Abstract: The in-situ processing mode of large components places very high requirements on the positioning accuracy of the processing equipment. The traditional iterative closest point (ICP) algorithm is prone to fail in registration when the initial poses of the source point cloud and the target point cloud are significantly different and has poor robustness in noisy environments, which may lead to incorrect matching. To address these issues, a component point cloud registration method based on the fast and robust-iterative closest point (FR-ICP) algorithm is proposed. The Stanford dataset and component point clouds are voxelally downsampled, key points are extracted based on intrinsic shape signatures (ISS), and the fast point feature histograms (FPFH) feature descriptors are calculated. The point cloud is coarsely registered using the random sample consensus (RANSAC) algorithm. The ICP algorithm from point-to-plane is improved by introducing the Tukey robust function, and the point cloud is finely registered. Experiments have shown that the improved algorithm can effectively enhance the robustness, convergence speed, and accuracy of point cloud registration in noisy environments as well as the registration success rate when there is noise and a significant initial pose difference.