Scholarlink

<p>Concrete strength optimization is crucial to enhance the durability of structures, especially in situations when supplementary materials like fly ash are added. Traditional models are unable to address the heterogeneity of concrete and variability of properties of fly ash that leads to suboptimal predictions and high uncertainty. In this paper, a strong framework for the prediction and optimization of concrete strength through the integration of multimodal data and application of hybrid machine learning techniques has been aimed at for this process. The proposed methodology combines Canonical Correlation Analysis (CCA) with autoencoders in order to fuse diverse datasets like historical and real-time as well as environmental data into a unifying feature set. Then, a hybrid CNN and LSTM model is utilized in an attempt to capture spatial features from images of concrete mixtures and temporal dependencies that are present in the curing process data. BNNs with Monte Carlo Dropout are implemented to quantify uncertainty and consequently provide confidence intervals for predictions. The experimental results were the Mean Absolute Error achieved of 2.1&#xa0;MPa, Root Mean Square Error of 3.1&#xa0;MPa, and Prediction Interval Coverage Probability of 96% in comparison with the other existing methods while giving an accuracy increase by 25% and the uncertainty reduction by 30%. In conclusion, the frame of the proposed study is significant steps forward to a reliable and accurate prediction of the strength of concrete that further has practical benefits toward sustainability in construction and advanced material optimization.</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

Application of Advanced Data Fusion and Hybrid Machine Learning Techniques for Strength Prediction and Optimization of Fly-Ash Based Sustainable Concrete

  • 期刊论文

  • 发表日期：

    2025年2月26日

查看全文
  • Sagar W. Dhengare,
  • Rajesh M. Bhagat,
  • Jayant M. Raut,
  • Prashant B. Pande,
  • Boskey V. Bahoria,
  • Ashish V. Chandak,
  • Nilesh Shelke,
  • Kavita Joshi,
  • Vikrant S. Vairagade

摘要

Concrete strength optimization is crucial to enhance the durability of structures, especially in situations when supplementary materials like fly ash are added. Traditional models are unable to address the heterogeneity of concrete and variability of properties of fly ash that leads to suboptimal predictions and high uncertainty. In this paper, a strong framework for the prediction and optimization of concrete strength through the integration of multimodal data and application of hybrid machine learning techniques has been aimed at for this process. The proposed methodology combines Canonical Correlation Analysis (CCA) with autoencoders in order to fuse diverse datasets like historical and real-time as well as environmental data into a unifying feature set. Then, a hybrid CNN and LSTM model is utilized in an attempt to capture spatial features from images of concrete mixtures and temporal dependencies that are present in the curing process data. BNNs with Monte Carlo Dropout are implemented to quantify uncertainty and consequently provide confidence intervals for predictions. The experimental results were the Mean Absolute Error achieved of 2.1 MPa, Root Mean Square Error of 3.1 MPa, and Prediction Interval Coverage Probability of 96% in comparison with the other existing methods while giving an accuracy increase by 25% and the uncertainty reduction by 30%. In conclusion, the frame of the proposed study is significant steps forward to a reliable and accurate prediction of the strength of concrete that further has practical benefits toward sustainability in construction and advanced material optimization.