Automated Post-Earthquake Building Damage Assessment Using Smart Devices

Abstract

Conventional practices to evaluate post-earthquake damage to buildings rely on reconnaissance teams deployed to the affected areas to carry out visual inspections of buildings. These inspections are done manually and are therefore time consuming and error prone. Motivated by these drawbacks, this dissertation explores the possibility and means for conducting automated inspections using smart devices, which are ubiquitous in modern communities. The premise is that smart devices can record acceleration data using their onboard sensors. The records can then be double integrated and processed to yield interstory drift ratios (IDRs), which are key indicators of building damage. The dynamic behavior of a smart device on an underlying surface subjected to seismic motion is studied first. The smart device and its frictional interactions with the underlying surface are represented using a modified friction model. The conditions under which the smart device slides on or sticks to the surface for a given earthquake intensity are investigated. The concept of a ‘probability of exceeding the slip limit curve’ is introduced to relate the probability of exceeding a given slip limit for a given structure and location. The presence of sliding motions in an acceleration record can contaminate the recorded data and make it impossible to estimate the motion of the underlying floor from smartphone measurements. To resolve this problem, stick-slip motion identification methods are studied based on two approaches. The first method relies on the theoretical observation that acceleration is constant during sliding. The second method employs two different types of machine learning algorithms to differentiate between sticking and slipping motions. It is shown that the developed techniques can yield reasonably high classification accuracies. Computation of IDR requires multiple steps, each of which is theoretically investigated and experimentally validated by using a shake table and multiple types of smart devices with different types of protective shells. The needed steps include record synchronization and warping, data fusion, and compensation for errors that are magnified by double integration (needed to compute IDR). The abilities of different types of smart devices to estimate displacement were compared and the error in displacement was shown to have a strong relationship to their mean square amplitude of stationary noise. The proposed IDR estimation process is validated using the results from previously published shake table experiments of a four-story steel frame structure. It is shown that reasonable estimates of IDR can be achieved by using the developed methods.