本研究針對可與折疊式輪椅結合的無外力感測器之動力輔助模組設計外力觀測器，這套外力觀測器將以卡爾曼濾波器(KF)估測使用者施加之外力，但KF需有準確的系統參數，才能準確地偵測使用者推力，所以使用最小平方法(LS)進行系統參數的辨識，但考慮到系統中存在隨環境改變或非時變的參數，故以遞迴最小平方法(RLS)進行參數辨識及更新，並通過模擬制定參數及外力的觀測與更新策略，最後以實驗驗證所提出之策略的可行性。
本研究從模擬結果中得知在以RLS估測參數時，所估測的參數間會有互相耦合之情形，因此提出於系統自由運行(free-run)時的減速情況下進行參數估測，以將參數間之耦合進行解耦，但在free-run時將有參數無法進行即時估測與更新，因此提出於系統啟動時，先以LS進行系統初始化估測系統參數之初始值，並假設使用者在使用輪椅時，其重量不會產生變動，所以其慣量亦不會改變，因此可將慣量與RLS的估測參數代入參數之關係式中，以求得相應參數，之後將KF與RLS結合，確定該策略可估得外力。最後於實驗過程中，將模擬時所提出之策略改為:於動力輪提供輔助力時，再估測free-run時無法估得之參數，而實驗結果顯示KF能估測到外力之趨勢。

In this study, an external force observer is designed for a power assist module without an external force sensor that can be combined with a folding wheelchair. In addition, the force observer will use Kalman Filter (KF) to estimate the external force which from the user. However, the Kalman filter needs accurate system parameters to accurately estimate the user's thrust. So, using the least squares method (Least Squares, LS) to identifying parameters for system. However, considering that there are parameters that change with the environment or time-invariant in the system, the Recursive Least Squares (RLS) method is used to identify and update the parameters. And through simulation to formulate update strategies for parameters and external force estimation.
This study knows from the simulation results that when the parameters are estimated by RLS, the estimated parameters will be coupled with each other. Therefore, it is proposed to perform parameter estimation under deceleration when the system is free-run for decoupling. However, there will be parameters that cannot be estimated and updated in real time during free-run. Therefore, it is proposed to use LS to initialize the system to estimate the initial value of the system parameters when the system is started. It is also assumed that when the user uses the wheelchair, its weight will not change, so its inertia will not change. Therefore, the estimated parameters of inertia and RLS can be substituted into the relational expression of the parameters to obtain the corresponding parameters. After that, KF and RLS are combined to determine that the strategy can be estimated. Finally, during the experiment, the strategy proposed in the simulation was changed to: when the power wheel provides auxiliary force, the parameters that cannot be estimated during free-run are then estimated. The experimental results show that KF can estimate the trend of external forces.

[1]MAX Mobility LLC, SmartDrive MX2 Product,
Available at:http://www.max-mobility.com/smartdrive#videos, Accessed 21 May 2020.
[2]Alber GmbH, e-motion Product,Availableat:https://www.alber.de/en/products/active-drives/e-motion/, Accessed 21 May 2020.
[3]葉治緯，2018，雙動力輪智慧電動輪椅之系統整合與控制，碩士論
文，國立臺灣大學，機械工程研究所，台北。
[4]C. C. Ou and T. C. Chen, 2012, “Power-assisted wheelchair design based on a lyapunov torque observer, ” International Journal of Innovative Computing, Information and Control, vol. 8, no. 12, pp. 8089-8102.
[5] D. Ding, R. A. Cooper, 2005, “Electric-powered wheelchairs: A review of current technology and insight into future direction,” IEEE Control Systems Magazine, Vol. 25, No2, pp.23-34.
[6]D. Y. Kim, Y. H. Kim, K. S. Kim, and J. M. Kim, 2016,“A Conttrol Method for Power-Assist Devices using a BLDC Motor for Manual Wheelchairs,”Journal of Power Electronics, Vol. 16, No. 2, pp. 798-804.
[7]林逸修，2013，鋰離子電池之多項式卡爾曼濾波器電量估測，碩士論文，元智大學，電機工程學研究所，桃園。
[8] Ming Yang , Zirui Liu, Jiang Long, Wanying Qu and Dianguo Xu, 2018, “An Algorithm for Online Inertia Identification and Load Torque Observation via Adaptive Kalman Observer-Recursive Least Squares,” Harbin Institute of Technology.
[9]江奇融，2007，卡爾曼綠波法於建築結構系統之參數識別，碩士論文，中原大學，土木工程學研究所，桃園
[10]V. van Geffen, 2009, A study of friction models and friction compensation, Technische Universiteit Eindhoven, Eindhoven.
[11]T. T. Phuong, K. Ohishi, C. Mitsantisuk, Y. Yokokura K. Ohnishi, R. Oboe, A. Sabanovic, 2018,“Disturbance Observer and Kalman Filter Based Motion Contral Realization”Journal of Industry Applications, Vol 7, No. 1, pp. 1-14.
[12]Karl Johan Åström Björn Wittenmark, 1995“Adaptive Control”2nd Ed., Addison Wesley, Boston.
[13]張喬崴，2017，應用第回最小平方識別法進行變勁度系統振動台與即時複合試驗之即時參數識別，碩士論文，國立成功大學，土木工程系研究所，台南。
[14]成功大學馬達科技研究中心，Available at:
http://km.emotors.ncku.edu.tw/emotor/worklog/EMTRC/motor_learn/elearning/chen/1.pdf, Accessed 21 May 2020.
[15]鄒應嶼，1996，直流馬達的模型化與參數辨別，國立交通大學，控制工程系所，Available at: http://pemclab.cn.nctu.edu.tw/peclub/W3cnotes/cn09/，Accessed 21 May 2020.