Structural Nonlinearities for a Single Degree of Freedom System-MATLAB I

This repository contains part of a large collection of synthetic displacement time-series datasets developed for the study “Data-Driven Classification of Structural Nonlinearities Using Interpretable Deep Learning on Time Series.”The datasets represent the dynamic response of single-degree-of-freedom (SDOF) systems exhibiting a variety of nonlinear behaviors, including cubic stiffness, Coulomb friction, clearance effects, and quadratic damping. Each sample consists of a 100,000-point univariate displacement time series, generated by numerically solving the nonlinear equations of motion under logarithmic sweep-sine excitation. The datasets were produced using both MATLAB (ode45) and Julia solvers, with systematic variation in forcing amplitudes, sweep rates, damping and stiffness parameters, and additive Gaussian noise.The suite includes both binary classification datasets (2 classes) and multi-class datasets (6 classes), enabling rigorous benchmarking of time-series classification and interpretability methods. Reason for Multiple Zenodo Deposits The complete dataset collection exceeds Zenodo’s maximum file-size limit for a single upload.To ensure full public accessibility, the benchmark suite has been divided into three coordinated Zenodo records, with each record hosting a separate subset of the datasets.Together, the three deposits constitute the full dataset release supporting the paper. Dataset Overview This deposit contains one subset of the following datasets (complete suite shown for reference): LNM1, LNJ1 – Binary datasets (linear vs. cubic stiffness) DM1, DJ1 – Six-class datasets (Duffing-type nonlinearities) DJ19 – Six-class dataset with 19 distinct forcing amplitudes ONM19, OWM19, ONJ19, OWJ19 – Reproductions and extensions of Ondra et al. (2017), with/without added noise These datasets differ in complexity, solver type, and noise levels, enabling robust evaluation of classification accuracy, generalization, and post-hoc interpretability performance. File Formats CSV (.csv) — displacement time series stored as rows or columns MATLAB (.mat) — MATLAB-compatible structured arrays for efficient loading Every sample includes: 100,000-point displacement signal Integer-encoded class label Metadata: solver type, forcing parameters, SNR, dataset identifier How to Cite Please cite the associated research: Abusalameh, B., Wei, J., & Mengaldo, G. (2025). Data-Driven Classification of Structural Nonlinearities Using Interpretable Deep Learning on Time Series. and cite this Zenodo record using its DOI.