الملخص الإنجليزي
Abstract :
Recent destructive earthquakes like the Indian Ocean earthquake and tsunami
(2004) in Indonesia, the earthquake and tsunami Sichuan (2008) in China, the Tōhoku
earthquake and tsunami in Japan (2011), the Gorkha earthquake and tsunami in Nepal
(2015), and the Turkey–Syria earthquakes (2023) in Turkey and Syria have
demonstrated structural deficiencies that have caused significant loss of life and
economic damage. Most of these buildings were lacked in seismic reinforcement and
had poor seismic designs. Inadequate seismic reinforcement in beam-column joints,
load-bearing walls, and columns are the major source of damage in reinforced concrete
structures during seismic event.
Most reinforced concrete (RC) buildings built in the 1980s were not seismically
detailed; instead, they were designed for gravity loads. In seismically active areas, these
kinds of structures are widely accessible worldwide. Beam column joints play an
essential role in integrating structural systems in reinforced concrete structures. Beamcolumn joints (BCJs) are critical components in reinforced concrete (RC) structures,
which undergo high stress under seismic events. The shear strength of these joints needs
to be predicted to ensure the overall safety of structures during an earthquake event.
One of the greatest motivations for this study is to predict the shear strength of
such joints that have no seismic reinforcement. The use of machine learning (ML)
approaches in conjunction with finite element analysis (FEA) approaches has recently
seen a lot of interest from the research community. In this study, ML-ANN has been
utilized to predict the strength of RC-BCJs with a lack of seismic detailing. Developing
ANN model needs data for training, validation and testing the ANN model. Thus, data
is crucial for ANN models to perform better. In this study, data sets needed for the ANN
model have been generated using FEA software ABAQUS. Different BCJs were
simulated in ABAQUS, considering seven influential design parameters as inputs for
the ANN model. The input parameters used for the collection of data sets were column
depth, beam depth, concrete compressive strength, beam tension reinforcement total
area, beam tension reinforcement yield strength, axial column load as a ratio of its
capacity, and the joint area. The total number of 6480 BCJs were simulated in
ABAQUS and analyzed. After reviewing the results of FEM analysis flexure failure
samples were discarded and the samples with joint shear failure were considered only.
C
The total number of samples with joint failure was 4320, which has been used to
develop the proposed model. The data set of FEA samples was divided into three sets:
1) training set 70% (3024 samples), 2) testing set 15% (648 samples), and 3) validation
set 15% (648 samples). These sets were utilized to develop an ANN model to predict
the shear strength of RC-BCJ as an output variable. The performance of the proposed
ANN model was validated using the testing data set of 648. The proposed model
showed 95.6% accuracy in predicting the shear strength of the BCJ samples data. In
addition, the testing data was also compared to the ACI code equation along with the
ANN proposed model. It has been found that the proposed model accurately predicts
the shear strength of the exterior RC-BCJs as compared to the ACI equation. Moreover,
the shear strength of 80 BCJs collected from the literature was predicted through the
ANN model. It has been observed that the proposed ANN model accurately predicts
the shear strength of BCJs with an average accuracy of 91.5%.
