Machine Learning-integrated digital twins for process optimization in Industry 5.0

Abstract

This study responds to the challenges of Industry 5.0 by proposing a hybrid model that integrates Ordinary Differential Equations (ODEs) with Machine Learning (ML) algorithms within a Digital Twin (DT) architecture. The proposal is applied to a detergent manufacturing plant, updating processes focusing on sustainability, resilience, and human-centeredness. The study was conducted in a real industrial plant, with operational data collected from SCADA, MES, and ERP systems. This paper proposes a hybrid model that integrates Machine Learning (ML), Ordinary Differential Equations (ODEs), and a Digital Twin framework for process optimization in the manufacturing industry. The variables were treated in modular architecture and tested within the ISO 23247 framework, with real-time visualizations through human-machine interfaces (HMI). The hybrid approach showed significant gains in predicting chemical solutions (R² = 0.80), sulfonic acid consumption (R² = 0.9998), and intelligent reactor allocation (80.7 % accuracy). In addition, the system predicted laboratory delays with 78.1 % accuracy and enabled significant reductions in loading times and operational deviations. In contrast, raw materials such as caustic soda, water, and laurel showed lower predictive performance, reinforcing the need for additional explanatory variables. The model enhances the potential of predictive AI combined with physical modeling for more sustainable, resilient, and human-centered decisions. Integrating ML and ODEs into a DT promotes operational and strategic gains for the detergent industry, aligning with the principles of Industry 5.0. The demonstrated approach is effective, scalable, and capable of transforming industrial data into optimized decisions, directly impacting the production process's efficiency, sustainability, and autonomy.