Sensor-corruption

Corruption Framework

A Corruption framework was built to corrupt the raw data. There are three components to the framework: Type, Granularity, and Severity.

Corruption Type

There are various different corruption types we applied.

• Stochastic: Adds random noise to the signal window, simulating stochastic sensor noise.
• Dropout: Randomly zeros out a fraction of timesteps in the signal window, simulating intermittent sensor failure or signal loss.
• Bias: Adds a constant offset to the signal window, simulating miscalibration or aging.
• Gain: Scales the signal by a constant multiplier, simulating sensor sensitivity degradation or amplification.
• Drift: Adds a linearly growing offset over the window, simulating gradual sensor drift over time.
• Resolution: Reduces the number of decimal places in the signal, simulating loss of sensor precision.

Granularity

Corruption can be applied at different levels of granularity by specifying which channels are affected. The dataset contains 6 channels across 2 sensors. Body accelerometer (x, y, z) and gyroscope (x, y, z). This allows corruption to be applied to an entire sensor modality, a single axis within a modality, or any combination, enabling targeted investigation.

Severity

Each corruption type has an associated severity parameter that controls the degree of corruption applied. The meaning of severity is type-sepcific.

• Stochastic: Corresponds to the fraction of the channel's standard deviation used as the noise scale.
  Typical values: (0.25, 0.5, 1.0, 1.25)
• Dropout: Corresponds to the fraction of timesteps zeroed out in the window. Typical value: (0.1, 0.25, 0.5, 0.75)
• Bias: Corresponds to a fraction of the channel's standard deviation added as a constant offset.
  Typical Values: (0.25, 0.5, 1.0, 1.5)
• Gain: Corresponds to a constant multiplier applied to the signal.
  Typical values: (0.5, 0.75, 1.25, 2.0)
• Drift: Corresponds to a multiplier of the channel's standard deviation by the end of the signal window.
  Typical values: (1, 2, 3, 4)
• Resolution: Corresponds to number of decimal places retained.
  Typical values: (3, 4, 5, 6)

Creating a python environment

First create a virtual python environment

$ python3 -m venv .venv

Activate environment

$ source .venv/bin/activate

Install required libraries

(venv) $ pip install -r requirements.txt

Run files

(venv) $ python src/data.py

Deactivate

(venv) $ deactivate