A methodology leveraging digital twins to enhance the operational strategy of manufacturing plants in unexpected scenarios

Currently, the increasing need to operate industrial plants more sustainably and efficiently, along with the uncertainty present in their operation, is forcing those responsible for operation and maintenance to adopt intelligence solutions that support optimized decision-making in these scenarios. Real-time decision-making frameworks face significant challenges due to the unpredictability of operational conditions and the growing pressure to achieve sustainability goals. Although digital twin has been explored in industrial applications, existing studies often lack real-time adaptability and fail to optimize operation and maintenance efficiency, environmental impact, and cost management simultaneously. To this end, this article proposes a digital twin-based methodology to dynamically optimize operating strategies and maintenance shutdowns in real-time. It adapts to process disturbances and ensures efficient production while reducing costs, carbon dioxide emissions, and surplus. These disturbances may include electricity price fluctuations, component degradation, or increased demand, among other factors. By applying the proposed methodology to a use case, reinforcement learning and neuronal networks enable real-time decision-making, achieving a 5% cost reduction per tonne, 5% lower carbon dioxide emissions, and a 30% improvement in demand adjustment. The results show that this approach enhances operational efficiency and sustainability in industrial plants. • Proposal of a methodology (DTM4SM) based on digital twin for real-time decision-making. • Reduced order models with measured and synthetic data to enhance simulation accuracy. • Deep learning algorithms to accelerate simulation process. • Machine learning and metaheuristics algorithms for real-time optimization. • Validation of DTM4SM through a use case for the steel industry.