Moment-Rotation Characteristics Prediction Models for Unique Boltless Steel Connections Using Machine Learning

Abstract

Beam-to-column connections (BCCs) in pallet rack structures are used for storing goods in industrial buildings, warehouses, and super-stores. BCCs must be easily demountable and reassembled to accommodate changing requirements over time. Common experimental tests for evaluating connection behaviour are expensive and time-consuming, so this study developed three prediction models using different algorithms to assess the moment-rotation behaviour of different connection types. The models were based on Support Vector Machine (SVM), Deep Learning (DL), and Decision Tree (DT) algorithms and trained using 70:30 split ratios, with further testing of 60:40 and 80:20 ratios. The models were evaluated using root mean square error, mean absolute error, and relative coefficient. The modified 60:40 DT Least Square model outperformed the other models in predicting moment-rotation behaviour, with consistent performance across all split ratios. The SVM Radial model performed poorly due to classification errors, and the DL Rectifier model made inconclusive predictions due to small sample size. The study highlights the accuracy and feasibility of various algorithm techniques in predicting BCC behaviour, enabling cost-effective and efficient testing of connections in pallet rack structures.