DOI: 10.66106/skygay.20250207 ISSN: 3105-7500

面向人工智能的紧固件制造检验检测高质量数据集构建方法（A High-Quality Dataset Construction Method for AI-Based Fastener Manufacturing Inspection and Testing）

张骄阳 Jiaoyang Zhang, 李梦 Meng Li, 朱松 Song Zhu, 董敬 Jing Dong, 郑天翼 Tianyi Zheng
Show PDF Cite
Abstract：With the deep advancement of Industry 4.0 and intelligent manufacturing, artificial intelligence (AI) technologies, particularly deep learning, have shown tremendous potential in industrial visual inspection. As fundamental industrial components, fasteners directly affect the safety and reliability of equipment. However, the effectiveness of current AI applications in fastener defect detection is severely limited by the quality and scale of training data. This paper systematically investigates the methodology for constructing high-quality datasets tailored for AI-based manufacturing inspection and testing of fasteners. First, the core requirements of industrial scenarios for datasets are analyzed, including defect diversity, data authenticity, and annotation accuracy. Subsequently, a complete construction process is proposed, encompassing five key stages: data requirement analysis and planning, multi-source data acquisition, professional data annotation, data preprocessing and augmentation, and dataset management with version control. This paper emphasizes data generation and augmentation strategies for few-shot and hard-sample defects and introduces a data quality assessment metric system to ensure dataset reliability. Finally, a case study on "bolt thread surface defects" validates the effectiveness and practicality of the proposed method. This research aims to provide a standardized and implementable technical solution for constructing high-quality AI datasets in the fastener manufacturing industry, thereby accelerating the industrial application of AI-based quality inspection.

More from our Archive