Zou Zhipeng, Peng Bin, Xie Lianghai, Song Shaoxun
School of Environment and Architecture, University of Shanghai for Science and Technology, Shanghai 200093, China.
Materials (Basel). 2024 Dec 16;17(24):6140. doi: 10.3390/ma17246140.
Ultra-high-performance concrete (UHPC) is widely used in engineering due to its exceptional mechanical properties, particularly compressive strength. Accurate prediction of the compressive strength is critical for optimizing mix proportions but remains challenging due to data dispersion, limited data availability, and complex material interactions. This study enhances the Gaussian Process (GP) model to address these challenges by incorporating Singular Value Decomposition (SVD) and Kalman Filtering and Smoothing (KF/KS). SVD improves data quality by extracting critical features, while KF/KS reduces data dispersion and align prediction with physical laws. The enhanced GP model predicts compressive strength with improved accuracy and quantifies uncertainty, offering significant advantages over traditional methods. The results demonstrate that the enhanced GP model outperforms other models, including artificial neural networks (ANN) and regression models, in terms of reliability and interpretability. This approach provides a robust tool for optimizing UHPC mix designs, reducing experimental costs, and ensuring structural performance.
超高性能混凝土（UHPC）因其卓越的力学性能，尤其是抗压强度，而在工程中得到广泛应用。准确预测抗压强度对于优化配合比至关重要，但由于数据离散、数据可用性有限以及材料相互作用复杂，这一预测仍具有挑战性。本研究通过结合奇异值分解（SVD）以及卡尔曼滤波与平滑（KF/KS）来增强高斯过程（GP）模型，以应对这些挑战。SVD通过提取关键特征来提高数据质量，而KF/KS则减少数据离散并使预测符合物理规律。增强后的GP模型能够更准确地预测抗压强度并量化不确定性，相较于传统方法具有显著优势。结果表明，在可靠性和可解释性方面，增强后的GP模型优于包括人工神经网络（ANN）和回归模型在内的其他模型。该方法为优化UHPC配合比设计、降低实验成本以及确保结构性能提供了一个强大的工具。
