Elimination of a measurement problem: A robust prediction model for missing eigenvector value to assess earthquake induced out-of-plane failure of infill wall

Abstract

One of the effective methods to assess local or global behavior of structural system is to perform shake table experiment. The biggest challenge of all experiments, either shake table or others, are missing data during the experiments due to removed instruments. This problem is arose especially during the out-of-plane tests. This problem is named as “Measurement” problem. This paper is focused on detection the measurement problem and solving this problem by using Adaptive-Network Based Neuro Inference System (ANFIS). Definition of the problem is briefly missing out-of-plane eigenvector value during the implementation of a shake table experiments on a full scale one bay one storey reinforced concrete structural system. For this aim, two shake table experiments were performed with reinforced concrete frame (RCF) and infill wall at the National Civil Engineering Laboratory (LNEC) Portugal. The out-of-plane failure of infill wall was assessed under combined bidirectional seismic load. Shake table experiments were conducted on two types of specimens. One of them is Unreinforced Brick Infill Wall (URBIF) composed of single layer 22 cm thick brick. The other is single layer 22 cm thick Infill Wall with Bed Joint Reinforcement solution (IwBJR). Both specimens were made of a single-layer brick infill wall enclosure with the RCF. Bed joint reinforcement term refers horizontal bed joint reinforcement as a strengthening technique. Shake table experiments were performed on each specimen at four stages. After third level of earthquake load, the accelerometers were removed on the wall to prevent damage to them. The removal of these instruments results in missing data. The missing eigenvector values were predicted with a robust Adaptive-Network Based Neuro Inference System (ANFIS) model to present failure mode of infill wall. The estimation quality was tested with R2, Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE) and Root Mean Square Error (RMSE) among with available data. The results of the quality indicator of R2 are 0.999 for URBIF model and 0.997 for IwBJR model, respectively. Moreover, rest of the indicator results are less than %1. The proposed approach is reliable on the base of quality indicator. Moreover, this study can be used for seismic engineering to predict eigenvector values to see local behavior of infill wall and mode shape evolution to determine which mod prone for partial and total collapse of infill wall.