Abstract
The certification of composite aerostructures traditionally follows a labor-intensive process involving extensive testing campaigns that encompass a range of environmental conditions, testing geometry, loading conditions, lay-up, and material architecture. This approach requires repeating tests from different batches to establish statistical significance and determine A-basis and B-basis design allowables, resulting in high costs and time consumption. Moreover, it limits design exploration and hinders the qualification and certification of new materials and designs. While composite failure simulation techniques have evolved over the past decade, they are often slow and overlook material variability and uncertainties. The calibration of high-fidelity simulation models is also complex. Low-fidelity models and reduced-order modeling have been used to address these issues, but their accuracy and applicability are in question. Multi-scale simulation methods have been developed to consider composite structures, yet they tend to neglect the impact of processing-induced defects on performance. Recent attempts to incorporate machine learning (ML) in the certification process have faced challenges due to the difficulty of obtaining necessary training data from high-fidelity and multi-scale simulations. In the proposed framework, material and processing uncertainties are considered by integrating stochastic process simulation with stochastic failure analysis, focusing on the effects of processing-induced defects on failure. This approach uses both low-fidelity and high-fidelity finite element (FE) simulations and multi-scale simulations to generate data for training ML models. These models serve to quantify uncertainty using a Monte Carlo approach, ultimately yielding digital design allowables. A case study with Hexcel AS4/8552 material illustrates the framework’s application, estimating design allowables while accounting for heat transfer uncertainties, residual stress variations, and volume fraction variations. These digitally estimated design allowables are subsequently validated against publicly available experimental data. This framework has the potential to revolutionize the certification process, reducing costs and time while enhancing design exploration and flexibility.
