REDUCING UNCERTAINTY, RISK AND COST OF VARIATION IN COMPOSITE NON-DIMENSIONAL ATTRIBUTES

How much are the structural integrity and performance of primary composite components affected by raw material flaws, laminate wrinkles, fiber waviness, laps, gaps, porosity, foreign object debris (FOD), fiber orientation and other non-dimensional attributes? The impact of variation in these factors is largely unknown and/or unquantified, which introduces uncertainty that must be addressed through risk-reduction measures such as overdesign and stringent manufacturing quality control. Such measures are necessary to ensure the quality, safety and performance of composite components, but they inflate the cost and increase the cycle time of producing these components. The advent of proven AI-enabled automatic in-process inspection technologies opens an opportunity to reduce this uncertainty. By digitally capturing and quantifying component features and anomalies during the fabrication process, automatic inspection generates digital images and data for an as-built digital twin of each fabricated component. Aggregated data on variation in non-dimensional attributes from multiple as-built digital twins serves as input for recently developed artificial intelligence/deep learning (AI/DL). This paper explains how applying AI/DL to captured as-built data provides composites fabricators with insights they may use to reduce uncertainty associated with non-dimensional attributes. Less overdesign and increased process tolerances may then be accommodated. The paper explores the resulting risk and cost reductions in several use cases.