This is the official code accompanying the paper Bayesian Deep Learning for Remaining Useful Life Estimation via Stein Variational Gradient Descent.
In this work, we explore the use of Stein variational gradient descent (SVGD) for training Bayesian deep learning models for remaining useful life (RUL) estimation. In particular, we investigate whether Bayesian deep learning models trained via SVGD can outperform in terms of convergence speed and predictive performance both the same models trained via Bayes by Backprop (BBB), which is the de-facto standard for training large scale Bayesian neural networks, and their frequentist counterparts trained via backpropagation (BP). For the experimental evaluation, we use the popular Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset.
First of all, install Miniconda.
Clone or download and extract the repository, navigate to <path-to-repository>, open a terminal and run:
conda env create -f environment.yml
Project dependencies (pinned to a specific version to reduce compatibility and reproducibility issues)
will be installed in a Conda virtual environment named bdl-rul-svgd.
To activate it, run:
conda activate bdl-rul-svgd
To deactivate it, run:
conda deactivate
To permanently delete it, run:
conda remove -n bdl-rul-svgd --all -y
To train one of the available deep learning model (d3 or c2p2) on a subset of the C-MAPSS dataset
(fd001, fd002, fd003 or fd004) using one of the available training algorithms (bp, bbb or svgd),
open a terminal and run:
conda activate bdl-rul-svgd
python train_<algorithm>.py <model> <subset>
Results (logs, metrics, checkpoints, etc.) can be found in the auto-generated experiments directory.
The experiments were run on an Ubuntu 20.04.5 LTS machine with an Intel i7-10875H CPU with 8 cores @ 2.30 GHz, 32 GB RAM and an NVIDIA GeForce RTX 3070 GPU @ 8 GB with CUDA Toolkit 11.3.1. To reproduce the experimental results, open a terminal and run:
conda activate bdl-rul-svgd
python run_experiments.py
@article{dellalibera2024bayesian,
author = {Luca Della Libera and Jacopo Andreoli and Davide Dalle Pezze and Mirco Ravanelli and Gian Antonio Susto},
title = {Bayesian Deep Learning for Remaining Useful Life Estimation via {S}tein Variational Gradient Descent},
journal = {arXiv preprint arXiv:2402.01098},
year = {2024},
url = {https://arxiv.org/abs/2402.01098},
}