top of page
Search
Writer's pictureArun Eswara

Ortho, An Intelligent Exoskeleton: Powered Orthosis via Neural Network for Rehabilitative Assistance

For this project, I was awarded Regeneron Science Talent Search Scholar, MIT Think Finalist Alternate, Paradigm Project Finalist, Diamond Challenge Texas 2nd Place, Texas State Science Fair Finalist, Texas Junior Academy of Science 2nd Place, and several others.

 

Ortho is a project I began in the summer of 2018. I worked on it for a total of about 2.5 years.


When one of my friends was injured in a biking accident, he was forced to wear an 'arm orthosis', an arm brace that maintains bone alignment in the arm and/or leg during injury recovery. Orthoses reduce the stress on the Lateral Epicondyle and Medial Epicondyle tendons to facilitate movement and allow muscle recovery.

Orthoses

However, there are various problems with orthoses - they make muscle flexion difficult due to the weight of the brace and they slow recovery since the wearer tends to limit that arm or leg's usage.


The solution to this is a powered orthosis - an orthosis that is mechanically actuated by a motor at the primary joint. Unfortunately, existing prototypes [1] [2] are limited in scope and feasibility - they are heavy and contain a large battery, have many protruding pieces (making daily use uncomfortable or even impossible), and limit rehabilitation because the muscle isn't activated.


The solution to this was creating 'Ortho, An Intelligent Exoskeleton'. I had 3 primary goals with Ortho: Intelligent (able to automatically and accurately detect when the person was moving their arm to provide assistance), ready for daily use (not protrusive or overly heavy), and rehabilitative (balanced mechanical actuation with actual muscle flexion to help the muscle steadily recover).

The final prototype follows a 3-phase methodology: Sensing, Movement Actuation, then Termination. First, embedded pressure, electromyographic, and accelerometer sensors, provide data to an Arduino Uno. Secondly, a speed vector is calculated from the accelerometer and a movement threshold is found from the pressure/neuromuscular inputs - a Bayesian Artificial Neural Network (created in Matlab and trained in ~100 trials) is used to continually improve the accuracy of the movement threshold. If it is determined that the user is trying to move their arm/leg (based on the output of the Artificial Neural Network), the motor is automatically actuated via the Arduino Uno. Finally, the motor is disabled when the user moves their arm or leg in the inertial direction.


Below is a more precise description of the software methodology:

Muscle rehabilitation is promoted by steadily decreasing aid from the motor - as the user gets stronger, the aid decreases. That shift in motor power is modeled in the graph below:

Here is a picture of the final prototype (with the components spread out rather than encapsulated in the casing):

Brace Diagram

And here is a video of the arm brace in action:

To test the effectiveness of the brace, 4 facets of the data were analyzed: accuracy (based on false positives and omissions), torque, time to complete each cycle, and user energy expenditure.


Here is the data from the experiment, with the data analysis in the subtitle:

Regardless of the user's initial strength, the brace has high (98%-99%) accuracy
The Ortho brace decreases the length of a movement cycle by ~0.2 seconds and increases torque by up to 5x
The Ortho brace requires ~20N less force at its peak and far less prolonged force for arm movement
The Bayesian Regularization Artificial Neural Network required ~200 movements to gain usable (>95%) accuracy
The oscillating pattern of movement compared to that of the Neural Network's prediction is approximately the same

Finally, here is the statistical verification of the data:

I also calculated an industry verification to compare Ortho to an existing prototype (found here):

After completing the brace, I conducted an algorithmic analysis to make the brace as efficient as possible. Although battery optimization was initially an issue, efficiency-testing and hardware optimizations made the brace useable.

I then conducted a medical analysis to compare Ortho's viability to that of existing orthoses and prototypes, as shown below:

Finally, I conducted a simple cost-feature analysis to prove market viability:

 

I have reached the following conclusions from creating this project:

- Torque increased following use of Ortho system by approximately 5x under normal conditions in traditional movement patterns

- Energy Expenditure decreased significantly to the point where severely atrophied muscles could be rehabilitated more rapidly

- The time per movement cycle decreased approximately 0.2 seconds per flexion and extension or 0.4 seconds per cycle


Therefore, Ortho is more efficient for everyday use than a traditional orthosis in terms of the measured variables


The following future development could be done to improve the brace further:

- Incorporating other sensors to increase overall accuracy (more electromyographic sensors, dedicated electroencephalogram)

- Modelling a ball-and-socket join for omnidirectional motion and applicability/testing on all bodily joints that undergo significant structural damage and rehabilitation

- Creating a powered orthotic device for lower limbs to help with biomechanical walking gait characteristics, such as a powered Knee-Ankle Foot Orthosis (KAFO)

 

For a more thorough explanation of the project, you can scroll through this embedded Google Slides presentation that I used at several scientific conferences:



Thanks For Reading! If you have any questions, or would like to get in contact, you can reach out to me at aruneswara@icloud.com.

392 views
bottom of page