We are designing a framework for structural health monitoring that functions fully online, without any prior information about the structure. As part of this effort, we are exploring several key principles that could help to accomplish this goal:
- Machine learning strategies that can handle high dimensional data, in order to avoid information loss due to dimensionality reduction.
- The use of Generative Adversarial Networks to overcome data scarcity and data imbalance.
- Alarm threshold tuning based on reliability analysis to reduce sensitivity to user-defined parameters.