Failure detection in single plate shear connection with LSTM network

The single plate shear (SPS) connection is a cost-effective choice for beam to column connections. The failure detection method for the connections helps people to respond and receive an early warning, allowing them to take action earlier and prevent serious consequences. Thus, a quick solution is needed to address the structural safety monitoring issue and predict the response of the SPS connection with accuracy and speed. The present study recommends the behaviour of SPS connection to predict its failure mode using long-short term memory (LSTM) networks. The LSTM models were designed utilising the datasets generated by applying a finite-element method (FEM) of SPS connection. Experimental Validated model was used for finite element analysis of 48 SPS connections with different seven parameters like distance, the depth of the shear plate connection plate, thickness of web, thickness of shear plate, the number of bolt columns, Grade of bolts, the beam span, and a number of common wide flange beam shapes. The LSTM model utilise von Mises stresses for analysing the failure mode of the SPS connection was compared to the results of the artificial neural network (ANN) and the FEM. The output of an LSTM model was nearly identical to an ANN model, with ANN model performing slightly better. In all three outputs, the ANN model's performance is satisfactory (r > 0.8). The training and testing of the ANN models required an average of six seconds, whereas the analysis of the deep LSTM network consumed almost an hour. The comparison shows that the ANN model predicted stresses more accurately than LSTM, at least for the current work, which reduces the necessity of using LSTM for the said task.