Generalized Input Preshaping Vibration Control Approach for Multi-Link Flexible Manipulators using Machine Intelligence

Abstract

In this study, an open-loop vibration control through parameter tuning method for multi-link flexible manipulators is generalized using the machine learning paradigm. The experimental studies part of the study consists of two stages. In the first stage, a decision tree model is created using the C4.5 algorithm to predict transient (during motion) and residual (stationary) vibration levels, and decision rules are derived from this model. The paper utilizes the C4.5 (decision tree) algorithm on a broad experimental dataset and compares its performance to that of other four well-known traditional machine learning algorithms (Naive Bayes (NB), Multilayer Perceptron (MLP), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN)). The reason for choosing the decision tree method is that it has a transparent decision-making mechanism that works in tandem with the Explainable Artificial Intelligence (XAI) perspective. The experiments showed that the C4.5 algorithm achieved the best classification performance for predicting both root mean square (RMS) values of residual and transient vibrations (RMSres and RMStrans) with 92.25% and 83.75% accuracy rates. In the second stage, the generalized coefficients for use in parameter tuning were found by interpreting the rules obtained from this model. Furthermore, the rules derived from the tree in this study were applied to different systems. Control applications showed that an average of 93.9% and 84.5% suppression ratios in residual and transient vibrations could be achieved with this method. In addition, vibration settling times were shortened by an average of 41.4 seconds. Following the experimental studies, the simulation was used to compare results with the aid of Matlab Simscape-Multibody. The results prove that the rules obtained from the tree in this study generate successful results in several different systems.