Solving the Inverse Kinematics of a Five Axis CNC Machine Using Shallow and Deep Neural Networks

This study discusses the potential of using neural networks to solve inverse kinematics problems for a 5-axis Computer Numerically Controlled (CNC) machine. While neural networks have been used to solve inverse kinematics equations for robots, the complexity of the CNC machine makes it challenging. This study emphasizes the importance of selecting the correct neural network architecture and training parameters to obtain accurate and precise results. The study's contributions include neural networks that have accurate predictions, the identification of factors relating to poor prediction, and methods to enhance accuracy and precision beyond the trained dataset. Multiple shallow and deep neural networks were developed using MATLAB with different dataset sizes and training parameters to solve the inverse kinematics equations of a 5-axis CNC machine. The shallow and deep neural networks were trained using a set range of input variables, such as P-Q, α, and γ, with varying dataset sizes of 0.5%, 1%, and 10% of the larger dataset that was produced which had 179685 rows of data. This was done to ensure that the neural network would train faster and still achieve accurate results. The neural networks were then evaluated using two performance metrics, MAE and MSE, by predicting P-Q input variables within and outside of the trained dataset ranges and following a 3D quadratic spline trajectory that is within the trained dataset's range. The results showed accurate and precise predictions depending on factors such as the training dataset's size, the network's depth, the dataset's range of values, and the appropriate training parameters. Larger dataset can improve accuracy and precision if the network is overfitted, and a deeper neural network with larger dataset can make more precise predictions outside of its trained dataset. However, the factors on achieving an accurate and precise model should be taken into account to achieve better results.