Dataset for: Residual Network-Based Damage Detection Using Data Augmentation and t-SNE Feature Analysis

This dataset contains vibration time-series for structural health monitoring (SHM) of a long-span steel truss railway bridge (Nam O Bridge, Da Nang, Vietnam). Data were generated from a calibrated finite-element method (FEM) model under realistic traffic loading and environmental perturbations, then curated for supervised damage classification and representation learning. The corpus supports research on deep learning for damage detection, data augmentation for time-series, and feature-space interpretability (e.g., t-SNE).

Provenance & Generation

• Structure: 4 simply supported 75 m spans, steel truss railway bridge.

• Modeling: MATLAB-based FE model (81 nodes, 175 beam elements; uniform steel properties; bearings via fixed/roller supports; added spring/rotational stiffness for model updating).

• Loading: AASHTO HL-93 three-axle truck, eight traversals at 70 km/h; added 500 kN impact and random noise to emulate operational uncertainty.

• Sensing layout: Two instrumented nodes (IDs 1002 and 3002), each with lateral Y and vertical Z acceleration DOFs.

• Sampling: 400 Hz for ~100 s per traversal.

• Scenarios: 11 distinct damage cases (0–50% stiffness reduction on selected members).

Contents & Structure

• Raw set (per scenario): 28 acceleration features × ~40,000 time points

  • Original array shape: (11, 28, 40,000)