Precise localization and mapping are core challenges in autonomous navigation. LiDAR SLAM faces degradation issues in unstructured environments, where pose estimation becomes inaccurate in the presence of sparse or complex features, leading to distorted or failed mapping. To address the difficulties of LiDAR point cloud feature extraction and registration in such degenerate environments, this paper proposes a multi-sensor fusion framework that integrates Inertial Measurement Unit (IMU), LiDAR point cloud, and wheel odometry data. The system first compensates for motion distortion in LiDAR scans using IMU data. Then, it employs a probabilistic degeneracy-aware Iterative Closest Point (ICP) algorithm to align the current point cloud frame with local keyframes in a sliding window, providing a prior pose estimate. Subsequently, the current frame in the world coordinate system is fused with IMU and wheel odometry data using an efficient tightly-coupled Error State Kalman Filter (ESKF) to obtain the final optimal pose. The current frame is then added to the global map, which is maintained using an incremental k-d tree (ikd-Tree) data structure. Experimental results demonstrate that the proposed method achieves higher localization accuracy and robustness in feature-degraded environments compared to traditional LiDAR-inertial navigation approaches.