Among a wide range of fiber-reinforced composites, those with randomly oriented short fibers, which are also known as random-chopped fiber-reinforced composites (RaFCs), are the most common composites owing to its ease of manufacturing, flexibility of composite shapes, and good material properties, including light weight and high stiffness. These properties of RaFCs are involved with the lengths and distributions of fibers inside the composites. However, inspecting the fiber lengths and distribution remains a challenging problem, particularly when the lengths and locations of individual fibers need to be distinguished using only X-ray transmission images. The main difficulty arises from the variety of fiber widths and their frequent intersections. To address this problem, this paper proposes a comprehensive software system to localize fibers and measure their lengths. Our system is inspired by a previous work for tracing human hair strands. To adopt the previous method for RaFCs, our system extends classic Gabor filter to explore the locally best parameter sets to suit different fiber shapes. With this adaptive filter, we can extract the locations and orientations of local fibers more robustly for RaFCs. Then individual fibers are traced by solving an initial value problem of an ordinary differential equation. To avoid erroneous tracing which typically occurs at intersections, our method traces only the non-intersecting parts of the fibers initially. After that, we connect the fiber segments using the proximity of their endpoints and the orientations. Through experimental validations on different fiber samples, we demonstrate the stability of the fiber tracing and the robustness of the fiber length calculation. Our system works properly even for X-ray radiographic images of heavily tangled fibers in carbon-fiber-reinforced thermoplastic laminates taken by X-ray Talbot–Lau interferometer.

Using a 3D scanner to inspect a 3D print is becoming more significant as the spectrum of applications for fused filament fabrication expands. We use the tool path, which describes the 3D printing process, as the nominal data to evaluate the manufacturing errors. However, since the 3D print’s coordinate system in the scan data is unknown, the 3D print must be aligned with the tool path before the nominal-actual comparison. Therefore, we present a novel alignment based on the 3D print’s microstructure. The proposed method first estimates the build direction and then aligns the 3D print with the tool path along the build direction. Finally, each layer of the 3D print is aligned with its corresponding layer of the tool path in the horizontal direction. The alignment result can be confirmed through the slice view. In addition, the performance, extension, and applications of the proposed method are discussed. GRAPHICAL ABSTRACT

The Feldkamp, Davis and Kress algorithm is a computationally efficient reconstruction method for three-dimensional cone-beam computed tomography. However, it suffers from severe artefacts when the number of projections is insufficient. Although recent deep learning–based methods have succeeded in reconstructing such sparse-view projections, it is still challenging to reconstruct a large 3D volume efficiently because of heavy memory consumption and difficulty in obtaining sufficient training data. Therefore, we propose a deep learning method to overcome these drawbacks. Our method consecutively reconstructs short bars in 3D CT volume using the intensities of pixels on the moving trajectory projected on the detector of these bars. Then, we feed such pixel intensities into a neural network and train it to simulate the filtering and back-projection processes of the FDK algorithm. Since the reconstruction volume is separated into bars, the neural network can be used with only a small amount of memory. Furthermore, the network can be trained sufficiently with only a few training samples because plenty of bar data can be extracted even from a single CT image. We experimentally demonstrate that our approach works efficiently for both simulated and actual sparse-view CBCT data using training data extracted from only a single CT image.