NON-DESTRUCTIVE EVALUATION OF FIBER-REINFORCED POLYMER COMPOSITES USING ACTIVE ULTRASONICS FOR INLINE APPLICATIONS

Abstract

New technologies are consistently under development as tools for evaluation and quality assurance (QA) of manufactured parts in order to minimize wastage during production. These tools should ideally offer robust operation, affordability and real time feedback to ensure their rapid integration into new production systems. This thesis focuses on developing an acoustic monitoring system featuring active ultrasonic sensors paired with an artificial intelligence model to assess a fiber reinforced polymer composite material accurately in real time. The research in this thesis first examined how to map signature ultrasonic frequencies in a detected signal to more complex morphological features of a composite, namely fiber orientation in this case to ultimately build an offline QA test. It was found that transforming the ultrasonic spectrum with a continuous wavelet transformation and pairing it with artificial neural networks allowed for highly accurate fiber orientation predictions. The signature resonating frequencies related to the fibers were unaffected by orientation; however, use of neural network modelling revealed changes over time in the frequency pattern as the sound wave passed through the material that were orientation sensitive. With strong evidence on the usefulness of this frequency-based evaluation tool, at least in off-line material characterization, the subsequent step to this research was to examine composite manufacturing inline on an extrusion system. The in-line monitoring tool found that signature frequencies previously correlated to properties of the material off line would shift should the flow rate change while the process was being monitored, thus iv causing inaccuracies in the quantification of those correlated properties. Multiple causes were explored for this behavior, with melt temperature variation derived by viscous dissipation being the most likely explanation for the frequency shift. Changes to the melt temperature varied the dispersion modes of the propagating sound through the polymer melt. Lastly, improvements to the capabilities of the off-line evaluation tool were made by developing a new expression for attenuation of a propagating ultrasound signal through in a fiber-filled composite that considers the porosity and fiber content of the material. Two signature spectral behaviors were identified and analyzed in the attenuation spectrum, the first being significantly influenced by the symmetric dispersion mode and resonance frequency present due to the air and fiber content; this region was successfully modelled with the aid of genetic programming to identify mathematical terms associated with the attenuation of the heterogenous material but not being completely dissimilar to models for homogenous materials. The second region had limited influence by both the porosity and fiber content, with dampening effects being the main relation, which was modelled using traditional classifiers.