EMG driven musculoskeletal model for robot assisted stroke rehabilitation system
thesisposted on 27.02.2017, 06:04 by Jauw, Veronica Lestari
Neuro-rehabilitation is a medical process aimed at restoring the sensory and motor functions of the nervous system via repetitive learning and training process. The success of neuro-rehabilitation is simply measured by how fast the patients can recover from the disability. However, the continuous growth of the number of disabled patients and the lack of resources and rehabilitation facilities has then questioned the justification of its success in the future. There are questions about whether the present approaches of neuro-rehabilitation are still efficient to keep up with the future growth of the disabled patients which is growing exponentially. Hence, there is a need for constant advancement in neuro-rehabilitation engineering to improve the efficiency of automated rehabilitation. By doing this, the recovery rate of the patients can be increased, allowing more patients to be rehabilitated even with the lack of resources and rehabilitation facilities. The current study explores the utilization of 9 different upper limb’s muscles to find the best correlation between the electromyography (EMG) signals of the muscles and the individual torque of the upper joints via mathematical models. The EMG signals are acquired in real time and processed by rectification and filtration to eliminate the interference of the signal. These processed signals are then used as the input variables in GA; training the optimized correlation coefficients to minimize the discrepancy between the actual and simulated torque. The best correlation will be used as a control model in the rehabilitative robot’s controller as a decision making. Should the patient require assistance; this model will calculate the amount of assistive force required and supply it to the patient through the servo to complete the rehabilitation cycle. A comparative study, supported by the quantitative data of the simulation results, indicates that the individual torque of the upper limb’s joints is best modeled as the inverse logarithm of the EMG signals of the upper limb muscles with a constant. The estimated torque calculated from this model is the closest to the actual torque among the other mathematical models and thus, it has the lowest discrepancy. The average of 32% discrepancy is calculated throughout 100 generations in ±34 minutes. However, this figure is not attainable upon the integration of the inverse logarithm mathematical model into the controller mainly due to the communication rate between the DAQ system and the server. There is a discrepancy of approximately 10% between the experimental and simulated results which deems to be quite significant.