Physics and Data Driven Modeling for Local Damage Evolution of Cold-Formed Steel Rack Uprights: Cases Study

Abstract

The steel rack upright is an essential member of the storage pallet rack system, which is made from mostly cold-formed thin-walled perforated steel profiles. Due to frequent storage and retrieval operations, pallet rack structures are likely subjected to accidental impact from forklifts or other material-handling robots. These impacts can cause local damage to the upright members, probably leading to the progressive collapse of the whole rack structure. In this paper, a physics and data-driven model is proposed for structure engineers to rapidly and quantitatively evaluate the upright local damage. First, finite element models of local bending damaged uprights are developed to accurately simulate the physical experimental results from the compression tests. It is shown that the corner damage of uprights is the most dangerous impact pattern compared to web damage and flange damage. Under certain special conditions, such as smaller upright sections and intermediate stiffeners, the upright with about 1 mm of local bending deformation at the corners can lead to an average decline of the carrying capacity up to 31.02%. Subsequently, a convolutional neural network (CNN) model is trained based on finite element method simulation data to predict the residual carrying capacity of damaged uprights quickly. The results obtained from the cases study indicate that the predicted values of CNNs are in good agreement with the FE numerical values, with mean absolute percentage error being 3.46%, which is a valuable decision-making tool for system engineers in the refined preventive maintenance of the automated storage and retrieval system.