The Prediction of Natural Frequencies of Cracked Plates Using the Machine Learning Methods

Authors

1 Faculty of Engineering, ShohadayeHoveizeh Campus of Technology, Shahid Chamran University of Ahvaz, DashteAzadegan, Iran

2 Department of Mechanical Engineering, Arvand International Branch, Islamic Azad University, Abadan, Iran

10.22044/jsfm.2026.17003.4017

Abstract

In this study, the first to fourth natural frequencies of a cracked plate with simply supported boundary conditions are predicted using machine learning methods, namely Random Forest and Linear Regression. Initially, a dataset comprising 200 observations was generated based on experimental and theoretical data to facilitate the prediction of natural frequencies. The crack characteristics (length, depth and location) were selected as the key parameters influencing the natural frequencies. Furthermore, 80% of the data was randomly assigned for training and 20% for testing the developed models. To evaluate the performance of the proposed models, various statistical indicators were analyzed for both the training and testing datasets. The results demonstrated that both methods yield highly accurate predictions of the natural frequencies for both training and testing data. The highest correlation coefficients for the training and testing datasets were observed to be 0.999 and 0.992, respectively. Moreover, the highest root mean square error (RMSE) values for the training and testing datasets were found to be 0.43 and 2.7, respectively, indicating the high precision of the proposed models. Additionally, an assessment of the reliability and generalizability of the models revealed that the Random Forest-based model exhibited greater stability and generalization capability across all frequency modes.

Keywords

Main Subjects

Journal of Solid and Fluid Mechanics
Volume 16, Issue 3
July and August 2026
Pages 283-296

Files

Share

How to cite

Statistics