An LLM-guided SD-LDM Digital Twin Construction Strategy (LSDT) for multi-industrial scenarios: Enhancing adaptability and efficiency

In the rapidly evolving landscape of Industry 5.0, Digital Twin (DT) have emerged as a transformative technology across various industrial sectors. However, as DT theory and practice progress, a critical issue arises: the prolonged building time associated with implementing DTs. To address this challenge, this paper proposes a rapid DT construction method: LLM-Guided SD-LDM Digital Twin Construction Strategy (LSDT) for multi-scenarios. Firstly, we introduce a cross-modal generation framework. This framework leverages Large Language Model (LLM)-Guided Stable Diffusion- Latent Diffusion Model (SD-LDM) technology, which is capable of swiftly constructing high-quality 3D models based on limited multimodal data. Subsequently, the generated models are transferred into the Digital twin construction framework. This framework incorporates both the DT construction method and the assembly and fusion method, enabling the realization of a multi-scale, multi-level DT construction. Finally, we conducted case study in R&D laboratories, prototype warehouses, and packaging units. The multi-dimensional scoring results showed that the model construction efficiency improved significantly, with peak values reaching 39 % (across models) and 73 % (single model), while usability scores peaked at 13.84. Furthermore, the constructed DT successfully met the core Ss requirements of the scenarios. These results indicate that the LSDT method accelerates the efficiency for DT construction and offers good adaptability.