Human-machine collaboration framework for structural health monitoring and resiliency

Abstract Post-earthquake damage assessment can be significantly expedited when machine learning (ML) algorithms supported by sensing technologies are used. ML has some limitations when it comes to seismic damage assessment. ML tools require data from each class while training but in the case of seismic structural health monitoring, vibration data are usually available from the undamaged structures. In this paper, a framework called the human-machine collaboration (H-MC) is proposed. The H-MC framework combines the ML tools and human (domain) expertise for damage assessment of real instrumented buildings with only data from undamaged cases. It uses novelty detection as the ML tool and structure-specific analytical model to enable rapid damage detection. Subsequently, the framework is applied to detect damage in real instrumented buildings. The results showed that the H-MC algorithm correctly detected the damaged cases. It also labeled all the undamaged events accurately eliminating false positive detection. Furthermore, it is revealed that the resiliency of a building can be improved when the H-MC method is implemented over traditional field inspection. The proposed framework can be a viable tool for rapid post-earthquake damage assessment which is essential for improving community resiliency.