Frequency Response Function Dataset of a Large-Span Steel Spatial Truss Roof (Bir Mourad Raïs Multimodal Station, Algiers) for Damage Identification

This dataset contains simulated ambient-vibration frequency response data from a finite element model of the roof structure of the Bir Mourad Raïs multimodal transport station, a large-span steel spatial truss located in Algiers, Algeria. The FEM model, built in SAP2000, represents the roof as a 3D spatial truss with an elliptical barrel-vault configuration (140.094 m length, 102.750 m width, 23.885 m maximum height). Bars are modeled as frame elements with moment releases at both ends (purely axial force transmission) connecting through spherical nodes, consistent with the bolted spherical-node connection typology of the real structure. The roof cladding is represented with zero-out-of-plane-stiffness membrane elements. All bars are S235-grade circular hollow steel sections (E = 210 GPa, G = 81 MPa, fy = 235 MPa, ρ = 7850 kg/m³), with all peripheral supports fully fixed (6 DOF restrained). Structural damage is simulated as a localized reduction in the material elastic modulus (E) within 18 predefined damage zones distributed across the structure supports. Each scenario assigns an independent randomized severity level to each of the 18 zones (0 = no reduction, higher values = greater stiffness loss). The dataset comprises 10,000 randomly generated damage scenarios plus 1 undamaged (healthy) baseline scenario, for a total of 10,001 cases. For each scenario, the structural response was obtained through a steady-state (frequency-domain) analysis simulating ambient vibration excitation over a 0–20 Hz bandwidth, sampled at 101 discrete frequency points. Responses were recorded at 10 sensor locations on the roof, each capturing motion along three orthogonal directions (U1, U2, U3), giving 101 frequencies × 3 directions = 303 response values per sensor per scenario. File contents: dataset.npy — array of shape (10001, 10, 303), float32. Axis 0: scenario index (index 0 = healthy baseline; indices 1–10000 correspond to the damage scenarios in full_data.csv, row-aligned). Axis 1: sensor ID (10 sensors). Axis 2: flattened FRF response (101 frequency points × 3 directions, U1/U2/U3). full_data.csv — shape (10000, 18). Each row is a damage scenario; each column is one of the 18 predefined damage zones, with values representing the severity of the elastic modulus reduction applied to that zone (0 = undamaged) for that scenario. Note on scenario alignment: row i of full_data.csv correspond to index i+1 of dataset.npy (offset by the healthy baseline at index 0)