Date of Award
Summer 2022
Document Type
Open Access Dissertation
Department
Mechanical Engineering
First Advisor
Ramy Harik
Abstract
The advanced manufacturing capabilities provided through the automated fiber placement (AFP) system has allowed for faster layup time and more consistent production across a number of different geometries. This contributes to the modern production of large composite structures and the widespread adaptation of composites in industry in general and aerospace in particular. However, the automation introduced in this process increases the difficulty of quality assurance efforts. Industry available tools for predicting layup quality are either limited in scope, or have extremely high computational overhead. With the advent of automated inspection systems, direct capture of semantic inspection data, and therefore complete quality data, becomes available. It is therefore the aim of this document to explore and develop a technique to combine semantic inspection data and incomplete but fast physical modeling tool into a comprehensive hybridized model for predicting and optimizing AFP layup quality.
To accomplish this, a novel parameterization of Gaussian Process Regression is developed such that nominal behavior is dictated through theory and analytic models, with latent variables being accounted for in the stochastic aspect of the model. Coupled with a unique clustering approach for data representation, it is the aim of this model to improve on the current state of the art in quality prediction as well as provide a direct path to process parameter optimization.
Rights
© 2022, Christopher M. Sacco
Recommended Citation
Sacco, C. M.(2022). Hybrid Theory-Machine Learning Methods for the Prediction of AFP Layup Quality. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/6939