Surface inspection of plastic products plays a crucial role in product quality—this is also the case at Reichmann & Sohn GmbH, which specializes in the development and manufacture of machines for grinding, cutting, and polishing.
Automation can simplify this process considerably. One of the biggest challenges here is reliably distinguishing actual scratch marks from similar-looking structural features, such as those found in background patterns, and from optical artifacts such as light reflections. Previous solutions have often been based on classic computer vision methods. These rely on static thresholds and reach their limits when anomalies or complex background structures are encountered.
Now, TTZ Günzburg has pursued more innovative approaches, creating a basis for new technological standards in the field of automated surface inspection at Reichmann & Sohn GmbH.
Scratches in the crosshairs: Deep learning models identify material anomalies
A specially prepared image database with a very high resolution of up to 0.2 micrometers serves as the basis for reliably distinguishing between relevant scratches and irrelevant structural elements. A new form of automation is achieved by combining precisely tailored, state-of-the-art deep learning methods that enable scratches to be correctly classified as shallow or deep and distinguished from background patterns and image noise. “After correctly detecting and segmenting the scratches and removing image noise, the system provides the corresponding coordinates for subsequent further processing of the damaged surface,” explains project manager Prof. Dr. Alexander Bartel.
At the same time, it was ensured that the results could be smoothly integrated into existing production processes. The AI system developed contributed significantly to significantly improving the quality of the existing system in the detection of surface defects. This creates an important basis for the further efficient automation of industrial manufacturing processes.
