International Journal of Industrial Engineering & Production Research

This paper proposes an explainable artificial intelligence (XAI)–based framework for automating dye recipe formulation in industrial textile manufacturing, with a focus on yarn rope dyeing for denim production. A deep learning multi-output regression model is developed to predict the resulting yarn shade components (L_cable, a_cable, b_cable)  from heterogeneous industrial inputs, including customer-defined fabric shade targets, cotton fiber characteristics, and washing recipe parameters. To ensure transparency and industrial interpretability, Shapley Additive Explanations (SHAP) are integrated to provide global and output-specific explanations of the model’s predictions. The analysis reveals the dominant influence of cotton fiber properties, such as tenacity, micronaire, and fiber uniformity, alongside key controllable process parameters, including neutralization time, cellulose treatment duration, and detergent temperature. The proposed framework enables a clear distinction between raw-material-driven variability and process-adjustable levers, transforming the predictive model into an interpretable decision-support tool. The approach is validated using real industrial data from a Tunisian denim manufacturer and is readily transferable to similar textile dyeing and finishing processes.