Scholarlink

<p>Lightweight expanded vermiculite (EV) mortars based on calcium sulfoaluminate (CSA) cement are promising for high temperature applications. However, predicting their residual strength after moderate thermal exposure (70–100℃) remains challenging. This study employs advanced acoustic emission (AE) monitoring and machine learning (ML) to address this. The key contributions are twofold: First, a novel Radial Basis Function (RBF) kernel-based approach has been introduced to dynamically classify failure modes in RA-AF analysis, overcoming the limitations of fixed threshold approaches. Second, a newly developed grouped Gaussian noise (GGN) technique has been used to augment the dataset, which has improved the performance of the LightGBM (LGBM) regression model. Experimental results indicate that while EV content reduces flexural strength, heating at 100℃ restores it by up to 48%, likely due to the formation of crack-filling hydration products. The RBF-refined AE analysis reveals a distinct transition from tensile to shear-dominated failure with accumulating damage. The optimized LGBM model, trained on GGN-augmented data, achieved high prediction accuracy (R<sup>2</sup> = 0.99, MAE = 0.18, MSE = 0.06), outperforming other mainstream models. This work proposes a combined diagnostic-predictive framework for assessing lightweight EV mortars under moderate thermal stress.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Acoustic Emission-Guided Damage Delineation and Machine Learning Prediction of Flexural Strength in Lightweight Mortar under Thermal Exposure

  • 期刊论文

  • 发表日期：

    2025年12月28日

查看全文
  • Linlin Jiang,
  • Jean Jacques Kouadjo Tchekwagep,
  • Zihao Li,
  • Fengzhen Yang,
  • Zhenxiang Chen,
  • Changhong Yang,
  • Shifeng Huang

摘要

Lightweight expanded vermiculite (EV) mortars based on calcium sulfoaluminate (CSA) cement are promising for high temperature applications. However, predicting their residual strength after moderate thermal exposure (70–100℃) remains challenging. This study employs advanced acoustic emission (AE) monitoring and machine learning (ML) to address this. The key contributions are twofold: First, a novel Radial Basis Function (RBF) kernel-based approach has been introduced to dynamically classify failure modes in RA-AF analysis, overcoming the limitations of fixed threshold approaches. Second, a newly developed grouped Gaussian noise (GGN) technique has been used to augment the dataset, which has improved the performance of the LightGBM (LGBM) regression model. Experimental results indicate that while EV content reduces flexural strength, heating at 100℃ restores it by up to 48%, likely due to the formation of crack-filling hydration products. The RBF-refined AE analysis reveals a distinct transition from tensile to shear-dominated failure with accumulating damage. The optimized LGBM model, trained on GGN-augmented data, achieved high prediction accuracy (R2 = 0.99, MAE = 0.18, MSE = 0.06), outperforming other mainstream models. This work proposes a combined diagnostic-predictive framework for assessing lightweight EV mortars under moderate thermal stress.