Influence of Optimization Algorithms and Computational Complexity on Concrete Compressive Strength Prediction Machine...

Area of Science:

• Civil Engineering
• Materials Science
• Artificial Intelligence

Background:

• Traditional concrete mix design methods struggle with modern complexity.
• Machine learning models show promise in predicting concrete compressive strength.
• Optimizing AI models is crucial for advanced concrete technology.

Purpose of the Study:

• To investigate the impact of computational complexity on AI model performance in concrete mix design.
• To evaluate the effectiveness of different optimization algorithms (QNM, ADAM, SGD) for AI models.
• To identify optimal AI strategies for accurate concrete compressive strength prediction.

Main Methods:

• Trained and tested forty-five deep neural network models with varying complexity.
• Utilized a comprehensive database of concrete mix designs and compressive strength data.
• Compared the performance of Quasi-Newton Method (QNM), ADAM, and Stochastic Gradient Descent (SGD) optimization algorithms.

Main Results:

• A significant interaction was found between optimization algorithms and model complexity in improving prediction accuracy.
• Models using the Quasi-Newton Method (QNM) demonstrated superior performance.
• QNM outperformed ADAM and SGD in reducing prediction errors (SSE, MSE, RMSE, NSE, ME) and increasing R².

Conclusions:

• The Quasi-Newton Method (QNM) is highly effective for optimizing AI models in concrete mix design.
• AI-driven approaches, particularly with QNM, offer more accurate and efficient concrete design solutions.
• This research advances AI applications in concrete technology, paving the way for future innovations.