methods in recognition accuracy on both public and custom datasets,demonstrating strong recognition precision and generalization capability.To further validate the comprehensive performance of the back exoskeleton system,a hardware-software integrated experimental platform is developed,incorporating a dual-mode assistive control framework based on a virtual impedance model.Through single-scenario and multi-scenario load-handling experiments,the variation patterns of EMG signals under different tasks are analyzed to verify the effectiveness of the exoskeleton as-sistance and the online recognition accuracy of the relational model.Experimental results confirm that the back exoskeleton system effectively reduces muscle activation levels and significantly alleviates spinal load.更多还原