Assessment of matting agent effect on polyurethane coatings' perceived blackness through geometric properties modeling: An AI approach

Abstract

This study investigates the effect of matting agent on the perceived blackness of polyurethane automotive finishes by examining their impact on geometric properties using an AI-based neural network model. Black polyurethane coatings were prepared with varying gloss levels by incorporating different amounts of matting agent. Observers ranked these samples based on perceived blackness using the pair comparison method. The data features included the colorimetric properties and geometric attributes of the black polyurethane panels, with visual scales obtained as target values. Various neural networks were employed: Single-Layer Neural Network for linear regression, Multi-Layer Neural Network, Deep Neural Network, and Dropout-Enhanced Deep Neural Network for non-linear regression. Feature selection was applied to eliminate irrelevant or redundant features. The findings indicated that the Multi-Layer Neural Network for non-linear regression, utilizing geometric features, achieved favorable evaluation metrics, including mean squared errors, and R² values. Additionally, this model predicted the visual scale values more accurately compared to other neural network models. By simulating this behavior, the research effectively eliminates the need for time-consuming visual assessment experiments traditionally used for blackness evaluations in industrial settings. This method greatly simplifies the evaluation process, conserving both time and resources while maintaining reliable and precise assessments.