Research and application of a digital twin model integrated with continual learning for burn-through point prediction

Abstract

The position of the burn-through point (BTP) is a critical parameter that significantly impacts the quality of the sinter and production efficiency. To address the limitations of traditional models, such as poor adaptability to dynamic industrial environments, weak visualization capabilities, and lack of guidance, this paper proposes a digital twin model for the sintering process, integrated with continual learning. By deeply merging digital twin technology with the sintering process, the model constructs a five-dimensional framework that includes physical entities, virtual environments, BTP prediction, twin data, and virtual-real connectivity. This model enables real-time monitoring and optimization of the sintering process. For prediction, the elastic weight consolidation-long short-term memory model is employed to forecast the BTP, allowing online model updates through continual learning, thereby preventing model ageing and catastrophic forgetting. Experimental results demonstrate that the proposed digital twin model exhibits strong stability and adaptability in dynamic sintering environments, offering an advanced approach to the digital transformation of sinter production.