Anomaly detection for industrial parts using PatchCore

• Main Author: Kuah, Zheng Xuan
• Other Authors: Yeo Chai Kiat
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/162909
• Institution: Nanyang Technological University

The ability to detect imperfect parts is essential for components in a large-scale industrial manufacturing. The decision of an anomaly detection revolves around a binary problem. This paper will delve into a state-of-the-art method of anomaly detection known as PatchCore and its effectiveness on various datasets. Several datasets are considered along with specific domain area such as the Magnetic tiles. By extending the usage of a memory bank for pixel level patch features from an auto encoder, PatchCore can achieve high level accuracy pixel-level anomaly detection score of up to 99.6%. Looking beyond traditional computing, the model will be considered for edge computing on Internet of Things for faster inference speed.