Viscoelastic behavior prediction of natural rubber blends via machine learning and its application for reverse engineering materials

With the increasing demand to focus on sustainability, natural rubber is a competitive alternative for materials within the automotive, footwear, aerospace, and consumer goods industry. Its ultrahigh molecular weight provides the user with a highly resilient linearly viscoelastic raw material capable of reaching strains up to 400%. Upon receiving the raw natural rubber, it then needs to be mixed with the appropriate additives to reach the desired specifications that the customer establishes. For that reason, formulators within the natural rubber industry are given the critical responsibility of providing the manufacturing team with the desired formulation based on the key performance indicators. For industry, these formulators must have acquired years of experience to minimize the number of iterations needed to arrive at the final formulation to save time and money. This study focuses on merging machine learning and polymer science to provide the industry with a reverse engineering tool capable of predicting natural rubber blend formulations based on key performance indicators dictated by the application. To test this methodology, the goal was to create a formulation that would reach similar compressive viscoelastic properties to an athletic footwear midsole by utilizing Artificial Neural Networks. The algorithm requires three inputs: the maximum stress reached during relaxation, the rate at which stress decays during relaxation tests, and the tanδduring cyclical testing in compression; all large-amplitude tests used to simulate actual deformations during walking/standing for an extended period of time. The algorithm would output the sulfur content, foaming content, and plasticizer content needed to create a material with a specific durometer reading, tano, max stress in relaxation, and a specific relaxation decay behavior with an accuracy of 96.75%.