Enhancing Paint Formula Innovation Using Generative AI and Historical Data Analytics

Abstract

This work implements a generative artificial intelligence approach and a search algorithm to recreate and innovate paint formulas on a hardware-near level. We use clustering techniques to explore data latent space and then implement Markov Chain Monte Carlo techniques to sample from the posterior distribution. The approach is designed for multi-objective optimization problems, where in the case of paint formulas, the objectives are the cost and the color difference from a reference formula. We demonstrate the feasibility of the idea by building and testing water-based acrylic paint formulas. The approach is general and can be applied to many different formulations at chemical or hardware-near levels.

The paint industry is very competitive and new formulations must be assembled quickly while adhering to predefined criteria regarding sustainability, cost, and performance. In recent years with the advent of generative artificial intelligence approaches, large-scale and rapid assembly and design of formulations became possible. Current commercial approaches to develop new formulations are based on an AI-trained linear regression type approach that does not produce chemical-related formulations when trained with data at the hardware-near level and can also lead to a lower diversity of ideas since it is limited to extrapolating small changes around the reference samples. In this work, we implement a generative artificial intelligence approach that permits to sample of the posterior search space by using tools such as Markov Chain Monte Carlo techniques, to provide access to a wide diversity of potential new formulations. Our approach accomplishes this by employing a search sub-algorithm that relies on prescribed objective functions such as cost and color difference with a reference formulation, that is sampled for many different formulations, and pre-processes the data. We tested our idea using compositional data from water-based acrylic paint formulas that were developed and tested. Our implementation is general and can be used to develop new formulations for many different products at the chemical formulation or hardware-near levels.