Deep Learning based 3D Volume Correlation for Additive Manufacturing Using High-Resolution Industrial X-ray Computed Tomography

TL;DR

The paper tackles voxel-level Digital Volume Correlation (DVC) between CAD models and high-resolution XCT scans in additive manufacturing, a challenging modality-mismatch problem with large data sizes. It introduces an unsupervised 3D registration approach based on VoxelMorph that employs a dynamic patch-based training pipeline and HDF5 data handling to estimate voxel-wise deformations from CAD to XCT. A Binary Difference Map (BDM) is proposed as a complementary metric to Dice Score to quantify voxel-wise matches and mismatches. On TPMS-based AM specimens, the method improves Dice from 82% to 94.7% and BDM0 from 68.27% to 89.93%, while reducing inference time from days to minutes and outperforming SPAM, enabling faster quality control and potential generation of compensation meshes for closed-loop AM processes.

Abstract

Quality control in additive manufacturing (AM) is vital for industrial applications in areas such as the automotive, medical and aerospace sectors. Geometric inaccuracies caused by shrinkage and deformations can compromise the life and performance of additively manufactured components. Such deviations can be quantified using Digital Volume Correlation (DVC), which compares the computer-aided design (CAD) model with the X-ray Computed Tomography (XCT) geometry of the components produced. However, accurate registration between the two modalities is challenging due to the absence of a ground truth or reference deformation field. In addition, the extremely large data size of high-resolution XCT volumes makes computation difficult. In this work, we present a deep learning-based approach for estimating voxel-wise deformations between CAD and XCT volumes. Our method uses a dynamic patch-based processing strategy to handle high-resolution volumes. In addition to the Dice Score, we introduce a Binary Difference Map (BDM) that quantifies voxel-wise mismatches between binarized CAD and XCT volumes to evaluate the accuracy of the registration. Our approach shows a 9.2\% improvement in the Dice Score and a 9.9\% improvement in the voxel match rate compared to classic DVC methods, while reducing the interaction time from days to minutes. This work sets the foundation for deep learning-based DVC methods to generate compensation meshes that can then be used in closed-loop correlations during the AM production process. Such a system would be of great interest to industries since the manufacturing process will become more reliable and efficient, saving time and material.

Figures (6)

• Figure 1: CAD models and corresponding XCT scan cross-sections for TPMS samples with different level-set parameters C.
• Figure 2: Overview of VoxelMorph-based 3D registration framework.
• Figure 3: Overlay of XCT (grey) and CAD (green) of the central slices for the sample with C = -0.6, shown before (a, b, c) and after (d, e, f) non-linear registration. Views include: (a, d) XZ plane, (b, e) YZ plane and (c, f) XY plane.
• Figure 4: Visualization of the deformation field estimated by the VoxelMorph model on the same slice shown in Fig. \ref{['fig:ctandmovedct']}. (a, b, c) shows the magnitude of the displacement field in the central slices along the XZ, YZ, and XY planes, respectively. The bottom row (d, e, f) presents the corresponding displacement field of the same slices.
• Figure 5: 3D visualization of registration results. a) Reference CAD model b) Original XCT volume c) Registered XCT volume aligned to the CAD geometry using the VoxelMorph model d) Masked displacement magnitude map using registered XCT volume. The red boxes highlight a region with severe structure breakage.