[4A6] Predictive models for estimating front position in resin infusion processes using ultrasonic guided wave sensors
C Calistru¹, V Tunukovic¹, E Mohseni¹, S G Pierce¹, C MacLeod¹, D Lines¹, R Vithanage¹,
I Bomphray², T Weis², G Munro³ and T O’Hare³
¹University of Strathclyde, UK
²National Manufacturing Institute Scotland (NMIS), UK
³Spirit AeroSystems, UK
Resin infusion and out-of-autoclave curing of carbon fibre composites offer an economical and sustainable alternative to autoclave processing in aerospace manufacturing. However, incomplete ply impregnation in the absence of the autoclave’s high pressures introduces porosity defects, affecting mechanical integrity and limiting adoption of these processes in safety-critical applications. In-situ monitoring of the resin flow is essential to prevent void formation and ensure composite quality. Certain propagating modes of ultrasonic guided waves (UGWs) react to the presence of fluids, providing a sensitive mechanism for resin front tracking.
In this study, three piezoelectric transducers were embedded in the upper lid of an infusion mould to generate and receive UGWs. Experimental and simulation-based dispersion analysis identified the fundamental antisymmetric mode as suitable for liquid monitoring, its amplitude varying with the proportion of the propagating path covered by resin. A test-rig designed to ensure consistent and repeatable fluid measurements correlated ultrasonic data with time-stamped resin-front positions extracted by a machine-vision algorithm. A parametric study optimised sensor spacing to balance sensitivity and coverage.
Initial statistical models relating UGW attenuation to front location yielded an average error of approximately 10%. A machine learning model reduced this error to 5%, demonstrating robust predictive performance with limited data.
In this study, three piezoelectric transducers were embedded in the upper lid of an infusion mould to generate and receive UGWs. Experimental and simulation-based dispersion analysis identified the fundamental antisymmetric mode as suitable for liquid monitoring, its amplitude varying with the proportion of the propagating path covered by resin. A test-rig designed to ensure consistent and repeatable fluid measurements correlated ultrasonic data with time-stamped resin-front positions extracted by a machine-vision algorithm. A parametric study optimised sensor spacing to balance sensitivity and coverage.
Initial statistical models relating UGW attenuation to front location yielded an average error of approximately 10%. A machine learning model reduced this error to 5%, demonstrating robust predictive performance with limited data.
