Journal of the Japan Society for Precision Engineering
Online ISSN : 1882-675X
Print ISSN : 0912-0289
ISSN-L : 0912-0289
Volume 87 , Issue 1
Showing 1-32 articles out of 32 articles from the selected issue
Address by the President
Special Issue on Precision Engineering Develops a Future Society
Lecture
My Experience in Precision Engineering
Gravure & Interview
Student Forum
Introduction of Laboratories
 
Paper
  • Hiroaki AIZAWA, Kyosuke KOMOTO, Kunihito KATO
    2021 Volume 87 Issue 1 Pages 61-70
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Anomaly detection is a challenging and fundamental issue in computer vision tasks. In recent years, Generative Adversarial Networks (GAN) and Bi-directional GAN based anomaly detection methods have achieved remarkable results. However, the performance of these approaches for anomaly detection depends on the ability to reconstruct a given normal image and to predict its latent variables. Therefore, we design a novel Bi-directional GAN-based anomaly detection model to improve these abilities. Especially, in order to reconstruct the image, we introduce the consistency loss for ensuring mutual mappings in both image and latent space. Moreover, we propose introducing the projection discriminator as an alternative of concatenating discriminator in order to perform efficient conditioning in the Bi-directional GAN model. In experiments, we evaluate the effectiveness of our model in MNIST and MVTec Metal Nut. Our experiment showed that our model allows us to detect various real anomalies such as bent, scratch, color, and flip, and outperforms the conventional ones.

    Download PDF (3870K)
  • Tadafumi NISHIMURA, Trong Huy PHAN, Kazuma YAMAMOTO, Makoto MASUDA
    2021 Volume 87 Issue 1 Pages 71-77
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Visual object detection (cars, pedestrians, etc.) from vehicle mounted camera faces great difficulty when the target objects are mutually occluded or partially hidden behind background objects. In this paper, in order to realize a occlusion-robust object detector, we propose to use 1) MDCN1), a SSD2)-based detector which utilizes contextual information surrounding the target objects, together with 2) Soft-NMS3), a bounding box unification technique catering to close-by objects. Experiments with the publicly available KITTI4) data acquired from vehicle mounted camera proved the effectiveness of the proposed method.

    Download PDF (3195K)
  • Ryo Takahashi, Yuji Sato, Junko Furuyama, Megumi Yamaoka, Masamoto Tan ...
    2021 Volume 87 Issue 1 Pages 78-82
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    In order to train a classifier for self checkout system in convenience store at low cost, a few-shot domain adaptation problem has to be solved. Since the system treats a classifier for large number of classes, conventional method of few-shot domain adaptation should be extended for many classes. This paper propose to exploit meta-class information by executing the adaptation on the normal-class level and the meta-class level simultaneously. The proposed method are shown to be effective for improving adaptation accuracy of a classifier for many classes. The results of our ablation study implies that i) the meta-class should be decided by using k-means clustering method rather than clustering manually, and that ii) the ratio between the number of normal-class and the number of meta-class should be fixed.

    Download PDF (2513K)
  • Yoshihiro FUKUHARA, Takahiro ITAZURI, Hirokatsu KATAOKA, Shigeo MORISH ...
    2021 Volume 87 Issue 1 Pages 83-91
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    In this paper, we address the open question: “What do adversarially robust models look at?” Recently, it has been reported in many works that there exists the trade-off between standard accuracy and adversarial robustness. According to prior works, this trade-off is rooted in the fact that adversarially robust and standard accurate models might depend on very different sets of features. However, it has not been well studied what kind of difference actually exists. In this paper, we analyze this difference through various experiments visually and quantitatively. Experimental results show that adversarially robust models look at things at a larger scale than standard models and pay less attention to fine textures. Furthermore, although it has been claimed that adversarially robust features are not compatible with standard accuracy, there is even a positive effect by using them as pre-trained models particularly in low resolution datasets.

    Download PDF (4367K)
  • Munetaka MINOGUCHI, Kai WATABE, Ryosuke YAMADA, Hirokatsu KATAOKA, Aki ...
    2021 Volume 87 Issue 1 Pages 92-98
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    In order to improve the accuracy of fashion style classification, we study effective learning methods for both pre-learning and transfer learning. In the pre-training, 16 cities classification learning is performed by Fashion Culture DataBase v2 which is a large-scale fashion database. In transfer learning, augmentation is performed by replacing the background region of the learning image with the Places365 data base. The experimental results show that the proposed prior learning is effective when the domain of the transfer learning destination data is close. In addition, the proposed transfer learning is more accurate than the baseline by applying background replacement to the learning images.

    Download PDF (2651K)
  • Saki IWATA, Yasutomo KAWANISHI, Daisuke DEGUCHI, Ichiro IDE, Hiroshi M ...
    2021 Volume 87 Issue 1 Pages 99-106
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Elderly monitoring systems are gaining attention in the modern aging society. For the purpose, Far-InfraRed (FIR) sensors are often used, because they can avoid privacy concerns and are robust to environmental lightings. The authors have previously proposed several methods for human skeleton estimation from an extremely low-resolution FIR image sequence whose resolution is 16 × 16 pixels. For more accurate estimation, this paper proposes a method that is robust to variations of human positions and actions in the FIR sequences. Specifically, to extract features robust to the human positions from the images by using a Convolutional Neural Network (CNN), a global max-pooling layer is inserted into the last layer instead of multiple pooling layers which are not suitable for low-resolution inputs. Also, a network with two branches is introduced that focuses on capturing spatial and temporal information respectively. Moreover, the network has a weighted sum mechanism of their outputs, which depends on the human actions. For evaluation, a dataset was created by capturing action sequences of a human at various positions in the FIR images. Through an experiment, we confirmed that the human motion can be smoothly estimated and that the estimation accuracy is improved by the proposed method.

    Download PDF (3296K)
  • Kodai NAKASHIMA, Yutaka SATOH, Hirokatsu KATAOKA
    2021 Volume 87 Issue 1 Pages 107-113
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Datasets play an important role in determining the features that deep neural networks can acquire, but they can also contain unintended biases when constructing datasets. The BDD100K dataset, famous for its semantic segmentation task, was collected to include traffic scenes for multiple weather conditions. However, due to differences in frequency of occurrence, there is a bias in the number of data for each weather condition. Therefore, the segmentation network trained by BDD100K has poor recognition performance in some weather conditions. Semantic segmentation is an urgent issue because it is expected to be applied to traffic scene recognition systems. In this paper, we aim to improve the performance of semantic segmentation by designing a method that generates images of desired weather conditions and uses them for data augmentation. In our experiments, we first show that the image generation method we have developed produces images of a quality that can be used for data augmentation. Next, we examine the effect of data augmentation on the semantic segmentation task. As a result, compared to baseline, the mean intersection over union (mIoU) improved by about 15% in wet weather, about 9% at night, and about 7% overall.

    Download PDF (2604K)
  • Eri YAMAGUCHI, Hiroshi HIGUCHI, Atsushi YAMASHITA, Hajime ASAMA
    2021 Volume 87 Issue 1 Pages 114-119
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    This paper proposes a method to improve the accuracy of SLAM in environments with glass by combining LRF and a polarization camera to detect a wide range of glass. Glass confidence of each point detected by the LRF is calculated using the polarization from the polarization camera. The polarization camera acquires the intensity of light passing through the polarizers in four directions. The degree of polarization is calculated from the light intensity of four directions, and is used as the glass confidence. Every time the robot moves, the map and the glass confidence are updated. The robot’s position is estimated using the generated map. Accuracy of the map is improved by considering the glass probability. Improved accuracy of the map also improves the accuracy of self-localization. The accuracy of glass detection was confirmed by the experimental results. The AUC of glass detection was 0.942. As a result, the proposed method was able to produce a more accurate map in the glass area.

    Download PDF (2604K)
  • Shunsuke NAKATSUKA, Kunihito KATO
    2021 Volume 87 Issue 1 Pages 120-126
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    In the case of industrial products, visual inspection is essential to improve the quality of products. In recent years, automation with neural networks has been considered. However, the conventional model discriminating good or defective products requires a large amount of good and defective samples for learning. In fact, it is difficult to ensure a large amount of defective samples. Therefore, in this research, anomaly (=defective samples) are detected by modeling the normal distribution and its complement from only a large amount of good products. Since the defect is a part of the image and its size varies, we propose the structure of Multi-scale Patch Discriminator in this paper.

    Download PDF (4650K)
  • Koki JIMBO, Toshitake TATENO
    2021 Volume 87 Issue 1 Pages 127-133
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Additive manufacturing (AM) technologies makes it possible to fabricate complex and internal structures that are difficult to fabricate with conventional processing methods. The material extrusion method, which is one type of AM, can fabricate composite material structures such as carbon fiber-reinforced plastics (CFRP). However, there are some problems to analyze a structure made with CFRP. First, the conventional FEM requires a lot of elements for approximation of composite materials, which leads to a long calculation time. Second, it is difficult to identify the model's parameters. This study proposes a method that can efficiently analyze the stiffness of a structure fabricated by AM. In conventional FEM, shape models were divided into small elements and the analysis was conducted by an element. In the present method, the mesh element was approximated to infill pattern structures fabricated by material extrusion AM. Simulation software using this analysis method was implemented, and the results were evaluated. Tensile and bending simulation with proposed method and conventional FEM software were performed and mechanical tests on specimens fabricated with CFRP were conducted. From the results, the reliability of proposed method was confirmed. Furthermore, a robot hand with a compliant mechanism was designed and its deformation was well simulated.

    Download PDF (2856K)
  • Kensuke TOBITANI, Aya SHIRAIWA, Kenji KATAHIRA, Noriko NAGATA, Kunio N ...
    2021 Volume 87 Issue 1 Pages 134-139
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    The platform of personal fabrication is developed by technologies such as 3D printing. However, most people do not have enough professional knowledge or skills to design these products. A promising approach relates subjective impressions to physical parameters to support intuitive design. This study aimed to build a model that predicts a “high-class feeling,” which is a product value, using physical features. The present study constructed a model that relates the subjective “high-class feeling” of products to physical features. Using a cosmetic compact as an example, a comprehensive high-class feeling and its 5 sub-factors were rated by 20 participants, and regression models were built to estimate those ratings based on physical parameters. This study suggests that a comprehensive high-class feeling is well explained by “elegance” and “luxuriance” and better estimated indirectly via these sub-factors than directly from physical features.

    Download PDF (4132K)
  • - The Effects of Ultrasonic Agitation -
    Unkai SATO, Hideki KAWAKUBO
    2021 Volume 87 Issue 1 Pages 140-145
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    This paper discusses etching performance and surface roughening performance of electrolyzed oxidizing water (hereinafter it is referred as EO water) on the rolled material of oxygen free copper with ultrasonic. Firstly, etching experiment against the surface as the rolled material of oxygen free copper, using Na2SO4-EO water, NaCl-EO water and chemicals, and with ultrasonic, was carried out. The results showed the ultrasonic agitation is effective for EO water compared with chemicals. Next, by the observation using SEM image, the effects of ultrasonic agitation on the surface micro shape was clarified. The following was clarified from the experimental results. About non-heat treated surface, when using NaCl-EO water, the surface becomes roughest. About heat treated surface, when using Na2SO4-EO water compared with H2SO4 solution, there is not occurrence of alien substance and the surface becomes smooth. Thirdly, in order to elucidate the effects of ultrasonic agitation, experiment on characteristics change of EO water with using ultrasonic, and measurement of oxygen bubble was carried out. By this study, we received the suggestion that EO water could be applied to the surface treatment of oxygen free copper.

    Download PDF (2492K)
  • - Chatter Vibration Suppression in Cutting Workpiece with Low Stiffness -
    Yuta YOSHIDA, Masahiro TAKANO, Hiroyasu MIYAKAWA, Kenichi HIROSAKI
    2021 Volume 87 Issue 1 Pages 146-150
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    Cutting thin cylindrical workpiece with low stiffness causes chatter vibration. The vibration increases machined surface roughness and cutting tool wear. This study aims to stabilize chatter vibration arising in turning workpiece with low stiffness. For the purpose of stabilizing vibration, the simple attenuation mechanism of bringing a damping beam into contact with workpiece is proposed. By this method, following results are obtained. Workpiece dampening could be improved using a damping beam with spring constant that maximizes strain energy of beam and modal damping ratio of vibration mode. Damping device could cut workpiece without chatter vibration compared with the case of not using it.

    Download PDF (2141K)
  • Masao KOHZAKI, Wataru YAMATO
    2021 Volume 87 Issue 1 Pages 151-156
    Published: January 05, 2021
    Released: January 05, 2021
    JOURNAL FREE ACCESS

    The TiB2-MoS2 composite films had been produced by DC magnetron sputtering for developing new lubricious films used at high temperatures. The composite films had excellent frictional properties with a friction coefficient of about 0.07 at R.T. even in dry conditions. Furthermore, the friction coefficients of the composite films became as low as 0.01 at 200 ˚C, although those of TiN-MoS2 composite films were above 0.2 at high temperatures. However, the frictional properties of the TiB2-MoS2 composite films deteriorated due to desorption of B by heating above 500 ˚C. Therefore, it was required to improve the frictional properties at above 500 ˚C for applying them to sliding parts used at high temperatures. In this study, B was excessively added to the composite films in order to prevent the desorption of B in heat treatments of 500 ˚C. The TiB4.4-MoS2 composite films showed a low friction coefficient of 0.03 at 200 ˚C and the excellent wear properties were maintained after heating at 500 ˚C. The formation and melting of boron oxide on the sliding surface was thought to be a reason for the low friction coefficients even after heating at 500 ˚C.

    Download PDF (4177K)
feedback
Top