<![CDATA[This article addressed a machine learning approach for determining a solution model for the direct kinematic problems of parallel manipulators. A 5R planar parallel manipulator was utilized for that approach because it had the solution in the closed form. Initially, a dataset was created from an inverse kinematic solution of the manipulator for one of its assembly modes. Then, this dataset was fed as the input (the joint space) and the output (the platform space) for modeling the direct kinematic solution of the manipulator using one of the machine learning algorithms, which was the decision tree. To avoid overfitting during the training, hyperparameter tuning was employed to find the best parameters for the decision tree model, which was later called the best model. Hence, the best model can be validated by using the closed form solution. If the best model failed to model the direct kinematic solution in the validation, remodeling had to be performed and executed the same steps again. For remodeling, the training dataset consisted of all assembly modes of the manipulator. Consequently, the best model after remodeling was able to present the direct kinematic solutions for all possible input domains. Unfortunately, around 5% of solutions shown a higher deviation which had to be investigated in the future.]]>
Copyrights © 2021
