Machine Learning-based expert system for the coating manufacturing industry for predicting coating performance

Although Jotun may be best known for decorative house paints, exterior and interior, we are one of the largest suppliers of protective coatings for ships, infrastructure and energy industry. We have a strong focus on developing sustainable solutions for our customers, and it is therefore very important for us to accurately predict how long our coating systems can protect, e.g. an offshore wind turbine. New technology enables collection of large amounts of data on the performance of our coating systems, and in this project we will explore the opportunities to use machine learning for protective coating lifetime as well as optimization of coating formulations for improved performance and more sustainable products. The coating manufacturing industry is experiencing the effect of many variables on its manufacturing and development processes. If one is trying to approach the question of achieving an optimal coating recipe, that will provide properties, desired by a customer, the recipe development process happens to be a combination of an educated guess and a “trial-and-error” approach. Objective 1: To address the existing issues in the knowledge reusability, and the lack of detailed insight into “coatings property – coatings recipe” relations in the coating manufacturing industry via creating Machine Learning Predicting tools, that will contain a series of Machine Learning models, which will be trained on currently available R&D data on the performance of protective coatings systems, and the physicochemical parameters of coating’s raw materials and the coatings itself. Objective 2: The secondary goal of the project is to shift from a categorical representation of components in the recipe to a more uniform representation via textual identifiers like InChi or SMILES (simplified molecular-input line-entry system: one-line string molecular description), which will contain both chemical and physical information about the components in a way, easily interpretable by machine learning algorithms. Such identifiers would help to describe raw materials in terms of their chemistry rather than generic or trade names.

This project employs interdisciplinary approaches. The objective of the project is to build a Machine Learning tool, that will contain a series of machine learning models, which will be trained on currently available Research & Development data on the performance of protective coatings, with the aim to obtain an expert system that will assist with estimating organic coating lifetime and will suggest optimal coating recipes that would have good performance at certain environmental conditions. While the coating is being in-service, although data about its performance can be collected, no formalized connection normally is established between coating composition and properties. Typical organic coating recipes contain numerous constituents, such as polymer resins, solvents, extenders, rheological additives, and other additives, that might come from various sources. Each of these constituents, while having a function of its own, has an effect on the functionality of all other components present in the coating, and, overall, does affect the final properties of an organic coating. Therefore, the connection between coating performance and composition appears to be multiparametric and non-linear, which is very difficult to assess. The coating manufacturing industry is experiencing the effect of many variables on its manufacturing and development processes. If one is trying to approach the question of achieving an optimal coating recipe, that will provide properties, desired by a customer, the recipe development process is happening to be a combination of an educated guess and a “trial-and-error” approach. Therefore, developing machine learning tools, that would assist R&D professionals, revealing connections between coating composition and coating performance seems a justified cause for a study. Such tools would allow the development of better-performing and more sustainable organic coating recipes in a shorter time span, increasing the efficiency of research activities.