AI-Enhanced Motor Power Analysis for Joint Kinematics Prediction and Fault Detection in Rehabilitation Systems

Rehabilitation is essential for post-stroke patients in restoring motor functions. More importantly, the safety of the patients on using assistive systems for therapy is a major concern. This paper introduces to a novel redundant mechanism adopted fault-alert exoskeleton based wheelchair design effective for upper and lower limb. Primarily the device facilitates robotic-assisted physiotherapy exercise system for post-stroke rehabilitation targeting wrist, elbow, shoulder, knee, ankle and backrest. Accurate joint angle measurement in limb is crucial in rehabilitation device to ensure precise feedback. Commonly available angle detection sensors such as Inertial Measurement Units (IMU's) and rotatory encoders often suffers from drift and experience slippage that affects measurement accuracy. To detect these errors, we propose a joint angle prediction system built on power analysis extracted from current sensor using machine learning (ML) model. Additionally, based on the physicality of the user under observation, the proposed Collective Generator Network (CGN) synthesizes power profile for the action to be performed. Hence both these models augment the rehabilitation system to detect electrical and mechanical disconnects from the motors, and encoder issues. The ML, CGN models, and datasets are made freely available for further adoption by the designers and researchers community.Clinical relevance-Integration of joint kinematics prediction unit built on motor power analysis, and CGN model generating the power profile for the actions to be performed aims to detect subtle fault variations within the system which otherwise remains unnoticed and helps in delivering the best service to the patients under usage.