Anomaly Detection in Automated Fibre Placement: Learning with Data Limitations

TL;DR

This work tackles AFP defect detection under limited labelled data by combining local sample extraction with a convolutional autoencoder trained on normal tows. Depth maps from an OCT sensor feed into a dense local-sample dataset that enables end-to-end anomaly detection and 1D-to-2D defect localization via blob analysis. The method achieves strong discrimination between normal and defective regions with a 16D latent space and yields bounding boxes with competitive localization performance (IoU around 0.708) without requiring defect samples for training. Its data-efficient, unsupervised approach offers practical AFP inspection that generalizes to various anomaly types and reduces reliance on extensive defect-labelled datasets.

Abstract

Conventional defect detection systems in Automated Fibre Placement (AFP) typically rely on end-to-end supervised learning, necessitating a substantial number of labelled defective samples for effective training. However, the scarcity of such labelled data poses a challenge. To overcome this limitation, we present a comprehensive framework for defect detection and localization in Automated Fibre Placement. Our approach combines unsupervised deep learning and classical computer vision algorithms, eliminating the need for labelled data or manufacturing defect samples. It efficiently detects various surface issues while requiring fewer images of composite parts for training. Our framework employs an innovative sample extraction method leveraging AFP's inherent symmetry to expand the dataset. By inputting a depth map of the fibre layup surface, we extract local samples aligned with each composite strip (tow). These samples are processed through an autoencoder, trained on normal samples for precise reconstructions, highlighting anomalies through reconstruction errors. Aggregated values form an anomaly map for insightful visualization. The framework employs blob detection on this map to locate manufacturing defects. The experimental findings reveal that despite training the autoencoder with a limited number of images, our proposed method exhibits satisfactory detection accuracy and accurately identifies defect locations. Our framework demonstrates comparable performance to existing methods, while also offering the advantage of detecting all types of anomalies without relying on an extensive labelled dataset of defects.

Figures (15)

• Figure 1: The industrial AFP setup is shown in two separate views. On the left is an overall view of the fibre placement machine (a), and on the right is a close-up shot of the robotic tool applying carbon fibre tows. The OCT sensor is visible to the upper left of the roller.
• Figure 2: Different representations of a composite part manufactured with an AFP machine. Above (a) is a 3D point cloud measured using OCT Technology. The bottom left (b) shows the depth map generated from the 3D point cloud, and a real photograph of the same composite part is shown in the bottom right (c) for comparison.
• Figure 3: An overview of the defect detection process shows the necessary steps.
• Figure 4: The centerline detection procedure contains two main steps: detecting horizontal and vertical lines (a) and estimating tow centerlines from the detected lines (b).
• Figure 5: A dataset is created from cropped sections of the depth maps, using the sliding window method. Normal samples are shown on top and abnormal samples are shown below.