Abstract: The mine hoist is the core equipment for production and transportation in coal mines, and its safe and stable operation is a critical foundation for ensuring continuous production and safe operations. General-purpose large language models face challenges in hoist operation and maintenance (O&M) scenarios, such as insufficient understanding of physical signals, weak visual perception capabilities, and inadequate reliability in intelligent decision-making. To address these issues, a knowledge-embedded vertical domain large model for intelligent O&M of coal mine hoists is constructed, establishing an intelligent O&M framework that integrates vibration signal mechanism perception, visual structured evidence generation, and domain knowledge-enhanced decision-making. The model aligns the physical features of vibration signals with the semantic space of the large language model through a multi-head cross-attention mechanism, and introduces a mechanism knowledge injection strategy to enhance the interpretability of fault diagnosis. An external operator intelligent scheduling mechanism for multi-vision tasks is established to convert the states of key components, such as head rope sway, tail rope entanglement, and residual coal in skips, into structured semantic evidence. By incorporating retrieval-augmented generation technology, domain knowledge including safety regulations, maintenance manuals, and historical logs is integrated to improve the reliability of model outputs. Test results on public bearing datasets show that the model achieved an average fault diagnosis accuracy of 99.44%, while on a self-built on-site visual dataset for coal mine hoists, the visual average detection accuracy reached up to 96.67%. Based on the constructed vertical domain large model, an intelligent O&M large model platform for coal mine hoists was developed, integrating functions such as digital twin state monitoring, large model intelligent question answering, dynamic knowledge base management, and full-process management of visual models. Engineering application results at the Xutuan Coal Mine in Huaibei demonstrate that the platform supports tasks including multi-source heterogeneous data access, abnormal state identification, domain knowledge retrieval, and standardized O&M report generation, thereby enhancing the interpretability, standardization, and engineering applicability of O&M decision-making. This provides a new technological pathway for the transformation of coal mine hoisting equipment from traditional condition monitoring to knowledge-driven intelligent O&M.