محمدجواد رضايي

Predicting the evolution of microstructure after macroscopic deformation is essential for designing materials with specific properties and performance. Computational models based on crystal plasticity (CP) have been extensively developed to understand the elastoplastic deformation behavior of materials at the microstructure level under various loading conditions. However, the high computational cost of multiscale crystal plasticity finite element (CPFE) modeling poses a significant challenge. To address this issue, this study combines CP models with machine learning (ML) techniques to effectively predict texture evolution in commercially pure aluminum wires under torsional loading. Initially, the microstructure of the as-received aluminum wire was characterized using electron backscatter diffraction (EBSD). These data were used as input for the multiscale CPFE model. The CPFE framework was calibrated based on experimental results to simulate the microstructural response during torsional deformation. The evolution of the microstructure, including pole figures and texture components, was examined at different rotations. To reduce computational costs, two ML approaches—artificial neural network (ANN) and random sample consensus-based regression (RANSAC Regressor)—were employed to predict texture evolution based on the Euler angles (φ₁, Φ, φ₂) of individual grains. The models were trained using multiscale CP simulation data at different rotations, ranging from 0.5 to 2.5 turns. Validation with experimental EBSD data confirmed the effectiveness of the ML methods. A comparative analysis revealed that the RANSACRegressor performed better than the ANN, successfully predicting 81.7% of crystal orientations with an error of less than 10 degrees for half a turn and 73.5% for 2.5 turns, whereas the ANN achieved only 5.9% and 21.8% accuracy, respectively. This study also analyzed microstructure features during torsional loading, showing that certain texture components were the most preva‎lent. Quantitative eva‎luation using the Taylor factor (TF) indicated that this value ranged between 2.65 and 3.04 as the strain increased from 0.5 to 2.5 turns. Additionally, increasing the number of turns led to an 11% rise in hardness near the outer surface of samples with an initial grain size of 55 micrometers. In summary, this research presents a novel hybrid approach combining multiscale CPFE with ML techniques to predict texture evolution in polycrystalline materials. The integration of RANSACRegressor with CPFE offers a cost-effective and high-accuracy method for predicting microstructural behavior, providing valuable insights into material responses during plastic deformation. This approach can be useful for optimizing material properties in various industrial applications, paving the way for more efficient and precise material design strategies.

مدلسازي چند مقياسي , كريستال پلاستيسيته , رويكرد يادگيري ماشين , تحولات ريزساختار , كد DAMASK

Multi-scale modelling , Crystal plasticity , Machine learning approach , Microstructure evolution , DAMASK code

MohammadJavad Rezaei

Mohammad Sedighi