Optimal Selection and Process Design of High-Performance Polymer Composites for Smart Construction Based on the "Agent AI" Framework

Against the backdrop of rapid advancements in smart construction and digital twin technologies, traditional building material selection and construction decision-making processes face challenges of inefficiency, lack of systematicity, and imprecision due to excessive reliance on manual expertise. To address this issue, this paper proposes a novel decision-making framework based on "agentic AI" (AI with multiagent capabilities), aiming to achieve automated optimization and process design for high-performance polymer composites in smart construction. The core of this framework is a multiagent system. A conductor agent based on a large language model (LLM) parses complex project requirements and coordinates multiple specialized agents to execute tasks, including data retrieval, performance prediction, multiobjective optimization, and interpretable reporting. The performance prediction module innovatively employs an "LLM-XGBoost" hybrid cascading architecture. It leverages the LLM's reasoning capabilities to intelligently tune XGBoost model hyperparameters, enabling high-precision quantitative predictions of material properties, cost-effectiveness, and construction efficiency. To ensure transparency and credibility in the decision-making process, the system integrates an explainability (XAI) module based on Shapley Additivity Propensity (SHAP) analysis, enabling quantitative assessment of each input parameter's contribution to the final recommended solution. Case studies on two typical scenarios — "anti-corrosion coatings for cross-sea bridges" and "self-healing concrete for high-rise residential buildings" — validate that this framework can accurately recommend optimal material solutions such as "0.7% graphene-modified epoxy resin" and "microcapsule-based self-healing polymer concrete." It predicts potential savings of 60–80% in full-lifecycle maintenance costs and reductions of 25–35% in the carbon footprint. This research not only provides a data-driven, explainable, end-to-end intelligent decision-making tool for engineering management but also demonstrates the immense potential of agent-based AI in driving digital transformation and sustainable development within the construction industry. This study offers crucial theoretical and technical support for accelerating the adoption and widespread use of novel green building materials.