As a in-process measurement method, we focused on sparks emitted when thermal energy generated by the grinding process flows into chips. However, it has not been possible to estimate the processing state from the observation of sparks. Therefor e, the objective of this study was to observe sparks generated during the grinding process by image processing and to investigate the relationship between the number, area, and brightness of sparks and the amount of metal removal and thermal effects on machined surface in the machining state. In this experiment, two types of grinding wheels were used to grind S45C and SUS316 by varying the depth of cut. To simplify the observation of sparks, all experiments were performed using dry grinding. The experimental results show that grinding resistance increases linearly with increasing depth of cut, and the number and area of sparks increase exponentially. The same trend was observed when the grinding wheel was changed. Furthermore, as the depth of cut was increased and an oxide film formed on the surface of the work piece, the tensile stress increased linearly and the number, area, and brightness of sparks increased exponentially. These results suggest the possibility of estimating the amount of metal removal and thermal effects on machined surface by observing sparks.

Since the invention of modern machine tools, their fundamental design has not changed. This paper proposes a novel machine tool design that enables the machining of a workpiece simultaneously with multiple spindles, for drastically enhancing the machining efficiency. Using parallel linear axes, each of which is regulated in synchronization with a rotary table, 2.5-dimensional tool path machining becomes possible. In this new configuration, the risk of collision among the spindles is entirely eliminated. This paper presents a tool path generation algorithm so that the multiple spindles are simultaneously engaged in the machining of contour-parallel paths. To prove the competence of the novel multi-spindle machine tool configuration, a prototype has been designed and constructed. Finally, to evaluate the higher efficiency of this configuration, a machining test with multiple spindles is conducted.

In surface anomaly detection tasks, achieving high-precision classification between normal and anomaly images is crucial. However, conventional pre-training using datasets like ImageNet often struggles to represent subtle differences in surface conditions. We propose an anomaly detection method based on pre-training with a semi-formula-driven image dataset, designed to enhance feature representations of surface defects with subtle differences. A large number of pseudo-anomaly images are generated by adding artificial defects to normal images using random Gaussian Mixture Model (GMM) parameters. By estimating GMM parameters that represent the color, position, size, and shape of artificial defects, the CNN learns to form discriminative feature representations even for subtle differences between normal and anomaly images. Experiments on the MVTec AD benchmark demonstrate that the proposed method outperforms or matches conventional ImageNet-based pre-training. In particular, the proposed method achieved Pixel-AUROC of 94.2 in categories where ImageNet previously showed lower performance (Pixel-AUROC of 90.5). The results indicate that feature representations learned through defect parameter estimation improves the precision and generalizability of anomaly detection.

This study clarifies the inherent redundancy between Position-Dependent Geometric Errors (PDGEs) and Position-Independent Geometric Errors (PIGEs) in five-axis machine tools. We establish a mathematical model showing that PIGEs can be represented by PDGEs and that the influence of PDGEs can be represented by the coordinate transformation into the workpiece coordinate system. Cutting experiments on a hexagonal workpiece, under the compensation for the equivalent PIGEs and PDGEs, confirm that replacing PIGEs with PDGEs does not degrade the machining accuracy. Moreover, based on the redundancy between PIGEs and PDGEs, it is possible to eliminate the compensation at a certain angular position, without influencing the geometry of the finished workpiece. These insights simplify indirect measurement and error-compensation methods, offering greater flexibility for both users and manufacturers.