This paper proposes a three-dimensional mapping system for in-hold environments oriented toward cleaning operations, to address the operational characteristics of bulk carrier holds—namely enclosed spaces, weak geometric features, and the continuous removal of bulk materials during cleaning, which leads to a gradual decrease in pile height and constantly changing surface morphology—aiming to achieve a stable and updatable 3D representation of the hold environment. A multi-LiDAR and inertial measurement unit (IMU) collaborative 3D mapping system is built on a hold-cleaning machine platform. Spatiotemporal consistency across sensors is ensured through time synchronization and extrinsic calibration. Motion compensation of LiDAR point clouds is performed using IMU preintegration, and pose estimation is carried out via a tightly coupled LiDAR and IMU odometry method. Combined with sliding-window local mapping and an incremental voxel-based map representation, the system enables real-time construction and continuous updating of the in-hold 3D map. Verification through multiple field experiments shows that the proposed system can robustly estimate the pose of the hold-cleaning machine under feature-sparse and highly dynamic operating conditions, while producing structurally complete and continuously updated 3D point cloud maps. The generated maps effectively capture key structures, including hold walls, the hold floor, and bulk material surfaces, thereby meeting the requirements for remote visualization and operational awareness. The proposed system is well suited to bulk carrier hold-cleaning scenarios and provides a reliable foundation for three-dimensional environmental perception in autonomous hold-cleaning operations.