Abstract: To solve the problem of precise prevention and control of rock burst in deep mines and promote the engineering application of digital rock mechanics technology, the EarthDive model is proposed as the core engine and systematic research is carried out on mine digital rock mechanics and rock burst tendency evaluation. Firstly, the core connotation of digital rock mechanics is defined, and its core differences from traditional rock mechanics are systematically compared from four dimensions: research paradigm, data acquisition, characterization scale and engineering adaptability, so as to clarify the technical advantages of digital rock mechanics. Standard coal and rock samples are prepared through laboratory tests, and multi-dimensional data including macro mechanical parameters, mesoscopic structural characteristics and physical response signals are collected. The dataset covers 8 major coal-producing provinces including Shanxi, Shaanxi, Inner Mongolia and Shandong, with burial depths ranging from 300 to 1 200 m, encompassing 5 coal ranks (gas coal, fat coal, coking coal, lean coal and anthracite) and two structural backgrounds (tectonic coal and primary structural coal), totaling 800 samples. The dataset is divided into training set (560), validation set (160) and independent test set (80) at a ratio of 7∶2∶1. A non-destructive prediction framework for rock burst tendency based on the EarthDive model is constructed, embedded with the professional knowledge graph of rock mechanics, and the laboratory accuracy verification is completed. A coal gene database covering macro-meso-micro data is built, and its core connotation, construction content and function realization are clarified to provide high-quality data support for model application. Finally, taking Tangshan Mine as the engineering background, a three-level on-site application architecture of “data layer-model layer-application layer” is constructed to complete the on-site evaluation of rock burst tendency and the analysis of application effects. The results show that the non-destructive prediction framework based on the EarthDive model has a high coefficient of determination (R²≥0.92) for mechanical parameter prediction, which has good reliability and applicability. The coal gene database effectively solves the pain points of scattered traditional coal and rock data and poor sharing, and realizes the standardized management and efficient reuse of coal and rock data. In the on-site application of Tangshan mine, this method accurately evaluates of rock burst liability. The closed-loop technical path of “laboratory verification-gene database support-engineering application” constructed provides a new technical means for the precise prevention and control of rock burst in deep mines, promotes the engineering application of digital rock mechanics in the coal industry, and has important theoretical and practical significance for the digital and intelligent transformation of the coal industry.