A deep learning–assisted structural health monitoring (SHM) framework that integrates Finite Element (FE) simulations and Neural Twin architectures to predict, analyze, and visualize the dynamic behavior of structures under various loading and environmental conditions.
This project bridges the gap between traditional Finite Element Analysis (FEA) and data-driven learning through a hybrid Neural Twin model.
It enables high-fidelity digital twin generation, vibration response prediction, and damage localization — using simulation-generated data for training and real-world motion data for validation.
- Automated FE model generation for various structural configurations
- Mesh optimization and mode shape extraction
- Modal and transient response computation
- Integration with computational solvers for high-accuracy data
- Deep learning model trained on FE-derived datasets
- Predicts dynamic responses (displacement, strain, acceleration)
- Generalizes to unseen geometries and boundary conditions
- Capable of zero-shot transfer learning for new structural domains
- Damage detection and localization through vibration signatures
- Anomaly scoring for fatigue and stiffness degradation
- Real-time comparison between physical and virtual sensors
- Dynamic response plots (time, frequency, and modal domains)
- Comparative evaluation between FE simulation and Neural Twin predictions
- Error metrics: RMSE, correlation coefficients, and spectral error maps
- Programming Language: Python
- Core Libraries: PyTorch, NumPy, SciPy, Matplotlib, Pandas
- FE Modeling: ABAQUS / OpenSees / custom FE solver integration
- Visualization: Plotly / Matplotlib
- Learning Framework: Neural Twin architecture with hybrid training
Co-authored the paper:
"Integrated Finite Element-Based Digital Twin Model and Zero-Shot Transfer Learning Approach for Structural Dynamics and Damage Detection", submitted to Expert Systems with Applications (Elsevier, 2025 – Under Review).