Abstract: Most Simultaneous Localization and Mapping (SLAM) systems that utilize tightly coupled laser-inertial odometry are often under-constrained in underground coal mines, leading to mapping failures and hindering the direct deployment of robots in operational frontlines. This paper analyzes the factors contributing to odometry drift in degraded environments, elucidating the mechanisms by which unstructured point clouds induce drift. We propose a targeted detection method for unstructured regions and a two-step denoising approach that includes filtering and interpolation to address dust and fog. An adaptive tightly coupled odometry system is developed, incorporating a factor that characterizes the state of unstructured areas. In a test environment reverse-engineered from the WHU-TLS Tunnel dataset, our system achieves an E_AP-E_RMS of approximately 0.40 m, an E_AP-Mean of 0.37 m, a E_RP-E_RMS of 0.015 m, and an E_RP-Mean of 0.011 m, outperforming other methods significantly. For engineering applications with Ultra-Wide Band devices, we use a nonlinear optimization method to maintain global odometry poses, constructing a factor graph with unstructured state constraints, UWB factors, and loop closure factors. By performing parallel global optimization on all factors, high-precision global localization and mapping are achieved. In a kilometer-scale simulated tunnel with UWB signals, the continuous mapping results show an E_AP-E_RMS of 1.48 m, an E_AP-Mean of 1.32 m, an E_RP-E_RMS of 0.026 m, and an E_RP-Mean of 0.013 m. In a 2 000 m-long, loop-free field test at Cuncaota Coal Mine, the E_AP-E_RMS was 15.64 m, the E_AP-Mean was 14.53 m, the E_RP-E_RMS was 0.198 m, and the E_RP-Mean was 0.037 m, demonstrating the system's potential for practical engineering applications. Our SLAM system, built with a 16-line LiDAR and a 6-axis IMU that meet explosion-proof standards, provides accurate mapping in degraded scenarios, significantly advancing the practical application of coal mine robots.