Multi-fidelity model using GRNN and ANFIS algorithms-based fracture criterion for predicting mixed-mode I-II of sugarcane leaves/epoxy composite

Abstract

Artificial intelligence plays a huge role in solving engineering problems. In the fracture mechanics field, artificial intelligence is used to predict fracture behavior or parameters, based on testing data, which is a destructive type of testing, so it is difficult to get a large amount of data for the learning modeling process. Therefore, this study presents artificial intelligence modeling to predict the fracture toughness by reducing the use of the actual testing fracture toughness data, while replacing the data with a sufficient quantity of data by fracture toughness prediction from the fracture criterion, for which artificial intelligence was modeled by the concept called the “multi-fidelity model”. The concept of the modeling begins with calculating and predicting the difference between the actual data set and the criterion data set, then combining the data sets, which is the data from the criterion performed mathematically with the data from the above difference prediction to the actual data set to create a model based on the increased data volume of combining the two data sets. In this study, the mixed-mode I-II fracture toughness of sugarcane leaves/epoxy composite with different mixing conditions was performed on an inclined crack specimen via the three-point bending test configuration used as the actual data set, while the generalized maximum tangential stress criterion (GMTS), which is preliminary fracture toughness, was used as criterion data. The multi-fidelity model is proposed via different artificial intelligence algorithms, namely the general regression neural network (GRNN) and the adaptive neuro-fuzzy inference system (ANFIS). Once the modeling process was complete, the performance metrics showed a clear increase in the efficiency of multi-fidelity models compared to the original artificial intelligence models. © 2023 Elsevier Ltd