Knowledge Transfer Between Buildings for Post-Earthquake Damage Diagnosis Without Historical Data

Abstract

Automated structural damage diagnosis is important for improving the efficiency of disaster responses and rehabilitation. In conventional data-driven frameworks using machine learning or statistical models, structural damage diagnosis models are often constructed in a manner of supervised learning. The supervised learning utilizes collected historical structural sensing data and respective damage states (i.e., labels) for each building to learn the building-specific mapping between them. However, in post-earthquake scenarios, historical data with labels are often not available for many buildings in the affected area, which makes it difficult to construct a damage diagnosis model. Directly using the historical data from other buildings to construct a damage diagnosis model for the target building would lead to inaccurate results. This is because data-driven models assume that the data distribution of training dataset (from other buildings) is consistent with that of test dataset (from the target building), while in practice each building has unique physical properties and thus unique data distribution. To this end, we introduce a new modeling framework to transfer the knowledge learned from other buildings to diagnose structural damage states in the target building. This framework is based on an adversarial domain adaptation approach that extracts domain-invariant feature representations of data from different buildings. The feature extraction function is trained in an adversarial way, which ensures the extracted feature distributions are robust to changes in structures while being predictive of the damage states. With the extracted domain-invariant feature representations, the data distributions become consistent across different buildings. We finally adopt the damage diagnosis model learned from other buildings datasets to infer the damage state of the target building. We evaluate our proposed framework using data collected from 5 different buildings subjected to 40 different earthquakes. We transfer the model learned from multiple buildings to diagnose a new building subjected to new earthquakes. The results show an up to 33.8% improvement in damage detection accuracy compared to conventional methods.