Abstract: To address the urgent demand for transparency in geological support amid the intelligent construction of coal mines, and targeting the problems of difficult information acquisition, low recognition accuracy, and insufficient intelligence in existing methods for identifying formation lithology and working conditions while drilling in underground coal mines, this paper proposes the definition and technical connotation of intelligent perception while drilling in underground coal mines by integrating the development of artificial intelligence and new mine measurement-while-drilling technologies. It further elaborates on the technical path for realizing intelligent perception while drilling driven by the fusion of drill string dynamics mechanism and multi-source measurement-while-drilling data. The dynamic response characteristics of drill string is analyzed, the interaction mechanisms of bit-rock and drill pipe-borehole wall, along with the dynamic evolution laws of drill pipe sinusoidal and helical buckling during drilling, are revealed, laying the theoretical foundation and defining the data boundaries for intelligent perception while drilling technology. The acquisition technical schemes for multi-source heterogeneous raw measurement-while-drilling data are determined, and data preprocessing is conducted in accordance with the time series. Based on the communication network composed of three tiers: the access layer, convergence layer and core layer, a three-stage progressive fusion strategy for multi-modal drilling data across the data layer, feature layer and decision layer is proposed. And an edge-cloud collaborative architecture for drilling information transmission and processing is constructed. Combined with the actual production conditions of the Huainan-Huaibei mining area, an engineering application studies are carried out to verify the effectiveness and practicality of the proposed technical system. Based on the Fuzzy C-Means (FCM) clustering method, the perception of drilling-while-rock formation hardness for three types of rock formations in cross-measure borehole construction was realized without prior labels in Huaibei coal mine, and the performance outperforms that of traditional clustering methods. Based on the Whale Optimization Algorithm-optimized Kernel Extreme Learning Machine (WOA-KELM) method, effective prediction of stuck pipe working conditions in the construction of directional long boreholes was achieved in Huainan coal mine, which provides a decision-making basis for the closed-loop control of intelligent drilling. Engineering application results demonstrate that the proposed technical system significantly enhances the real-time geological information perception and intelligence of underground drilling in coal mine, offering critical technical support and engineering reference for transparent geology mine construction and the advancement of intelligent drilling technology.