Novel formulation optimization using Big Data, modeling, and predictive tools

 
Big data analytics and machine learning (ML) algorithms could be useful to improve accuracy of formulations optimization and performance predictions. A case study was conducted to explore European regionally relevant broad formulation spaces ranging from white to deep bases using novel matting resin that provides an engineered surface with controlled topology and functionality. Optimal design-of-experiments strategy combined with high-throughput methodology were utilized to evaluate about 1000 paint formulations which covered formulations space of greater than 38000 combinations. Data modeling of paint properties were carried out applying linear regression to ML algorithms for comparing accuracy of the models. Applied methodologies were Taylor-Series Expansion involving main effect and higher order interaction terms, Partial Least Square, Bootstrap Forest, and Neural Network. Predictive models were further utilized to optimize performance and simulate preferred compositional space for target balance of properties. Optimal formulations were validated using bench top approach.