Acoustic insulation optimization of walls and panels with functional graded hollow sections using graph transformer evaluator and probability-informed genetic algorithm

• Published in: Building and environment Vol. 270; p. 112550
• Main Authors: Wang, Hanmo, Nguyen, Tam H, Lu, Zhuyin, Rhiana, Febi, Fong, Dickson, He, Wei, Owyong, Shawn, Chen, Huan Ting, Lin, Alexander
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.02.2025
• Subjects: Acoustic insulation; Functional graded materials; Genetic algorithm; Graph neural network; Acoustic insulation; Functional graded materials; Genetic algorithm; Graph neural network
• ISSN: 0360-1323
• DOI: 10.1016/j.buildenv.2025.112550

•Acoustic insulation optimization of Bidirectional Variational Material (BVM).•Advanced Graph Transformer Network (GTN) for acoustic performance prediction.•A probability-informed genetic algorithm (GA) for design optimization. Sound pollution from urbanization negatively impacts health and well-being, making effective acoustic insulation in buildings. Partition walls made with Bidirectional Varied Materials (BVMs), a type of Functionally Graded Material (FGM), can improve sound insulation. However, traditional optimization methods for BVMs face challenges like (1) time-consuming Finite Element Methods (FEM), which hinder rapid design assessment, and (2) basic Genetic Algorithms (GAs) with random mutation selection, leading to inefficient searches and local optima. This paper presents an efficient optimization framework for BVM acoustic insulation. To address challenge (1), Graph Neural Network (GNN) models were employed to quickly evaluate acoustic performance, reducing reliance on time-consuming simulations. Among tested architectures, the Graph Transformer Network (GTN) outperformed others with a Mean Relative Error (MRE) of 0.8 %, capturing global structural patterns. To tackle challenge (2), a probability-informed GA was developed to enhance optimization by intelligently selecting mutation genes. Instead of random mutation, as in traditional GAs, the algorithm uses probabilistic insights to identify genes likely to improve the solution. By targeting these points, it increases mutation efficiency, focuses the search on promising areas, and significantly outperforms traditional GAs, particularly at higher mutation rates. Our method demonstrates that at 500 Hz, typical for indoor environments, the optimized BVM partition wall achieves a transmission loss of 36.92 dB. By enhancing evaluation and optimization efficiency through the established framework, this research introduces a novel approach for the efficient exploration of complex material configurations and offers a practical solution for developing FGM tailored to building's acoustic insulation needs.