Deep Generative Modeling of Infrared Images Provides Signature of Cracking in Cross-Linked Polyethylene Pipe

Area of Science:

• Materials Science
• Analytical Chemistry
• Data Science

Background:

• Hyperspectral infrared (IR) imaging provides rich chemical information but is challenging to analyze for complex materials.
• Cross-linked polyethylene (PEX-a) pipes are susceptible to aging and degradation, impacting their performance and safety.

Purpose of the Study:

• To develop a deep generative modeling approach for analyzing hyperspectral IR imaging data.
• To identify and characterize physicochemical factors of aging and degradation in PEX-a pipe.

Main Methods:

• Implementation of a β-variational autoencoder (a type of deep generative model).
• Training the model on hyperspectral IR image data of PEX-a pipe samples.
• Applying the trained model to analyze virgin, in-service, and cracked PEX-a pipe.

Main Results:

• The model successfully learned disentangled representations of physicochemical factors related to aging and degradation.
• Three distinct factors of variance were identified in the PEX-a pipe data.
• Detailed physicochemical changes during aging, degradation, and cracking were mapped to IR images.

Conclusions:

• Deep generative modeling, specifically β-variational autoencoders, significantly enhances the analysis of complex hyperspectral IR data.
• Representation learning provides valuable insights into material degradation processes.
• This method offers a powerful tool for analyzing heterogeneous samples in materials science and chemistry.