Predicting mechanical behavior of different thin-walled tubes using data-driven models

Abstract

The mechanical behavior of thin-walled tubes holds great significance in various engineering applications, ranging from aviation to civil engineering. This study introduces an innovative approach by utilizing machine learning techniques such as Gaussian Process Regression (GPR), Regression Trees (RTs), Artificial Neural Networks (ANNs), and Support Vector Machines (SVMs) to build data-driven models for predicting the mechanical behavior of different types of thin-walled tubes. To achieve this, we gather datasets encompassing various parameters, including material properties, pressure, and displacement. The dataset is a MATLAB array with dimensions of 2800×4. We partitioned the datasets into a training set (70 %, 1960 samples), a validation set (15 %, 420 samples), and a testing set (15 %, 420 samples). The R-squared values for the validation set are as follows: GPR (0.93), RT (0.88), ANN (0.84), and SVM (0.83). For the test set, the R-squared values are: GPR (0.80), RT (0.79), ANN (0.86), and SVM (0.82). Employing these machine learning techniques, we develop models that can predict mechanical properties for each tube category, such as compressional behavior and impact force. These models demonstrate promising accuracy and generalizability, making them valuable tools for engineering design and analysis.