Structural health monitoring is an increasingly popular tool for evaluating post-earthquake building damage. However, sensors are often not placed on all floors due to space, cost, and/or usage limitations. This may result in difficulty classifying the lateral deformation mode (e.g., total yielding or soft-story). However, the lateral deformation mode is important as it affects the estimation of the critical floor response and damage classification. As such, there is a need to identify the deformation mode based on the known building response. Past studies had shown that engineering features based on peak and yield displacement profiles of known building response could potentially be used in machine-learning classification applications. However, it is often difficult to capture floor displacements accurately, particularly if accelerometers were used to monitor building response due to low-frequency noise incurred when double integrating accelerations. This study looks to evaluate the suitability of non-displacement-based features for identifying the deformation mode of buildings without sensors on all floors. Building response parameters considered included peak floor acceleration, peak floor velocity, floor cumulative absolute velocity, among others. Firstly, inelastic dynamic analyses of multistory frame buildings considering a range of different deformation modes were performed. Afterwards, various techniques were applied to determine the importance of each feature. It was found that building response parameters normalized by the roof response were evaluated as being more important compared to their absolute response parameter counterparts. In addition, individual parameters were unable to properly indicate the deformation modes on their own, with the lowest misclassification rate being about 40%. However, when considering a combination of features for calibrating classification machine learning models in future applications, the misclassification rate was decreased significantly to 13%. This indicates that a combination of features should be used for developing the deformation mode classification model.
各種検索
サポート
ヘルプ