✅ Autoencoder-Based Anomaly Detection on 6DOF SHM Data 

In a 216-meter tall building experiencing intermittent lateral wind anomalies without visible damage, I developed an unsupervised anomaly detection system using autoencoders. After cleaning and standardizing 6DOF acceleration data (3 translational, 3 rotational), I trained the model to learn normal structural behavior and flag deviations based on reconstruction error. This approach enabled near real-time anomaly detection and reduced false positive rates by approximately 40% compared to traditional threshold-based methods. 

✅ Occupant Feedback App with Real-Time Alerts 

To address the lack of user-facing tools for capturing perceived comfort during structural events, I developed a mobile application using Flutter and Firebase. The app collects floor-specific feedback from occupants, sends real-time alerts during wind or seismic triggers based on acceleration thresholds, and links user responses to corresponding time-series SHM events. This created a two-way communication channel between residents and the building’s monitoring system, and was successfully deployed as a field-ready prototype for future user studies. 

1. Synthetic Wind-Response Generator + Anomaly Detection 

2. Comfort Threshold Explorer Dashboard (ISO 10137)

3. Mode Shape Visualizer Using FEM/Modal Analysis Theory

4. Noise Detection in SHM Signals 

... and with your patronage, many more successful projects to come!

More on Projects