PARAMETRIC OPTIMIZATION OF SOFT ROBOT BIONIC HAND MANIPULATOR USING SUPERVISED MACHINE LEARNING

• Main Author: Amajida, Fawaz
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85467
• Institution: Institut Teknologi Bandung

As much as 16% of global population are disabled with one of highest prevalence is caused by hand amputation. One of the solutions of that condition is to use bionic hand. Bionic hands in general are made of rigid material such as plastics and metals that are not as flexible and adaptive as human hand. Thus, soft material is applied to make it interacting more adaptively and safely. Biomechanics Research Team of ITB has developed soft robot bionic hand using combination of two materials and equipped with rigid structure and strain limiter to make the movement ‘hand-alike’. The movement of the bionic hand can be enhanced by making the range of motion of its finger to be 90 degrees. Unfortunately, the design optimization of soft robots requires huge computational load due to its complex geometry and nonlinearity. Approach of machine learning is conducted to find the optimum design with lower computational load in a shorter time. Linear regression, decision tree, random forest, k-nearest neighbours, and support vector machine algorithms are evaluated in this research. Those models are trained using datasets that are generated through finite element method simulation of parameters-permutated samples. Machine learning models’ predictions are validated by simulating the design using finite element method. Decision tree successfully predicted the design of index finger manipulator with 16% improvement on range of motion accuracy while support vector machine predicted the thumb and little finger manipulator design with 11% and 21% improvement on range of motion accuracy compared to previous research’s design. Further study is required to explore more algorithms and optimization methods widely.