Failure load prediction of different types of tensile carbon-fiber reinforced polymer specimens and vessels

Abstract

A total of 30 tensile carbon‑fiber‑reinforced polymer (CFRP) specimens and five type‑IV CFRP pressure vessels are tested using acoustic emission (AE) monitoring. Building on previously established approaches, Felicity, Shelby, and signal‑derived ratios are extracted from the AE data and used as input features for a neural network trained to predict failure loads. The ratio trends were found to be broadly consistent across the different test objects. For the tensile specimens with varying layup structures, the individually predicted failure loads show an average error of 4 kN (≈8% of failure load). Approximate burst pressures of the vessels can also be predicted using a model trained only on tensile‑sample data, albeit with larger uncertainties. Incorporating vessel data into the training set improves prediction accuracy. The study identifies a strong dependence on the chosen loading scheme, which represents a key limitation of the method and should be addressed in future work.