An optimized machine learning based moment-rotation analysis of steel pallet rack connections

The moment-rotation (M-o) response of steel pallet rack (SPR) beam-to-column connections (BCCs) is naturally complex and predicted through repeated experimental investigations motivated by the variety in the geometry of commercially available beam end connectors (BECs). Past literature has shown that even finite element modeling (FEM) was somehow unable to fully capture the structural behavior of SPR BCCs in the plastic region. This study proposes an innovative Support vector machine (SVM)-Discrete Wavelet transform (DWT)-based optimized model for M-O analysis of SPR BCCs. A data set of total thirty-two experiments on SPR BCCs was used to develop the model. The experimental investigations identified the most influential parameters affecting the M-O response of SPR BCCs which are column thickness, beam depth, and depth of the BEC. Those parameters were optimized using Firefly algorithm and selected as input parameters. Reliability assessment of proposed predictive model was performed using root-mean-square error (RMSE), Pearson coefficient (r), and coefficient of determination 'r-square' (R2). The findings of predictive model were juxtaposed with experimental outcomes and FEM results, available in the literature, and a close agreement was achieved. The R2 value of 0.958 and 0.984 were achieved for moment and rotation predictions, respectively. Hence, the proposed SVM-DWT model can be efficiently used to forecast the optimum and reliable M-O response of SPR BCCs and to minimize the need of repetitive testing.