Deep Learning Approach to Identifying New Infrared Spectroscopic Features Produced by Accelerated Aging of Cross-linked Polyethylene Pipe

We have implemented a deep generative modeling approach using a b-variational autoencoder (b-VAE) to learn disentangled representations of the major sources of variance in infrared spectroscopy data collected on cross-linked polyethylene (PEX-a) pipe subjected to accelerated aging [1]. This has allowed us to extract detailed information on the physicochemical changes that occur during aging, degradation and failure of PEX-a pipe. In the present study, we introduce defects on the inner surface of the pipes by scoring the surface with a sharp tool and then subjecting the pipes to accelerated aging. High resolution infrared images allow us to identify a new spectroscopic feature near the apex of the defect that develops during the aging of the pipes. This feature, associated with enhanced carboxylate absorbance, resulted in a large mean square error when processed with our b-VAE model trained on over 30,000 IR spectra collected on unused, aged and failed PEX-a pipe, indicating a new signature of pipe degradation.