Abstract
Advancement in Artificial Intelligence (AI), particularly Machine Learning (ML), continues to transform many facets of our lives. Focused on making recommendations based on data, Machine Learning models are trained by algorithms to make decisions based on the input data. This research aims to build machine learning models to predict shock absorber spring deformation when a load is applied to it, using data obtained from finite element analysis. Additionally, the project seeks to compare regression and neural network models to identify a more robust approach to predict shock absorber spring deformation on Python and MATLAB environments.
A shock absorber 3D model was designed using Autodesk Inventor 2022, and 125 simulations were performed to generate the data. The coil diameter, wire radius, and the number of active coils were varied for each simulation to obtain a different spring deformation at each stage. Simulation results were used to build models of prediction. These models were built successively using several regression approaches (Linear regression, Lasso regression) and algorithms (Random forest, XGBoost, Levenberg-Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient). These regression approaches and algorithms were selected to get an insight into the building of models on different platforms and identify the most robust approach to analyse the deformation of shock absorber spring.
Upon evaluating the regression approaches and neural network models, the Bayesian Regularization neural network model exhibited the highest prediction performance, with a regression value of 0.99981 corresponding to a configuration of 20 nodes. This also implies that Neural networks give a better approximation when compared to regression models in this particular application. The linear, random forest and XGBoost approaches yielded regression values of 0.927, 0.958782, and 0.9793, respectively, with the Lasso regression having a performance metric of 0.917. This study provides a flexible design approach and demonstrate the potential of machine learning for the estimation and the analysis of shock absorber spring deformation.
Keywords: Deformation, Machine Learning, MATLAB, Neural Network, Python, Regression Analysis, Shock Absorber Spring.