model

This module contains the necessary neural networks to build a RUL estimator.

In general, a RUL estimator network consists of two parts: a feature extractor and a regression head. The feature extractor is a network that transforms the input feature windows into a latent feature vector. The regression head maps these latent features to a scalar RUL value. The feature extractors can be found in the cnn and rnn modules. The regression head in the head module.

Some domain adaption approaches use a domain discriminator. The networks in the head module can be used to construct them, too.

Examples:

>>> import torch
>>> from rul_adapt import model
>>> feature_extractor = model.CnnExtractor(14,[32, 16],30,fc_units=8)
>>> regressor = model.FullyConnectedHead(8, [4, 1])
>>> latent_features = feature_extractor(torch.randn(10, 14, 30))
>>> latent_features.shape
torch.Size([10, 8])
>>> rul = regressor(latent_features)
>>> rul.shape
torch.Size([10, 1])