Simulation code and generated datasets for "Physics-Guided Sparse Domain Adaptation for Rotor Fault-Severity Estimation and Remaining Useful Life Prognostics"

Published: 1 September 2026| Version 1 | DOI: 10.17632/fzst734cks.1
Contributor:
Ravi Kant

Description

This dataset provides the finite-element (FE) rotor simulation code and the generated datasets supporting the article "Physics-Guided Sparse Domain Adaptation for Rotor Fault-Severity Estimation and Remaining Useful Life Prognostics." The study reframes rotor fault diagnosis as continuous fault-severity regression and extends it to remaining useful life (RUL) prognostics. A calibrated Euler–Bernoulli FE rotor model generates labelled vibration signals across a continuum of unbalance and misalignment severities (source domain); a perturbed, high-fidelity instance of the model serves as a controlled digital-twin target domain that emulates a simulation-to-reality distribution shift. A residual encoder shared by a severity-regression head and a gradient-reversal domain discriminator is aligned by Maximum Mean Discrepancy (MMD) and regularised by sparse penalties (L1 and weight-entropy), and evaluated under a leakage-free protocol. Contents: Simulation and analysis code: the FE rotor model, dataset generator, system-identification routine, training/evaluation pipeline (domain-adaptation baselines and sparsity), domain-shift study, RUL prognostics, and figure-generation scripts. Generated datasets and pre-computed result files (Python pickle format) so that all tables and figures can be regenerated. A data-and-code description document and reproduction instructions. All data are simulation-generated; no third-party, proprietary, or measured datasets are used. The pipeline runs on a standard CPU (Python 3.11). Full reproduction steps and dependencies are provided in the included README and requirements files.

Files

Categories

Machine Learning, Risk Assessment in Prognosis, Rotor Dynamics, Vibration Condition Monitoring

Licence