Prediction of impact behavior of notched composite pipes repaired with different composite patches by machine learning algorithms

In this study, the impact behavior of notched glass fiber reinforced composite pipes repaired by patch bonding is investigated and the impact performance is predicted by different machine learning algorithms. The effects of patch material, patch thickness, adhesive type and impact velocity on peak force (PF), absorbed energy (AE), maximum displacement (MD) and failure zones were investigated. In the finite element model, Progressive failure analysis based on the combination of Hashin failure criteria and Cohesive zone model (CZM) with Bilinear traction-separation law using MAT-54 material model with LS DYNA finite element program. In addition, Linear Regression (LR), Support Vector Machines (SVM), k-Nearest Neighbors (k-NN) and Random Forest (RF) machine learning algorithms were used to predict the maximum strength values of composite pipes under impact. The peak force value of the patched specimen increased by maximum 87.2% while the energy absorption efficiency value decreased by 78.9% compared to the unpatched specimen. It was determined that the patch reduced the impact failure zone of the specimen. KNN has the highest accuracy rate with 85% in predicting PF, while RF is more reliable than other algorithms in predicting AE and MD outputs with 97% and 87%, respectively.