Prediction of Fatigue an‎d Creep Failure in selec‎ted Metallic Alloys Using Physics-Based Artificial Intelligence

3/16/2027 12:00:00 AM

Fatigue an‎d creep, known as two common failure mechanisms in industrial components, can bring about catastrophic phenomena. Multiple methods can be employed to predict these phenomena. These methods necessitate various calibration parameters to implement analyses. Moreover, empirical an‎d numerical methods simplify the physical happenings indiscernibly that can cause undeniable errors, the difference between calculated an‎d real experimental data points. Unlike static classical fitting methods, AI effectively captures complex nonlinearities through experimental data, enabling superior learning an‎d continuous model evolution. However, Scarcity of experimental datasets due to high expenses an‎d being timely, have faced AI-based methods to failure an‎d overfitting issue. In recent years, Physics-based AI models have drawn significant amount of attention towards themselves, as they have decreased data dependency an‎d therefor increased generalizability. The adoption of Physics-based AI in this work effectively minimizes data dependency by guiding the network with physical principles. This approach yields far more precise an‎d dependable results in interpolation an‎d extrapolation tasks compared to purely data-driven models, particularly under complex temperature an‎d stress profiles. In this study, physics-guided method (enforcing limits on neurons) an‎d physics-enforced method (adding a physical formula to loss function) have been imposed on neural networks. Subsequently, these deep learning models’ performances, utilizing the mentioned techniques, have been eva‎luated. These two models’ performances have been eva‎luated on two distinct datasets an‎d finally, quantified metrics, various machine learning an‎d statistical metrics, have been employed. Notably, the mean absolute percentage error of less than 11% presents the accuracy of the proposed models. At last, suggestions for improvement an‎d enhancement of AI-based models are presented.

هوش مصنوعي , شبكه عصبي , خزش , خستگي , هوش مصنوعي فيزيك پايه

Artificial Intelligence , Neural Network , Creep , Fatigue , Physics-based Artificial Intelligence

Saba Abbaszadeh Montazeri

Prof Dr. M.R.Ayatollahi