Trained 3CH VAE model from 9/21/25

Code available at https://github.com/sandialabs/pvcracks/ - This VAE model is for the three channel model, it only uses the crack, busbar and dark area masks of the Electroluminescence images. These can be extracted using MultiSolSegment (also in the pvcracks repo) - The purpose is to parametrize the crack masks and compress the information into latent vectors. - Latent space visualization can be used to cluster the images in categories according to severity of damage - Latent vectors can be used to correlate to power loss using the powerloss model in the pvcracks repo