Surface Defects Detection of Cylindrical High-Precision Industrial Parts Based on Deep Learning Algorithms: A Review

• Published in: Operations Research Forum Vol. 5; no. 3; p. 58
• Main Authors: Wei, Li, Solihin, Mahmud Iwan, Saruchi, Sarah ‘Atifah, Astuti, Winda, Hong, Lim Wei, Kit, Ang Chun
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2024; Springer Nature B.V
• Subjects: Aerospace industry; Algorithms; Applications of Mathematics; Automation; Business and Management; Cameras; Catastrophic wear; Computer vision; Corrosion; Critical components; Data acquisition; Deep learning; Defects; Design; Image acquisition; Innovation and Infrastructure; Machine learning; Manufacturing; Math Applications in Computer Science; Mathematical and Computational Engineering; Neural networks; Operations Research/Decision Theory; Optimization; Performance degradation; Quality control; Review; State-of-the-art reviews; Surface defects; Topical Collection on Math for SDG 9 - Industry; Uniqueness; Deep learning; Computer vision; High-precision cylindrical parts; Image processing; Optical illumination; Anomaly detection; Defect detection
• ISSN: 2662-2556; 2662-2556
• DOI: 10.1007/s43069-024-00337-5

High-precision cylindrical parts are critical components across various industries including aerospace, automotive, and manufacturing. Since these parts play a pivotal role in the performance and safety of the systems they are integrated into, they are often subject to stringent quality control measures. Defects on the interior and exterior wall surfaces of these cylindrical parts can severely undermine their function, leading to degraded performance, increased wear, and even catastrophic failures in extreme cases. This article aims to comprehensively summarize the task definition, challenges, mainstream methods, public datasets, evaluation metrics, and other aspects of surface defect detection for high-precision cylindrical parts, in order to help researchers quickly grasp this field. Specifically, the background and characteristics of industrial defect detection are first introduced. Owing to the unique geometric features of cylindrical part surfaces, algorithms and equipment for image data acquisition used in surface defect detection are elaborated in detail. This article presents an extensive overview of state-of-the-art surface defect detection techniques designed for high-precision cylindrical components, all rooted in deep learning. The methods are systematically classified into three main categories: fully supervised, unsupervised, and alternative approaches, based on their data labeling strategies. Additionally, the paper conducts a comprehensive analysis within each category, shedding light on their unique strengths, limitations, and practical use cases. Concluding the discussion, the paper provides insights into future development trends and potential research directions in this field that will lead to manufacturing innovation.