Michael Thompson Michael Thompson Data-driven advances in manufacturing for batch polymer processing using multivariate nondestructive monitoring

Abstract

Incorporating advanced manufacturing philosophies in practice relies on efficient strategies that can use new available sensor technologies to improve quality monitoring and process understanding. One new technology is nonlinear ultrasonics, which is a multivariate nondestructive method for the characterization of produced plastic parts. Two approaches are proposed to integrate captured data for in-line quality classification and online monitoring, providing a cost-effective alternative to destructive testing. Cluster identification is evaluated with a combination of principal component analysis (PCA) and a soft class analogy method to consider products with differing quality based on information contained in the multivariate ultrasonic signal. In the second approach, a state-space dynamic model using subspace identification is applied to historical process data and correlated with the ultrasonic-based quality data for quality prediction, and an online visualization tool was proposed in combination with a nonparametric evaluation. Results were validated with experimental data from a polyethylene rotational molding process.