Neural Network-Based Estimation of Rebar Deflection for Robotic Rebar Installation of Vertical Structures

Abstract

In construction, most existing research on rebar placement focuses on guardrail system for large-scale planar slabs or multi robot arm system for off-site rebar cage assembly, frequently utilizing gantry systems. However, their substantial setup time and spatial requirements greatly reduce their suitability for smaller-scale or more spatially constrained tings. This is particularly pronounced in on-site application where the rebar installation of vertical structural elements occurs in fragmented and restricted workspaces, limiting the applicability of existing guardrails or multiple robotic systems. To address this gap, this study investigates the feasibility of employing a single arm robotic system to grasp and position rebar for vertical structural components. One critical challenge in this context is to account for rebar deflection during rebar placement operations. Moreover, existing analytical equations do not provide sufficiently accurate deflection estimates. To address this problem, a neural network model was developed to predict rebar deflection across a range of sizes and lengths, allowing the robot to adapt its position dynamically according to the target location. The proposed model achieved an R? accuracy of 0.9714 and outperformed other models in both the +0.01m and +0.092m thresholds, demonstrating its effectiveness in providing precise deflection estimates. This level of precision facilitates efficient robotic rebar placement using a single-arm system. system.