AI-driven data fusion modeling for enhanced prediction of mixed-mode I/III fracture toughness

• Objective : The study aims to enhance the accuracy of predictions for mixed-mode I/III fracture toughness in polymethyl methacrylate (PMMA) using artificial intelligence (AI). • Current Challenges : Traditional methods rely on destructive testing (ENDB specimens in three-point bending), which are expensive, time-consuming, and limited by data availability. • Proposed Solution : The research introduces a data fusion source modeling approach, combining fracture toughness test data with predictive data from the Maximum Tangential Stress (MTS) and Strain Energy Density (SED) criteria. • AI Models : The study uses adaptive boosting (AdaBoost) and general regression neural networks (GRNN) to improve predictive performance. • Key Findings : • Feature Importance : Mode mixity, specimen thickness, and radius are identified as crucial factors influencing fracture toughness. • Mode Mixity : The most critical factor. • Specimen Thickness : A reduction leads to decreased fracture toughness. • Radius : A complex factor with often negative impacts on toughness. • Conclusion : AI-powered models with data fusion can overcome data scarcity and improve predictions in fracture mechanics and other engineering areas with limited data sets. This research evaluates the use of artificial intelligence to enhance the accuracy of predictions for mixed-mode I/III fracture toughness in polymethyl methacrylate . Traditionally, assessing fracture toughness relies heavily on destructive testing methods, specifically using edge-notch disc bend specimens subjected to three-point bending tests. These established methods are not only expensive and time-consuming but also frequently limited by the availability of data. To address these challenges, this study introduces a data fusion source modeling approach. This approach integrates primary fracture toughness test data with secondary predictive data derived from the maximum tangential stress criterion and the local strain energy density criterion. By employing adaptive boosting and general regression neural network algorithms, the models developed in this research demonstrate a marked improvement in predictive performance compared to traditional primary source models. Additionally, a feature importance analysis using Shapley Additive exPlanations values reveals that the mode mixity parameter, specimen thickness, and radius are critical factors influencing fracture toughness. The study highlights that while mode mixity emerges as the most significant factor, a reduction in specimen thickness generally leads to decreased fracture toughness, whereas an increase in radius has a more complex, often negative, effect. The results of this study indicate that AI-powered models using data fusion can overcome limitations related to data scarcity, enabling more accurate predictions in fracture mechanics. Furthermore, this approach provides a pathway for utilizing AI in other engineering domains where data sets are limited.