Machine Learning Models for Predicting Deflection and Shape of 2D Cantilever Beams

Abstract

Abstract Efficient predictive models for large deflection of beams are critical for the design and analysis of compliant mechanisms. Machine learning (ML) models for predicting large tip deflection and shape of 2D cantilever beams are proposed for the analysis of compliant mechanisms in this paper. First, the training data is collected by running massive finite element simulations of cantilever beams subject to various loading conditions. Second, these training data are fed into several nonlinear regression learners. The result is a generalized predictive model for any 2D beams subject to a wide range of tip loads. The evaluation results show the maximum error is about 3.8%. The machine learning models with high prediction speed and accuracy are obtained. Two extreme loading conditions are studied to evaluate our machine learning model of deflections. Our results show that the error of our ML model is less than that of the conventional PRB 3R model, and the speed is 98 times that of the FEM in these two studies. A numerical example of 4-bar linkages with a 2D beam is applied to demonstrate the use of the machine learning model of the beam shape. The result shows the very close shape and high efficiency of our model.