Surface Defect Detection for Precision Manufacturing

---

At Neatco, we aim to explore the use of artificial intelligence (AI) for detecting and tracking surface defects on aluminium panels, supporting our commitment to waste reduction and the circular economy. Surface defects can compromise the quality of painted panels, potentially leading to wasted materials through rework or scrapping. Often, manual inspection methods involve a degree of subjectivity and can lead to inconsistencies in defect detection. In addition, traditional machine vision techniques are highly susceptible to changes in lighting condition, object placement or fixturing, and image sensor placement.  

Maintaining consistent lighting across large objects presents a significant challenge for machine vision fixturing. The objective of this project was to develop and evaluate an artificial intelligence based neural network that would provide consistent, automated detection of surface defects, ensuring precise identification of their location and size while reducing variability in the inspection process.  

Neatco’s engineering team leveraged advanced computer vision techniques to design and train a neural network capable of detecting surface defects on aluminium panels. The system was trained using a small dataset of images, enabling the network to recognize surface defects such as scratches under different lighting conditions and with varying surface textures. Key elements of the solution included:  

• Scratch Detection: The neural network detects and classifies scratches, identifying their location on aluminium panels and measuring their size. This data is output in real-time to support quality control and decision-making processes.  
• Accuracy and Consistency: The system delivers highly accurate and consistent results, eliminating the subjectivity inherent in manual inspection. It is designed to detect even subtle scratches that may be missed by the human eye.  
• Real-Time Feedback: When scratches are detected, the system keeps track of and reports its position and size, allowing for a quick assessment of part quality.  
• Scalability: The solution works with varying plate sizes and surface finishes, making it suitable for a wide range of production environments.  

By automating surface defect detection, the neural network helps ensure that every panel meets quality specifications and promotes a more sustainable production practices that minimizes waste at the source. 

The AI-powered surface defect detection system is currently in the refinement and integration phase of the project. Early results have been highly encouraging, with the neural network successfully detecting a wide range of scratch types and sizes across aluminium plates of various finishes.