vibration-soft-sensor

Documentation Status PyPI Testsuite

A general toolbox for the development of vibrational softsensors. You can find a lot of functionality for data preprocessing, model training and evaluation for linear and non linear methods. We have also adapt the general approach to a variety of signal types.

The main paper Stabilised Auto-Regressive Neural Networks (S-Arnns) for Data Driven Prediction of Forced Nonlinear Systems can be found on http://dx.doi.org/10.2139/ssrn.4720155.

Table of Contents

Content

The vibration-soft-sensor library provides comprehensive tools for analyzing vibration data and building soft sensors. Key components include:

Core Components

  • Model Types

    • autoreg_models: Autoregressive models including ARNN

    • recurrent_models: Recurrent neural networks for time series prediction

    • linear_methods: Transfer function and other linear methods

    • model: Base model implementations (LSTM, CNN)

    • arx: Tools for ARX modeling with sliding window approaches

Analysis Tools

  • frequency_methods: Frequency domain analysis including FDS (Fatigue Damage Spectrum)

  • visualization: Extensive plotting utilities for sensitivity analysis and model evaluation

  • eval_tools: Model evaluation and error computation functionalities

Data Handling

  • data_gen: Simulated data generation with various excitation signals (sine, sweep, white noise)

  • meas_handling: Measurement data preprocessing, including filtering

Advanced Features

  • stab_scheduler: Stability scoring and schedulers for neural networks

  • losses: Custom loss functions including PSD-based losses

  • ensemble_wrappers: Ensemble model implementations (Sync/AsyncEnsemble)

  • hyperparameter_optimization: Tools for optimizing model parameters

Sensitivity Analysis

The library offers multiple sensitivity analysis methods:

  • Gradient-based methods

  • SmoothGrad

  • Integrated gradients

  • Perturbation analysis

  • Uncertainty quantification

Installation

# Using pip
pip install softsensor

# Using poetry
poetry add softsensor

# Using uv
uv pip install softsensor

Installation for Development

# Clone the repository
git clone https://github.com/boschresearch/vibration-soft-sensor.git
cd vibration-soft-sensor

create virtual local environment

# Using conda
conda create -p ./.venv
conda activate ./.venv

# Using poetry
poetry env
poetry shell

# Using uv
uv venv .venv
.venv/bin/activate

install development dependencies

# Using pip
pip install -e .[dev]

# Using poetry
poetry install --with dev

# Using uv
uv pip install -e .[dev]

Maintainers

Tobias Westmeier: tobias.westmeier@iee.fraunhofer.de

Daniel Kreuter: danielchristopher.kreuter@de.bosch.com

License

softsensor is open-sourced under the Apache-2.0 license. See the LICENSE file for details.

For a list of other open source components included in pyLife, see the file 3rd-party-licenses.txt.