Predicting Deflections in Beams using Machine Learning Algorithms

Abstract

Prediction of deflections in structures is imperative for safety and stability. Classical approach and numerical solutions are currently being employed to predict the deflections of beams. However, the results from these approaches varies when compared to tests. But testing is time consuming and expensive. During the initial design stages, designers would like to quickly perform various design iterations. In order to get a quick and accurate prediction, Machine Learning models are used. In this study, regression algorithms like linear, Lasso and Ridge are used to predict the deflections in three different beams: cantilever, clamped-clamped and overhang. The database for training the algorithm is generated using the principle of superposition of forces. Around 100 datasets were generated for each type of beams in excel. Python Scikit Learn library is used to train and test the regression algorithm. The root mean square error (RMSE) for the three types of beams is nearly zero. Hence, the linear regression model resulted in high accuracy predictions. This study proposed an effective model that is cheap, accurate and efficient, to help designers predict deflections at an early stage of the design.