Modeling the 3D internal crack under compression entails complex fracture mechanics (mode I-II-III fracture), resulting in substantial computational costs and challenges in characterizing fracture morphology characterization for Phase Field Method (PFM) simulations. In the present paper, we developed a 3D adaptive crack propagation model integrated into the PFM to study the internal cracks in rocks. The proposed locally refined solution was implemented within the isogeometric analysis (IGA) framework, utilizing PHT-splines and employing the phase field variable as an error indicator to accurately capture non-smooth crack surfaces. The simulation begins with a coarser mesh, and the spatial discretization is adaptively refined in real-time as the computation progresses and the phase variable threshold is reached. By comparing our results with previous numerical tests, we validated the effectiveness of our model in simulating crack growth. Notably, a smaller length scale parameter led to a higher peak force in notched square plates under tension, and our adaptive PFM successfully reduced errors caused by length scale parameters. We further validated our model by comparing it with uniaxial compression experiments on 3D-printed resin samples with internal penny-shaped cracks, achieving good agreement with the experimental data. Using the 3D adaptive PFM, we predicted changes in crack morphology and fracturing mechanisms by varying the inclination angles of internal penny-shaped cracks. Our numerical results showed that as the inclination angle increased, wing cracks initiated closer to the end tips of the original crack, and their length decreased due to the wings wrapping effect. The corresponding fracturing mechanisms transition from mode I-III fracture (0°) to mode I-II-III fracture (30°, 45°, 60°), and eventually to mode I fracture (90°) as inclination angles increase. Comparing 2D and 3D models, we found that the growth of wing cracks in 3D was significantly influenced by the intermediate principal stress. We conclude that the extended 3D adaptive PFM-based model accurately predicted complex mechanical behavior and crack growth patterns in rock.