Tomoyuki Fujimori

Skeleton surface generation from volumetric models of thin plate structures for industrial applications

Computed Tomography (CT) is a powerful non-destructive measurement technology to generate cross-sectional X-ray images of objects, fromwhich three-dimensional volumetricmodels can be constructed. In this study, we focus on industrial applications for X-ray CT in the analysis of thin plate structures, and propose in particular a method to generate sufficiently accurate skeleton meshes from a volumetric model of a thin plate structure. We use a geodesic-based skeletonization algorithm to extract skeleton cells, which can then be contoured to generate a skeleton surface. Since the thin plate structure has junctions, the skeleton surface is non-manifold around these areas. Because of this, the Marching Cubes algorithm is extended to enable handling ofmultiple labels (signs) to generate non-manifoldmeshes. The generatedmesh is optimized to pass through the mid of the plate by interpolating mapping. Some experimental results for industrial samples are shown using our implemented prototype system, and it is proven that the accuracy of the generated mesh is sufficient for industrial applications.

Surface extraction from multi-material CT data

This paper describes a method for extracting surfaces from multi-material CT (computed tomography) data. Most contouring methods such as marching cubes algorithm assume that CT data are composed of only two materials. Some extended methods can extract surfaces from the multi-material (non-manifold) implicit representation. However, these methods are not directly applicable to CT data which are composed of three or more materials. There are two major problems that arise from fundamentals of CT. The first problem is that we have to use n(n - l)/2 threshold values for CT data contains n materials and select appropriately one threshold value for each boundary area. The second is that we cannot reconstruct only from CT data in which area three or more materials are adjacent each other. In this paper, we propose a method to solve the problems by using image analysis and demonstrate the effectiveness of the method with application examples construct polygon models from CT data of machine parts.

Separated medial surface extraction from CT data of machine parts

This paper describes a new method for extracting separated medial surfaces from CT (Computed Tomography) data of machine parts. Plate structures are common mechanical structures such as car body shells. When designing such structures in CAD (Computer Aided Design) and CAE (Computer Aided Engineering) systems, their shapes are usually represented as surface models associated with their thickness values. In this research we are aiming at extracting medial surface models of a plate structure from its CT data so as to be used in CAD and CAE systems. However, such a structure consist of many components which are adjacent each other. For example, car body shells are consist of many welded plates. CT imaging technology has some weak points in the area. One of them is that, if there are two or more objects made of same material, CT scanner cannot make the distinction between them. The problem is caused by the principles of CT imaging technology. Because CT image represents the mass distribution within a cross section, we cannot separate the objects only from image information. However, there are many requests for scanning assembled parts and separating objects made of same material. Therefore, we propose a method to separate each components. CT data has not enough information amount as has been metinoned, so we adopt other knowledge about model shapes. We conclude with experiments on welded mechine parts for effectiveness of our method.

Out-of-core distance transforms

This paper presents a method for computing distance fields from large volumetric models. Conventional methods have strict limits in terms of the amount of memory space available, as all volumetric models must be allocated to the random access memory (RAM) to compute distance fields. We resolve this problem through an out-of-core strategy. Our algorithm starts by decomposing volumetric models into small regions known as clusters, and distance fields are then computed by Local Distance Transform (LDT) and Inter-Cluster Propagation (ICP). LDT computes the distance transform for each cluster, and since it is independent, other clusters can also be saved to the storage medium. ICP propagates the distance at the boundary of the cluster to neighboring clusters to remove inconsistency in distance fields. In addition, we propose an efficient ordering algorithm based on the propagated distance to reduce LDT and ICP. This paper also demonstrates the results of distance transform from volumetric models with over a billion cells.