Defect Detection is the automated identification of flaws, anomalies, or deviations from specification in products, materials, or processes, typically performed using machine vision and machine-learning models on images or sensor data. It localises and classifies defects such as cracks, scratches, contamination, dimensional errors, or assembly faults, often in real time on a production line. Modern systems combine high-resolution imaging, deep-learning object detection and anomaly detection, and feedback to robotic or process controls. Defect detection underpins quality assurance and industrial inspection, reducing waste and ensuring consistency.
Overview
- Defect detection automates a task historically performed by human inspectors: spotting flaws that render a product non-conforming. It combines imaging hardware with learned models that classify and localise defects.
- Approaches range from classical image processing (thresholding, edge detection, template matching) to deep-learning object detection and unsupervised anomaly detection that flags deviations from a learned normal distribution.
- Integrated into production lines, defect detection closes a quality loop: detected faults trigger rejection, rework, or adjustments to the manufacturing process.
Mechanisms
- Imaging: cameras, line scanners, and specialised sensors capture the product surface under controlled lighting.
- Feature extraction and classification: learned features distinguish defective from acceptable regions.
- Anomaly detection: models trained on defect-free examples identify out-of-distribution patterns without labelled defects.
- Closed-loop control: detections feed back to robotic sorting and process adjustment for continuous quality assurance.
Applications
- Surface inspection in semiconductor, automotive, and textile manufacturing.
- Weld and assembly verification in robotic production cells.
- Predictive maintenance through detection of wear and material degradation.