Multimodal Large Language Models for Adaptive 3D Concrete Printing (2025-09)¶

Drawing upon contemporary research on AI-based automation and intelligent control systems, this paper investigates the practical application of Multimodal Large Language Models (mLLMs) for facilitating reactive and automatic adjustments within the 3D Concrete Printing (3DCP) process, which lacks real-time adaptiveness to dynamic environment changes during fabrication. The proposed computational framework integrates mLLMs with simulated sensor networks and monitoring apparatus to holistically assess the quality ofprinted elements while simultaneously analysing interrelated design specifications, environmental variables, and production parameters. As a primary output, the mLLM dynamically suggests the optimal settings and operational parameters required to achieve desired structural performance objectives under fluctuating conditions. As an empirical proof of concept, we simulate diverse environmental stressors (e.g., temperature fluctuations, humidity spikes) and stringent performance requirements, under which the AI-controlled adaptive 3DCP system autonomously adjusts its extrusion, material deposition, and geometric compensation operations accordingly and in real-time.