Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo

Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo Tóm tắt nghiên cứu điều khiển robot phục hồi chức năng chi dưới sử dụng cơ nhân tạo

MINISTRY OF EDUCATION AND TRAINING HANOI UNIVERSITY OF SCIENCE AND TECHNOLOGY Dinh Van Vuong STUDY ON CONTROL OF A PAM-BASED LOWER- LIMB REHABILITATION ROBOT Majors: Control Engineering And Automation Code: 9520216 SUMMARY OF DOCTOR DISSERTATION ON CONTROL ENGINEERING AND AUTOMATION Ha Noi - 2024 The Dissertation was completed at: Hanoi University Of Science And Technology SUPERVISOR Assoc.Prof.Dr Duong Minh Duc Dr Dao Quy Thinh Reviewer 1: Reviewer 2: Reviewer 3: The Dissertation was defended before the Doctoral Dissertation Evaluation Council of Hanoi University of Science and Technology meeting at Hanoi University of Science and Technology At …… hour, day … month … year ……… The Dissertation can be found at the library: Ta Quang Buu Library - Hanoi University of Science and Technology Vietnam National Library Abstract The necessity of the Dissertation Nowadays, rehabilitation robot systems are being researched and developed worldwide to replace physical therapists gradually Robots can assist patients systematically in performing preprogrammed rehabilitation exercises At the same time, robots can assist with long- term training without getting tired However, as robots interact directly with humans, safety is always a top priority in designing and controlling rehabilitation robots In addition, the robot's actuators must also be flexibly controlled to make patients feel most comfortable during training and avoid causing injury to the patient Recently, the rehabilitation system using pneumatic artificial muscles has attracted much attention from researchers due to the similarity between PAM and human muscles The PAM is lightweight and has a higher power-to- weight ratio than motorized transmission devices Moreover, PAM is quite flexible and suitable for robots that interact with humans, such as rehabilitation robots Several prototype systems of rehabilitation robots have been developed at research centres worldwide But, most of the above systems are still in the early stages of development In summary, all rehabilitation robot systems that use pneumatic artificial muscles domestically and internationally are only currently in the laboratory and have not been commercialized So, the potential for research and development is enormous Based on this reality, we have chosen the topic of "Study on Control of a PAM-based Lower-limb Rehabilitation Robot" Purpose of the research The purpose of this Dissertation focuses on control lower-limb robotic orthosis: Objectives and scope of research a Research objective The research objective is the BK-Gait 2-DOF robotic orthosis that covers hip and knee joints The robotic orthosis is actuated by pneumatic artificial muscles in an antagonistic configuration b Scope of research The research scope of this Dissertation focuses only on the study of controlled lower-limb robotic orthosis Therefore, the research project will be performed based on theoretical foundations and experiments: • The model parameters are collected based on the pneumatic artificial muscle (PAM) with the antagonistic configuration of the BK-Gait 2-DOF robotic orthosis • All measurements, control algorithms, and experimental results are performed and verified by experiments on the lower-limb rehabilitation robot model (BK-Gait) at Hanoi University of Science and Technology Research method The research method is a combination of theoretical research and experimental verification • The robot will be designed and simulated using software to ensure durability It will then be fabricated for testing purposes • A mathematical model with two muscles having an antagonistic configuration will be built to describe the system's dynamic characteristics The model parameters will be identified through the system's input/output data using an optimization algorithm implemented in Matlab/Simulink • Control algorithms will be applied to build trajectory-tracking and impedance controllers for all actuators and robots Then, it will be programmed on suitable controllers such as National Instrument's MyRio The control performance will be verified through experimental results The proposed controllers will be performed through experiments with the fabricated robot The scientific and practical significance of the Dissertation a Scientific significance The scientific significance of this Dissertation is to build trajectory- tracking control and impedance control algorithms for actuators and robot systems using PAM-based actuators that are accurate and suitable for rehabilitation applications b Practical significance The practical significance of this Dissertation is to build a rehabilitation system for the human lower limb with trajectory tracking and impedance control functions with good precision and applicability to rehabilitation systems in practice Structure of the Dissertation The Dissertation is structured into Chapters and a conclusion as follows: • Chapter Overview of the Rehabilitation Systems: This Chapter provides an overview of the rehabilitation robot system It highlights that the global research and development efforts in rehabilitation robots are substantial due to their notable advantages over traditional rehabilitation methods However, most research on robotic rehabilitation systems using artificial muscles is limited to the laboratory and has not been commercialized This indicates a significant potential for further research and development in the field • Chapter Modeling and Control of PAMs: This Chapter overviews pneumatic artificial muscles and the popular methods for modeling artificial muscles After that, we built a mathematical model for a PAM-based actuator Finally, we will apply advanced control algorithms to build a trajectory-tracking controller for a PAM-based actuator Multiple experiment scenarios will be performed to verify the effectiveness of these controllers • Chapter Trajectory Tracking Control of the BK-Gait Orthosis: This Chapter focuses on improving the control system for the BK-Gait lower limb robotic orthosis Firstly, we will build a mathematical model for the BK-Gait lower limb robotic orthosis Next, we will apply advanced control algorithms to build a trajectory-tracking controller for BK-Gait lower limb robotic orthosis Finally, multiple experiment scenarios will be performed to verify the effectiveness of the built controller • Chapter Impedance Control of the BK-Gait Orthosis: In this Chapter, a neural network-based method is chosen to estimate the patient's recovery, an essential factor for a gait training robot system powered by pneumatic artificial muscles The estimated patient recovery will be used as the control signal for the impedance controller to improve joint compliance coefficients • Conclusions and Recommendations: This section summarizes the results achieved in the thesis, the main contributions, and proposes future research directions The Contributions of the Dissertation This study mainly presents the control algorithms for a low-limb rehabilitation system by combining theoretical research and experimental verification The main contributions of the Dissertation: • Building a two-degree-of-freedom pneumatic artificial muscle- based exoskeleton robot for the human's lower-limb rehabilitation • Developing trajectory tracking control function for a prototype robot by employing some advanced control strategies • Integrating an impedance control function into a robot by using neural networks to approximate the interaction force of humans to the robot Chapter Overview of the Rehabilitation Systems 1.1 Motorized Lower-Limb Rehabilitation Orthosis Systems The Driven Gait Orthosis (DGO), also known as LOKOMAT (Hocoma AG, Volketswill Switzerland), is currently available in the market and is extensively researched in many rehabilitation centres as one of the best examples for gait orthosis that can be used for lower-limb disabilities This orthosis system is shown in Figure 1.1a It consists of three main parts: body weight support, treadmill, and powered leg orthosis Considerable control algorithms have been implemented into this system to improve its performance, such as position, adaptability, impedance controllers, etc Figure 1.1b shows the treadmill gait trainer system, which incorporated the electromechanical gait device with the treadmill/gait training, known as the LokoHelp (LokoHelp Group, Germany) The LokoHelp used a different mechanical system than the LOKOMAT, which implemented the powered leg orthosis The foot- powered orthosis, "Pedago'', used an electromechanical gait device to provide a gait motion during training sessions The control device helps move the patients' foot trajectory with a fixed step length of 400 mm, in which the gait cycle (GC) speed can be varied from to km/h The ReoAmbulator robotic system (Motorika Ltd, USA), which is also known as "AutoAmbulato'', is another example of existing treadmill gait trainers for lower-limb rehabilitation therapy, as shown in Figure 1.1c This system has been used in research centers and medical hospitals for rehabilitation therapies and educational research studies Figure 1.1 (a) LOKOMAT, (b) LokoHelp, (c) ReoAmbulator The evaluated motorized lower-limb gait rehabilitation orthosis systems mentioned above represent only a fraction of the existing rehabilitation orthoses However, it could be summarized from these examples that its development has advanced, whereby many lower-limb gait rehabilitation orthoses based on electrical motors have already been commercialized With its growth speed in mechanical design and the implementation of advanced control schemes and strategies, the space available for enhancements might closely reach its peak 1.2 Pneumatic Muscle Actuated Lower-limb Rehabilitation Orthosis Systems Figure 1.4 (a) The hip orthosis, (b) The ankle orthosis, (c) The ankle orthosis AFO Recently, a natural and low-cost actuator PAM has been widely implemented in developing rehabilitation systems Compared with conventional actuators such as electrical motors, series elastic actuators (SEA), and brushless DC motors, PAM has many advantages, including being naturally compliant, lightweight, and having a high weight ratio to power Despite inherent drawbacks such as very high nonlinear and uncertain characteristics and slow response in force generation, the applications of PAM in robotic rehabilitation fields are exponentially growing due to the demand on much high compliant human-robotic systems The first robotic orthosis actuated by PAM was developed by Claysson B Vimieiro et al in 2004 for supporting the hip flexion movement of patients As shown in Figure 1.4, this exoskeleton is designed with two main parts: the first one is a pelvic brace to provide the stability of the robot, and the second one is support for the thigh The clinical results showed that the exoskeleton was able to provide not only more stabilization but also a better condition for the patients during walking activity In summary, it can be seen that the rehabilitation robot system is being heavily researched and developed worldwide because of its outstanding advantages compared to traditional rehabilitation methods Robots can assist patients systematically in performing rehabilitation exercises that have been preprogrammed Several prototype systems of rehabilitation robots have been developed at research centers worldwide However, most of the above systems are still in the early stages of development The "assist-as-needed'' (AAN) function is indispensable for a rehabilitation robot to restore patient function Therefore, a rehabilitation robot must have sufficient rigidity to guide the patient's limb along the designated trajectory and estimate the patient's level of disability to provide the necessary support 1.3 The BK-Gait lower-limb rehabilitation system Figure 1.9 The BK-Gait lower-limb rehabilitation system Figure 1.9 demonstrate the schematic diagram of the BK-Gait rehabilitation system The overall rehabilitation system includes the following parts: • Part is the Body Weight Support • Part is a treadmill (walking assist device) • Part is the development of sample physical therapy exercises • Part is the Robot Orthosis The research scope of this Dissertation only focuses on the study of control lower-limb robotic orthosis Therefore, the research project will be performed based on theoretical foundations and experiments However, there are two control problems for lower limb rehabilitation robots: trajectory tracking and impedance control problems The aim of the Dissertation is to apply some advanced control algorithms to build controllers to solve the above two control problems 1.4 The experimental system The experimental models built within the research scope of this Dissertation all use pneumatic artificial muscles (PAM) as actuators There are many types of artificial muscles, as shown in Figure 2.1 of Section 2.1.1 However, this study will use the McKibben muscle due to lightweight, easy to fabricate with low cost, high power/weight ratio, and similar to the behavior of human muscle Although the McKibben mechanism is also commercialized on the market But it is very expensive Therefore, in this study, the McKibben muscles will be made manually with available materials and low cost 1.4.1 The experimental model for a PAM-based actuator Figure 1.16 The experimental model for a PAM-based actuator The experimental platform consists of two self-made PAMs, as shown in Figure 1.16, each 25 mm in diameter and 400 mm in length, arranged in an antagonistic configuration Two proportional electric control valves regulate the internal pressure of the PAMs The deflection angle of the pulley is measured using an angle sensor (WDD35D5T) with a precision of 1% The control platform comprises an embedded controller (National Instrument myRIO-1900) that can be monitored and interact with the field devices using LabVIEW software This experimental model will verify the control performance of the algorithms implemented in Chapter 1.4.2 The experimental model for the BK-Gait lower-limb robotic orthosis This research considers a BK-Gait-based lower limb rehabilitation system for the experimental works The system's main advantage is the suspension frame's direct attachment to the pre-shaped aluminum, which uk =  n m   (2.32) b1  yk +  yk−i − bjuk− j − pˆk + ek−1 − sk−1 + KSW sign(sk ) i =1 j=0  2.2.2 Adaptive Radial-Basis Function Neural Network Control of a Pneumatic Actuator This section offers an online adaptive controller based on RBF neural approximation to improve control precision and adapt to parameter variations Figure 2.18 depicts the structure of the proposed closed-loop neural-based control system Figure 2.18 RBF Neural network-based control diagram The control signal is: u = −g(x) + yd + MT E (2.52)   Figure 2.19 demonstrates the typical three-layer diagram of an RBF neural network Figure 2.19 Diagram of an RBF neural network From the previous part, the unknown nonlinear function 𝑔𝑥 the RBF neural network has represented The control signal is: u = −gˆ(x) + yd + MT E (2.56)   12 where 𝑔̂𝑥 is the estimated parameter for 𝑔𝑥: 2.2.3 Prescribed Performance Function Based Sliding Mode Control of Opposing Pneumatic Artificial Muscles to Enhance Safety The proposed control approach combines a prescribed performance function and a discrete-time sliding mode control to enhance the system's control performance Figure 2.23 shows the schematic diagram of the proposed control approach Figure 2.23 Block schematic for the recommended performance sliding mode control This prescribed performance function ensures that the set orbital tracking error will always be limited to a given domain k+1 = (1−  ) k +  2 (2.70) Defining 𝑒𝑘 is the tracking error of the measured value 𝑦𝑘 from its desired one 𝑦𝑘∗ From PPF (2.70), we have the expression for the convergence domain of the error as follows: −k  ek  k (2.71) The control signal 𝑢𝑘 as follows: uk =  yk+1  + a1 yk + a2 yk−1 − b2uk−1 − pˆk − pk  (2.86) b1 k+1   +   +   (k) −  b1  + e  Assuming that the estimation error of the disturbance 𝑝̃𝑘 is negligible, the control signal can be obtained as follows: 13 uk =  yk+1  + a1 yk + a2 yk−1 − b2uk−1 − pˆk  (2.87) b1 k+1   +   +   (k) −  b1  1+ e  2.2.4 Experimental results In this part, the proposed controller (PPF-ERL-SMC) is also compared to its counterpart, an Adaptive Sliding Mode Controller (ASMC) and an adaptive controller based on a radial basis function neural network (RBF) Figure 2.25 Experiment results Figure 2.26 Experiment results without a load with a load The experiments conducted in the first scenario were carried out without a load The purpose of this scenario was to evaluate the tracking performance of the proposed controller for sinusoidal signals with a frequency of 0.5 Hz The results are presented in Figure 2.25 Based on the result, it can be inferred that the PPF-ERL-based controller proposed in this research executes better control performance in the transient process during system startup The proposed controller demonstrates superior control quality when handling interference effects that can cause abnormalities in rehabilitation In the second scenario, the system is tested by suddenly adding a load of kg after it reaches the steady state The desired trajectories for the test are the same as the first scenarios The experimental results are shown in Figure 2.26 Based on the results obtained from the two experimental scenarios, it can be inferred that the PPF-ERL-based controller proposed 14 in this research executes better control performance in the transient process during system startup and when a load is applied The proposed controller demonstrates superior control quality when handling interference effects that can cause abnormality in rehabilitation The RMSE and MTE values of the three controllers are shown in Figure 2.27 Figure 2.27 The quantitative evaluation of three controllers when tracking 0.5 Hz sinusoidal signal in case of without load and load Chapter Trajectory Tracking Control of the BK-Gait Orthosis 3.1 The mathematical model of the BK-Gait Orthosis According to the Euler-Lagrange equation, the robot dynamics with two rotating joints as follows: T = M( ) + H(, ) + G( ) (3.1) From equation (2.17) and equation (3.1) we have:  = M-1 (−H − G) + M-1A1ΔP (3.3) By adding 𝜔(𝑡), which is an unknown disturbance existing in any system, the state-space model of the dynamic system (3.3) as follows: x1 (t) =  (t)  x1 (t ) = x2 (t ) (3.4) x2 (t ) = f (x1 (t ),x2 (t ),ω(t )) +  (ω(t )) + u(t ) 15 3.2 Nonlinear ESO-based ADRC Controller This section implements nonlinear ADRC for trajectory tracking controller in a 2-DOF PAMs-based rehabilitation exoskeleton Due to using PAMs for actuators, the robot is strongly nonlinear with time- variant parameters and unknown disturbance So, to effectively control the robot, a nonlinear extended state observer (ESO) is established Then, a nonlinear feedback controller is designed to handle the system following the desired trajectory In addition, a tracking differentiator (TD) is also proposed to create a physically feasible reference trajectory Figure 3.2 demonstrates the model of the ADRC controller Figure 3.2 ADRC controller diagram 3.2.1 Tracking Differentiator (TD) The conventional tracking differentiator y(t) of the signal r(t) is: y (t )  r (t ) − r (t − Ts ) (3.5) Ts Considering the estimation error of the tracking differentiator as follows:  e1   e1 (t )  2    = A  e2 (t )  + Br (t ) (3.12)  e2     3.2.2 Extended State Observer (ESO) The ESO is an improved development based on the fundamental state observer The ESO can estimate both the state variables and the system disturbances The unmodeled system dynamics, unknown control coefficients, external noise, etc., can cause system disturbance The state 16 variables of the proposed PAM-based robot:   xˆ1 (t ) = xˆ (t ) + 3   y (t ) − xˆ1 (t )        y (t ) − xˆ1 (t )  2 −1  (3.20) xˆ (t ) = xˆ (t ) + 3  + u(t )      y (t ) − xˆ1 (t ) 3−2 xˆ (t ) =       3.2.3 Feedback Controller The last important part of the ADRC approach is the feedback controller, which developed as the following equation: u(t) = (3.27) −(k +1) xˆ1 (t ) − z1 (t ) − (k +1) xˆ2 (t ) − z2 (t ) − xˆ3 (t ) b0 3.3 Discrete-Time Backstepping Sliding Mode Control for a 2-DOF PAM-Based Exoskeleton Figure 3.8 Block diagram of the proposed control strategy This section introduces the proposed BSMC technique Figure 3.8 depicts the control block diagram of the BSMC approach The backstepping control method decomposes the second-order system model into smaller subsystems At each stage, the virtual control law 𝑦1(𝑘) and 𝑦2(𝑘) for the corresponding subsystems are developed using the discrete-time Lyapunov stability theorem In step 3, the sliding- mode control approach guarantees that the system state trajectory 17 reaches the sliding surface and that the system disturbance current tracking error reduces to zero • Step 1: Aim to establish a tracking error vector that measures the difference between the controlled rotation angle y(k) and the reference signal y*(k) We have: V1 (k ) = Ts y1 (k ) − Ts y1 (k − e2 (k ) = Ts2e12 (k ) − e2 (k ) (3.34) ) • Step 2: To guarantee the convergence of the vector e1(k) to zero, we can choose the second Lyapunov function as: V2 (k ) = Ts y2 (k ) − Ts y2 (k − (1 − Ts2 ) e12 (k ) − e2 (k ) (3.39) ) = Tse2 (k )2 − (1− Ts2 )e12 (k ) − e2 (k ) By examining equation (3.39), it becomes evident that ∆𝑉2(𝑘) will become negative definite if 𝑒2(𝑘) equals Therefore, the next stage is determining the vector of 𝑒2(𝑘) that leads to convergence towards zero • Step 3: At this stage, a sliding-mode control approach is applied after completing the two steps in the backstepping design process: Equation (3.46) enables the selection of a set of numbers ∝, β and γ that ensures the stability of the Lyapunov function Therefore, the proposed backstepping sliding mode control guarantees the system's stability   2 V3 (k )  −s2 (k −1) − e(k ) + e2 (k )    e2 (k )  −(1− Ts2 ) e1 (k ) +  2 (3.46) 2(1− Ts )     2  2  22 −  − − − Ts  e2 (k )  4(1− Ts )  3.4 Experimental results Both control strategies demonstrate effective tracking performance in the first scenario without a load However, the BSMC controller 18