HOME
LINKEDIN
Sensors Network Architecture for Intelligent Robotic Gait Rehabilitation
Level: Final Year Project
Grade: 77%
Skills Devevloped: MATLAB, CAD (Fusion360), Prototyping (3D Printing), Circuit Design, Technical Drawings, Leading Experiments, Presentation, Primary and Secondary Research, Report Writing.
3rd Year Mechanical Engineering
2023
Stroke appears to be a common cerebrovascular disease that impact 1 out of every 6 people on earth with high morbidity, mortality, and disability rates. This individual design project focused on designing and building a low-cost, wearable quaternion-based IMU sensor network to operate alongside robotic rehabilitation devices for post-stroke patients to assess the kinetic and kinematic parameters of their gait cycles. Ultimately, this project achieved capturing joint angle accuraries of 93%, while cutting the costs by 94% compared to commercially used Xsens IMU sensors.
Or
Scroll for Project Overview
Scroll for Project Overview
BRIEF OVERVIEW
Robotic Rehabilitation Device (Problem)
The robotic-assisted-end-effector gait rehabilitation device guides a normal gait, with pressure sensors encompassed in the footplate to examine the spatiotemporal parameters and kinetic variables (such as force, momentum) which are important for assessing the strength and coordination of muscles and joints, identifying the asymmetries in muscle functions.
However, kinematics must also be examined to identify abnormal joint mobility in comparison to a healthy person for corrective feedback for the robotic assistive device.
Hence, I went on a journey to develop a set of low-cost IMU sensor network to quantify the kinematic parameters for knee-joint angles, and justified its accuracies with different exercises in comparison to commercially joint measuring techniques.
Design for Wearability
Assembled to Ensure Stability
Euler-Angle Validation
Two attachment methods (AM) were tested: AM1 has the IMU’s X-axis directed towards the ground; while AM2 has their Y-axis directed to the ground during the neutral standing position.
Then, the same exercises of deep squats and walking were performed by the subject for the more accurate attachment method to be identified from the comparisons of the results.
Technical Drawings for Casings of the Selected IMU Casing Design (Attachment Method 2)
Data Collection (Mathematical Model)
MATLAB was used to receive the quaternion orientation data from the 2 IMU sensors, and calculations were performed with equations below to identify the joint angles in between the IMU sensors.
To translate the quaternions from the two IMU sensors into joint angles, three-dimensional rotation matrices had to be computed using the equation below:
Where R(q) is the rotation matrix of the quaternions from the global reference frame, q0 is the scalar demonstrating the rotational angle; q1, q2, and q3 portray the axis of rotation around where 𝑞0 is performed.
The rotation of the knee joint angles could be presented by locating the relative rotation from Sensor 1 to Sensor 2 between 2 rotational matrices.
Where 𝑅(𝑞)12 is the rotation matrix from Sensor 1 to 2, rxy correlated to the row and column of the results in the relative rotation matrix, 𝑅(𝑞)2 is the global rotation matrix of the quaternions at Sensor 2, and (R(q)1) T is the transpose of the rotation matrix of the quaternions at Sensor 1.
This allowed the phi (φ), theta (θ) and psi (ψ) angles (flexion/extension, abduction/adduction, and internal/external rotations) to be calculated respectively:
Experiments and Validation Methods
Goniometer
OpenCap
Testing for Temporospatial Parameters
Data Collected and Analysed on MATLAB
The Attachment Method 2 was selected due to its achievement of 93% accuracy with merely 12.32 degrees of RMSE. Being only 0.32 degrees away from the commercial IMU sensors, yet cutting the cost by nearly 94% compared to an Xsens IMU sensor.
Robotic Rehabilitation Device
Knee Flexion/Extension vs Time
Collected from OpenCap
Ground Reaction Forces vs Time
Collected from the end-effectors of the robotic rehabilitation device
This project succesfully integrated the kinematic and kinetic analysis of a person on the rehabilitation treadmill. The IMU sensors were developed to accurately capture joint kinematics and spatiotemporal parameters (stance-swing ratio), with an accuracy of 94%. While the robotic rehabilitation device captured the ground reaction forces to identify any foot drop conditions. Although OpenCap is more efficient as it is capable of analysing most body parts all at once, no reliable estiomations of the ankle joint could be made, which is a concern. Hence, the best combination on the robotic treadmill is to implement OpenCap for joint kinematics, IMU solely for ankle analysis, and the end-effectors from the treadmill for measuring the kinetic parameters.