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

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.