LayerWatch – A Benchmark for Layer-wise Multi-Class Visual Defect Detection in Fused Filament Fabrication

Automatically detecting and classifying defects in Fused Filament Fabrication (FFF) reduces material waste, machine time, and manual inspection effort, which makes it an important task in modern manufacturing. Typically, this task is reduced to manual visual inspection or automated anomaly detection. However, automating the detection and classification of visual defects in FFF remains challenging.
This is because defects can emerge gradually during printing, only becoming visible several layers after their physical occurrence. Also, they can vary substantially depending on the part geometry, material and recording conditions. Despite growing interest in vision-based monitoring, there is still a lack of benchmark datasets that capture the temporal, multi-class, and distribution-shifted nature of real FFF processes. Hence in this work, we introduce a novel benchmark for visual defect detection and classification in FFF. Therefore, we first formalize the problem of visual defect detection and classification in FFF and derive requirements that a benchmark must satisfy to capture the complexity of the problem. Based on the requirements, we record a benchmark dataset and evaluate it using state-of-the-art object detection models. We report their performance as baselines and publish dataset and code online at https://github.com/SEilermann/LayerWatch.