| 摘要: | 提升上肢運動缺陷的康復效果是康復醫療的一個重要目標，特別是針對嚴重
殘疾患者的康復治療。近期研究顯示，利用上肢外骨骼機器人進行重複性訓練能顯著提升中風患者的運動控制能力，並改善康復結果。與傳統使用電動馬達或液壓缸作為執行器的外骨骼系統相比，氣動肌肉執行器（Pneumatic Muscle Actuator, PMA）因其輕量化、低功耗、高柔性與適應性強等特性，成為更適合於醫療輔助設備的選擇。然而，PMA 在充氣與放氣階段表現出非線性遲滯現象，受摩擦、材料彈性及內部氣壓動態影響，使得 PMA 難以達到精確的控制效果。本研究開發了一種由腦波控制的氣動外骨骼系統，以實現癱瘓患者的主動康復訓練。外骨骼系統利用腦波信號驅動，透過腦-機介面（Brain-Computer Interface, BCI）系統解讀
患者的運動意圖。這些意圖通過模糊神經網絡（Fuzzy Neural Networks, FNN）轉換為控制信號，用以驅動 PMA 進行外骨骼的抬升和下降。為應對 PMA 在充氣與放氣階段的非線性特性，本研究提出雙路模糊神經網絡（Dual-Path FNN, DFNN），分別優化外骨骼的抬升與下降的控制策略，並進行多負載條件下的性能測試。在實驗中，我們比較了 DFNN 與單路模糊神經網絡（Single-Path FNN, SFNN）的控制性能。結果顯示，DFNN 在不同負載（無負載、1 公斤、2 公斤）條件下均表現出更低的誤差、更穩定的反應速度及更好的控制精度。例如，在 1公斤負載下，DFNN 的平均角度誤差為 1.23°，而 SFNN 則為 3.26°。此外，我們
進一步測試了外骨骼系統在腦波控制模式下的實用性，三位受試者通過腦波意圖成功驅動外骨骼完成手臂的抬升與放下動作。本研究表明，外骨骼系統在腦波解讀與運動執行方面均展現出穩定性與準確性，為未來基於腦波控制的外骨骼系統研究奠定了基礎。;Improving rehabilitation outcomes for upper limb motor deficits is a critical goal in
medical rehabilitation, especially for patients with severe disabilities. Recent studies have
demonstrated that repetitive training using upper limb exoskeleton robots significantly
enhances motor control and rehabilitation outcomes in stroke patients. Compared to
traditional exoskeleton systems that use electric motors or hydraulic actuators, pneumatic
muscle actuators (PMAs) stand out as a more suitable option for medical assistive devices
due to their lightweight design, low power consumption, high flexibility, and adaptability.
However, the inherent nonlinear hysteresis behavior of PMAs, caused by friction, material
elasticity, and internal air pressure dynamics during inflation and deflation, poses challenges
for precise control.
This study developed a brainwave-controlled pneumatic exoskeleton system aimed at
enabling active rehabilitation training for paralyzed patients. The system employs brainwave
signals as control inputs, leveraging a brain-computer interface (BCI) to interpret patients’
motor intentions. These intentions are translated into control signals via fuzzy neural
networks (FNNs) to drive the PMA for lifting and lowering the exoskeleton. To address the
nonlinear characteristics of PMAs during inflation and deflation, this study proposed a dual path fuzzy neural network (DFNN) architecture, which optimizes the control strategies
separately for lifting and lowering motions. The system was evaluated under various load
conditions to assess its performance.
In the experiments, the performance of the DFNN was compared with a single-path
fuzzy neural network (SFNN). Results showed that the DFNN exhibited lower errors, faster
response times, and higher control precision across different load conditions (unloaded, 1 kg,
and 2 kg). For example, under a 1 kg load, the average angular error of the DFNN was 1.23°,
compared to 3.26° for the SFNN. Furthermore, the practical usability of the exoskeleton
iii
system under brainwave control was tested. Three participants successfully drove the
exoskeleton to perform arm-lifting and lowering tasks using motor imagery-based brainwave
intentions.
This study demonstrates that the exoskeleton system provides stability and accuracy in
both brainwave decoding and motion execution. The results highlight the potential of using a
brainwave-controlled pneumatic exoskeleton for rehabilitation and lay the foundation for
future advancements in brainwave-driven exoskeleton systems. |
