Model-Based Adaptive Machine Learning Approach in Concrete Mix Design.

Area of Science:

• Civil Engineering
• Materials Science
• Computer Science

Background:

• Current concrete mix design methods, often based on the Three Equation Method, struggle to meet modern demands for structural properties, production efficiency, and environmental friendliness.
• Existing analytical and laboratory procedures can lead to difficulties in predicting final concrete properties and necessitate precautionary oversizing.
• The increasing complexity of concrete mixtures requires more sophisticated prediction tools.

Purpose of the Study:

• To develop a novel, adaptive machine learning approach for predicting concrete compressive strength.
• To address the limitations of current methods in accurately reflecting the variability of modern concrete mixes.
• To provide a reliable tool for estimating concrete properties based on ingredient composition.

Main Methods:

• Development of a machine learning model utilizing a deep neural network architecture.
• Training the model on an extensive database of concrete recipes.
• Translating the trained model into a predictive mathematical formula.
• Testing and comparative analysis against contemporary design methods (Bolomey and Fuller).

Main Results:

• The proposed adaptive algorithm significantly outperforms models without adaptive features when trained on the same dataset.
• The model accurately estimates concrete compressive strength based on main ingredient composition and two batch observations.
• The developed algorithm demonstrates superior performance in predicting concrete strength compared to traditional methods.

Conclusions:

• The novel adaptive machine learning algorithm offers a significant advancement in concrete mix design.
• This tool can effectively serve as a concrete strength checking mechanism within the design process.
• The method enhances the prediction accuracy for concrete compressive strength, accounting for mix variability.