Application of a Multilayer Perceptron Neural Network for Classifying Software Platforms of a Powered Prosthesis through a Force Plate

Abstract

The amalgamation of conventional gait analysis devices, such as a force plate, with a machine learning platform facilitates the capability to classify between two disparate software platforms for the same bionic powered prosthesis. The BiOM powered prosthesis is applied with its standard software platform that incorporates a finite state machine control architecture and a biomimetic software platform that uniquely accounts for the muscle modeling history dependence known as the winding filament hypothesis. The feature set is derived from a series of kinetic and temporal parameters derived from the force plate recordings. The multilayer perceptron neural network achieves 91% classification between the software platforms for the BiOM powered prosthesis conventional finite state machine control architecture and biomimetic software platform based on the force plate derived feature set.