Modeling Self-Healing of Concrete Using Hybrid Genetic Algorithm-Artificial Neural Network

Area of Science:

• Civil Engineering
• Materials Science
• Computational Intelligence

Background:

• Self-healing concrete aims to enhance infrastructure durability and reduce maintenance costs.
• Predicting self-healing performance is complex due to numerous influencing factors.
• Optimization techniques are crucial for developing accurate predictive models.

Purpose of the Study:

• To develop a hybrid genetic algorithm-artificial neural network (GA-ANN) model for predicting intrinsic self-healing in concrete.
• To optimize the artificial neural network (ANN) using a genetic algorithm (GA) for improved prediction accuracy.
• To evaluate the model's capability in capturing the effects of various self-healing agents.

Main Methods:

• Implementation of a genetic algorithm (GA) as a stochastic optimizer for ANN weights and biases.
• Training and validation of the GA-ANN model using a custom database from experimental studies.
• Utilizing inputs such as cement content, w/c ratio, supplementary cementitious materials, bio-healing agents, and additives.
• Measuring self-healing performance via crack width reduction.

Main Results:

• The GA-ANN model successfully predicted the self-healing performance of concrete.
• The model demonstrated the ability to avoid local optima, achieving a global optimum.
• The approach effectively captured the complex interactions of various self-healing agents.
• Accurate prediction of crack width reduction was achieved based on material inputs.

Conclusions:

• The proposed GA-ANN model offers a robust approach for predicting concrete self-healing.
• This predictive capability can aid in the design of more durable and sustainable cement-based materials.
• The hybrid optimization strategy enhances the reliability of predictive models in materials science.