This project focuses on deep learning-based detection of surface defects on metal industrial components. These flaws affect not only visual quality but also fatigue resistance, durability, and corrosion resistance. Some stripe-like textures closely resemble scratches, causing false alarms, while matte-like textures can conceal tiny defects and lead to missed detections. To address these issues, the project investigates a new deep learning framework that combines defect representation learning, so that subtle defect patterns can be distinguished more reliably under complex textured backgrounds.
Aim
The aim of this project is to develop a novel deep learning algorithm for accurate and robust metal surface defect detection in complex industrial scenarios. Specifically, it seeks to improve discrimination between true defects and confusing background textures on textured metal surfaces, while maintaining sensitivity to very small and subtle defects. The project also aims to overcome the practical limitation of insufficient defect samples, which often restricts the performance of supervised industrial inspection systems. By designing a new network architecture informed by deep learning, the study intends to learn the manifold distribution of samples in feature space and strengthen similarity-based understanding of defective and non-defective regions. In addition, synthetic samples will be introduced to enrich defect representation during training, improving the model’s generalisation ability and detection reliability in real manufacturing environments.
Objectives
The main objectives are to:
- analyse the visual characteristics of defects and textures;
- design a new deep neural network architecture for surface defect detection under complex texture interference;
- model the distribution of normal and defective samples in feature space to improve separability and robustness; and
- evaluate the proposed method in terms of detection accuracy, false alarm reduction, missed defect reduction, and adaptability to diverse industrial textures. These objectives support the development of a practical and scalable inspection algorithm for industrial processes.
Significance
This project addresses a critical quality control problem in modern industry processes: the reliable detection of tiny and ambiguous defects on textured metal surfaces. Improved defect detection can directly support product quality, service life, and safety by reducing the risk of performance degradation caused by surface damage. From a technical perspective, the project contributes to industrial inspection research by integrating deep learning with surface defect analysis, offering a promising alternative to traditional hand-crafted feature methods. The use of synthetic samples also provides a practical solution to the common problem of defect data scarcity in real scenarios. For industry, the proposed method may enable more automated, accurate, and scalable inspection of precision metal parts, reducing manual effort and lowering the costs associated with false rejection, missed defects, and downstream failure.
Ideal Candidate
We are looking for a self-motivated PhD candidate with excellent organisation, problem-solving and project management skills. Candidates with strong quantitative skills, including familiarity with deep learning and mathematical modelling are desired for this project. Additionally, the applicants should meet the eligibility criteria for entry into a PhD program at Curtin University.
This project is open to Domestic applicants only.
Scholarship
If you are identified as the preferred candidate for this project, you may be considered for an RTP scholarship.
Enquires and How to Apply
For enquires about this opportunity contact Professor Honglei Xu at H.Xu@curtin.edu.au
To formally apply submit an Expression of Interest to Professor Honglei Xu during the Central Scholarship round (July 1st – July 31st 2026)