A measured data correlation-based strain estimation technique for building structures using convolutional neural network

Abstract

A machine learning-based strain estimation method for structural members in a building is presented The relationship between the strain responses of structural members is determined using a convolutional neural network (CNN) For accurate strain estimation, correlation analysis is introduced to select the optimal CNN model among responses from multiple structural members. The optimal CNN model trained using the response of the structural member with a high degree of correlation with the response of the target structural member is utilized to estimate the strain of the target structural member The proposed correlation-based technique can also provide the next best CNN model in case of defects in the sensors used to construct the optimal CNN. Validity is examined through the application of the presented technique to a numerical study on a three-dimensional steel structure and an experimental study on a steel frame specimen.