非破壊検査
Print ISSN : 0367-5866
最新号
選択された号の論文の5件中1~5を表示しています
  • 山﨑 惇史, 林 高弘, 森 直樹
    2024 年 73 巻 7 号 p. 267-273
    発行日: 2024/07/01
    公開日: 2024/07/01
    ジャーナル フリー

    This research focuses on the automatic defect detection of metal additive manufacturing components combining machine learning and the scanning laser source technique. In the experiment, an additive manufacturing component with a circular artificial defect was put on the plate on which transducers were attached. By laser scanning, the waveform generated at each laser spot was obtained through these transducers. The waveforms obtained were Fourier-transformed, and the peak values were extracted as features of machine learning. Moreover, each place was labeled to indicate whether a defect was present or not. The logistic regression model and the random forest classifier model learned these features and labels. After that, another dataset of waveforms was prepared, and the prediction of defective places was conducted by using these models. As a result, the random forest classifier model predicted the defective area precisely, but the logistic regression model generated comparatively noisy outputs. These results demonstrate the possibility of applying random forest classifier model to automatic defect detection during AM processes.

  • 神谷 大樹, 清水 鏡介, 伊藤 洋一, 大隅 歩
    2024 年 73 巻 7 号 p. 274-279
    発行日: 2024/07/01
    公開日: 2024/07/01
    ジャーナル フリー

    Non-destructive testing can take a long time when applied to structures with thin metal walls such as pipes and tanks for maintenance. This is because the structure is large, and the inspection range is vast. To solve this problem, we have developed a non-destructive testing method for thin metal plate structures using the scanning technique with an airborne ultrasound source. This method can detect defective areas from their reflections and diffractions. However, the defects are visualized by visual inspection by testers, who may overlook them due to human error. In this study, we constructed a system to support defect detection by using a generative adversarial network. As a result, we confirmed that images can be generated for defect detection using Pix2Pix.

  • 北澤 聡, 坂田 聡
    2024 年 73 巻 7 号 p. 280-285
    発行日: 2024/07/01
    公開日: 2024/07/01
    ジャーナル フリー

    To improve railway vehicle reliability, data-driven quality control has been attempted in manufacturing processes from the DX viewpoint. Railway vehicles have many welds joining structural parts, and their quality is controlled by UT to verify their durability. Real-time collection and evaluation of the digital UT data would allow in situ repair during the welding processes. The challenge here is establishing the UT technique for curved weld surfaces, such as fillet weld, which makes measurement difficult due to the contactability of the ultrasonic probes and the weld surface. Flexible array probes and FMC/TFM imaging techniques have been applied to overcome this issue. In TFM, coordinates of piezoelectric elements in an array probe are necessary to generate an inspection image. In this study, strain gauges were applied to obtain the shape of an array probe, corresponding to the element coordinates. A real-time imaging algorithm, using a prototype flexible probe with strain gauges, showed the feasibility of fast inspection of various curved surface welds in railway vehicles.

  • 中畑 和之, 一色 正晴, 井門 俊, 浅川 濯, 伊津美 隆, 大平 克己
    2024 年 73 巻 7 号 p. 286-292
    発行日: 2024/07/01
    公開日: 2024/07/01
    ジャーナル フリー

    Non-destructive Evaluation 4.0 (NDE 4.0) is a concept that aims to disseminate a cutting-edge approach integrating advanced technologies to the NDE community. One key technology in Non-destructive Evaluation 4.0 (NDE 4.0) is visualizing the internal target’s three-dimensional (3D) shape in an easy-to-understand manner. In this study, we proposed a superimposing technique of 3D ultrasonic imaging results and the realistic shape of target concrete. As the imaging method, Full-waveforms Sampling And Processing (FSAP) was applied using a matrix array probe. Here, based on beam radiation simulation, we manufactured array elements in a low-frequency ultrasonic range in order to penetrate the ultrasonic wave in deep areas of concrete. The ultrasonic imaging result was overlaid on a 3D shape captured by a Structure from Motion (SfM) technology. Furthermore, the result can be displayed with Augmented Reality (AR) technology. The format of the superimposed inspection image will be the subject of future discussion in terms of 3D data retainability and exchange, while AR overlaid interactive 3D inspection images will require high-performance computing on a device in real-time.

  • 遠藤 英樹, 山根 佑之, 佐々木 昇, 芦田 強, 森本 勉, 岡本 陽
    2024 年 73 巻 7 号 p. 293-296
    発行日: 2024/07/01
    公開日: 2024/07/01
    ジャーナル フリー

    This paper discusses efficient maintenance methods for steel structures built over 50 years ago. One of the challenges faced in the maintenance efforts of a steel mill, in which the authors are involved, is the efficient detection of fatigue cracks that occur in frequently used overhead cranes and their associated equipment. In particular, screening methods are needed for runway girders supporting the crane’s running rails, which require time-consuming inspections for fatigue cracks. Therefore, this paper considers an efficient screening method for fatigue cracks that occur under the triangular ribs of runway girders. While inspection methods using thermoelastic effect have been proposed in the past, they have limitations in detecting crack shapes and measuring crack lengths. Therefore, a new method for detecting fatigue cracks using infrared radiation is proposed. Deep learning techniques are also considered to improve the efficiency of detecting fatigue cracks from the captured thermal images. The proposed method was evaluated on the runway girders of a steel mill, demonstrating its ability to detect fatigue cracks of 10cm or more.

feedback
Top