Anomaly Detection for Industrial Applications, Its Challenges, Solutions, and Future Directions: A Review

TL;DR

This survey examines vision-based Industrial Anomaly Detection (IAD) from 2019 onward, detailing data acquisition, preprocessing, learning mechanisms, evaluation, and datasets. It analyzes a spectrum of learning paradigms—from supervised to unsupervised and semi-supervised—along with advanced techniques like few-shot, zero-shot, and teacher-student models, and discusses practical challenges such as real-time constraints and data quality. The authors synthesize major datasets (e.g., MVTec AD, Real-IAD, VisA) and articulate how dataset limitations shape model performance, highlighting Real-IAD as a growing benchmark. They conclude with future directions including explainable AI, large vision-language models, and broader, more diverse datasets to accelerate real-world adoption of IAD systems in industry.

Abstract

Anomaly detection from images captured using camera sensors is one of the mainstream applications at the industrial level. Particularly, it maintains the quality and optimizes the efficiency in production processes across diverse industrial tasks, including advanced manufacturing and aerospace engineering. Traditional anomaly detection workflow is based on a manual inspection by human operators, which is a tedious task. Advances in intelligent automated inspection systems have revolutionized the Industrial Anomaly Detection (IAD) process. Recent vision-based approaches can automatically extract, process, and interpret features using computer vision and align with the goals of automation in industrial operations. In light of the shift in inspection methodologies, this survey reviews studies published since 2019, with a specific focus on vision-based anomaly detection. The components of an IAD pipeline that are overlooked in existing surveys are presented, including areas related to data acquisition, preprocessing, learning mechanisms, and evaluation. In addition to the collected publications, several scientific and industry-related challenges and their perspective solutions are highlighted. Popular and relevant industrial datasets are also summarized, providing further insight into inspection applications. Finally, future directions of vision-based IAD are discussed, offering researchers insight into the state-of-the-art of industrial inspection.

Figures (12)

• Figure 1: Automated inspection paradigm in a general manufacturing setting. Showcases the data collection, decision making, and quality control process.
• Figure 2: An overview of a supervised anomaly detection framework that learns three disentangled types of abnormalities ding2022catching. Image samples are first fed into a neural network, which extracts feature maps representing key characteristics of the input. Features are then passed to a classifier that assigns labels (e.g., normal or defective) or identifies specific abnormalities based on the feature patterns.
• Figure 3: Diagram of a self-supervised model's flow, from image input to the classification of steel surface defects zabin2023contrastive.
• Figure 4: Semi-supervised defect classification model, outlining the modules and the propagation of the labeled and unlabeled samples in the framework shi2024efficient.
• Figure 5: IAD-related publication trends from 2019 to June 2024 and across scientific databases.