Data augmentation-aided fatigue delamination shape prognostics in composites

Abstract

Delamination will inevitably occur during the fatigue process in composites, and its shape contains substantial information like the area and center, which is useful for prognostics. However, developing a deep learning-based shape prognostic model typically requires a comprehensive dataset of delamination images, which is difficult to obtain in practical scenarios. To deal with the limited data availability, this paper proposes a novel data augmentation (DA)-aided shape prognostic solution. First, the delamination shape is discretized into certain key lengths, which are then processed through data augmentation to provide synthetic data. Next, a recursive prognostic model is built with both original and synthetic data and then used to project future shapes with one single observed image only. The proposed framework is demonstrated by experimental tests of fatigue delamination growth in composites with ultrasonics C-scan monitoring.