Characterization and optimization of mechanical properties in design materials using convolutional neural networks and particle swarm optimization

Recent years have seen the rise of sophisticated materials in design and manufacture. They are crucial in many industries because of their strength, flexibility, and durability. While effective, conventional techniques of describing these qualities may be time-consuming and may not fully use the considerable quantity of data available. Integrating machine learning and intense learning with material science processes may solve the issue. In this research, we use convolutional neural networks (CNNs), known for image and pattern recognition, to understand material microstructure patterns and forecast their mechanical properties. PSO is used to enhance and optimize these predictions. Our technique uses a CNN-based system to create, train, and validate models for material property predictions and a unique PSO-CNN integration to maximize model parameters and prediction accuracy. Our studies showed that the dataset's mean ultimate strength is 572.75 MPa, a standard for CNN training. Data variability necessitates complex CNN architectures with PSO-optimizing parameters. Poisson's ratio and density fluctuations suggest material modification. This study establishes a framework for characterizing and optimizing mechanical characteristics in design materials and links computational methods to real-world material science applications. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2023.