Speaker
Description
Cross-linked polyethylene (PEX) pipes are being used increasingly for water transport and heating applications. The pipes are exceptionally reliable but sometimes unexpected failures occur. We are using infrared microscopy, accelerated aging and machine learning analysis techniques to understand the mechanisms of pipe aging, degradation and failure [1-3]. In previous studies, we implemented linear dimensionality reduction [1] and deep generative modeling [2,3] approaches to learn disentangled representations of major sources of variance in our large body of infrared spectroscopy data collected on PEX-a pipe subjected to accelerated aging conditions. In the present study, we compare two machine learning approaches – principal component analysis (PCA) and a β-variational autoencoder (β-VAE) – to identify and analyze a new spectroscopic feature that develops during accelerated aging of PEX pipes in which scoring marks of adjustable depth have been introduced on the inner surface of the pipes. This spectroscopic feature, which corresponds to localized enhanced carboxylate absorbance, was not observed in unused or in-service pipes. Both PCA and β-VAE allow the identification of the variables (either principal components or latent variables) that are responsible for the largest variance in the data sets, but the deep learning β-VAE approach has the added advantage of identifying the physical significance of each variable.
[1] M. Grossutti, J. D’Amico, J. Quintal, H. MacFarlane, A. Quirk and J.R. Dutcher, J. Phys. Chem. Lett. 13, 5787 (2022).
[2] M. Grossutti, J. D’Amico, J. Quintal, H. MacFarlane, W.C. Wareham, A. Quirk and J.R. Dutcher, ACS Appl. Mater. Interfaces 15, 22532 (2023).
[3] J. D'Amico, M. Grossutti and J.R. Dutcher, ACS Appl. Polym. Mater. 6, 534 (2024)
| Keyword-1 | Machine Learning |
|---|---|
| Keyword-2 | Infrared Spectroscopy |
| Keyword-3 | Material Characterization |