Feature Extraction

Datasets like FEMTO or XJTU-SY provide raw vibration data. It may be useful to extract hand-crafted features, i.e. RMS or P2P, from this vibration data to use in your network. The RulDataModule provides the option to use a custom feature extractor on each window of data.

The feature extractor can be anything that can be called as a function. It should take a feature array with the shape [num_windows, window_size, num_features] and a target array with the shape [num_windows]. The return value should be the transformed feature and target arrays. Depending on whether a window_size is supplied to the data module, the expected output shape of the feature extractor is:

window_size is None: [num_new_windows, new_window_size, features]
window_size is not None: [num_windows, features]

An example would be taking the mean of the window with lambda x: np.mean(x, axis=1). Because this function reduces the windows to a single feature vector, we set window_size to 10 to get new windows of that size:

>>> import rul_datasets
>>> import numpy as np
>>> reader = rul_datasets.FemtoReader(fd=1)
>>> dm = rul_datasets.RulDataModule(
...     reader,
...     batch_size=32,
...     feature_extractor=lambda x: np.mean(x, axis=1),
...     window_size=10
... )
>>> dm.setup()
>>> features, _ = dm.to_dataset("dev")[0]
>>> features.shape
torch.Size([2, 10])

The new features have a new window size of 10 instead of 30, the readers default. Each of the two features contains the mean of the corresponding feature in the original data window.

The number of samples will reduce by num_runs * (window_size - 1) due to the re-windowing of the data:

>>> reader = rul_datasets.FemtoReader(fd=1)
>>> dm_org = rul_datasets.RulDataModule(reader, batch_size=32)
>>> dm_org.setup()
>>> dm_extracted = rul_datasets.RulDataModule(
...     reader,
...     batch_size=32,
...     feature_extractor=lambda x: np.mean(x, axis=1),
...     window_size=10
... )
>>> dm_extracted.setup()
>>> len(dm_org.to_dataset("dev"))
3674
>>> dm_extracted.to_dataset("dev")
3656

If your feature extractor produces windows itself, you can set window_size to None. This way, no new windows are extracted. An example would be extracting multiple sub-windows from the existing windows.