Position Analysis of the Slider-Crank (R-RRT) Mechanism Using Artificial Neural Networks

Abstract

The slider-crank mechanism is a common mechanical linkage that converts rotary motion into reciprocating motion. It finds wide applications in various fields, including internal combustion engines, pumps, compressors, presses, robotics, and human-powered vehicles. Due to its widespread use, several textbooks have covered its position, velocity, and acceleration analyses, with different researchers proposing various analysis solutions. Recently, artificial neural networks (ANN) have been utilized in diverse research areas, including inverse and forward kinematic analysis. However, there has not been a specific use of ANN for position analysis of slider-crank mechanisms. This study aims to addresses that gap by presenting the position analysis of the in-line type slider-crank (R-RRT) mechanism using the ANN algorithm. For this purpose, the Levenberg-Marquardt backpropagation algorithm is selected due to its advantages, such as speed, stable convergence of training error, and the combination of Gauss-Newton training algorithm and steepest descent method. To train the algorithm effectively, 50 data sets are carefully chosen and randomly split for training, validation, and testing. Moreover, an additional 200 data sets are reserved for testing the trained algorithm to evaluate its performance. This study presents the result of the neural network algorithm training, as well as the outcomes of additional testing of the trained algorithm. These results are thoroughly discussed and analyzed.