An AI-Driven Low-Cost Cyber-Physical Digital Twin Framework for Steel Column SHM: Integrating IoT Telemetry and Machine Learning Surrogate Models

Civil infrastructure failures pose profound socioeconomic risks, demanding advanced diagnostics over traditional Structural Health Monitoring (SHM), which relies on expensive wired networks and periodic inspections that miss rapid real-time buckling in slender columns. This paper presents a low-cost Cyber-Physical Digital Twin (CPDT) framework linking affordable IoT hardware with rigorous numerical validation. A 25 mm × 6 mm × 490.22 mm flat-plate steel column was tested under incremental axial compression; a nonlinear Abaqus model (incorporating elasto-plasticity and a 1.5 mm geometric imperfection) predicted the 3,150 N failure load within 0.6%. A custom $25.30 Arduino-based IoT node calibrated to this baseline converged within 0.28% of the peak load benchmark. Training nine machine learning algorithms on a 1,000-sample Latin Hypercube dataset, Gaussian Process Regression (GPR) emerged superior, achieving R²=0.993 (cross-validation) and R²=0.999 (testing). Parallel-threaded Python middleware guarantees sub-millisecond latency for live synchronization of Abaqus Von Mises stress contours. An autonomous closed-loop alert successfully triggered pre-failure warnings at 90.3% of the ultimate load. Furthermore, a Multi-Geometry Scaling Framework—utilizing an Inverse Strain Recovery algorithm (\(\mathcal{E} = \frac{\mathrm{Papp}}{\mathrm{ExAbase}}\)) and Double V-Lookup database interpolation—scales the system to arbitrary dimensions without physical recalibration, yielding a 1.75% mean absolute percentage error across ten blind-validation specimens. This provides a scalable, verified blueprint for autonomous structural risk mitigation, reducing costs by over 97% versus conventional SHM.