TY - GEN
T1 - Solving the forward kinematics problem of a parallel kinematic machine using the neural network method
AU - Zhang, Dan
AU - Shi, Qi
AU - Lei, Jianhe
PY - 2010
Y1 - 2010
N2 - In this paper, the forward kinematic problem of a 3-DOF asymmetrical parallel kinematic machine has been solved using multi-layered neural network method. The mathematical expression of the multi-layered feed forward neural networks with back-propagation is introduced. Neural network of the 3-DOF parallel manipulator was trained based on the data generated from the closed-form inverse kinematic solution. Levenberg-Marquardt algorithm is applied to train the moderate-sized feed forward neural network, which leads to the fastest convergence of the network. After the learning stage, the network was tested on a random selected data and the simulation results showed that the neural network configuration can receive the expected accuracy of the parallel manipulator. The optimal artificial neural network model of the 3-DOF parallel manipulator has been found with tan-sigmoid transfer function for the middle hidden layer. The trained neural network model of the forward kinematics of the 3-DOF parallel manipulator can be used for real-time control purpose.
AB - In this paper, the forward kinematic problem of a 3-DOF asymmetrical parallel kinematic machine has been solved using multi-layered neural network method. The mathematical expression of the multi-layered feed forward neural networks with back-propagation is introduced. Neural network of the 3-DOF parallel manipulator was trained based on the data generated from the closed-form inverse kinematic solution. Levenberg-Marquardt algorithm is applied to train the moderate-sized feed forward neural network, which leads to the fastest convergence of the network. After the learning stage, the network was tested on a random selected data and the simulation results showed that the neural network configuration can receive the expected accuracy of the parallel manipulator. The optimal artificial neural network model of the 3-DOF parallel manipulator has been found with tan-sigmoid transfer function for the middle hidden layer. The trained neural network model of the forward kinematics of the 3-DOF parallel manipulator can be used for real-time control purpose.
KW - Back-propagation
KW - Forward/inverse kinematics
KW - Multilayer perceptron (MLP)
KW - Neural network
KW - Parallel kinematic machine
UR - http://www.scopus.com/inward/record.url?scp=78149372914&partnerID=8YFLogxK
M3 - Conference article published in proceeding or book
AN - SCOPUS:78149372914
SN - 0872638685
SN - 9780872638686
T3 - Transactions of the North American Manufacturing Research Institution of SME
SP - 727
EP - 733
BT - Transactions of the North American Manufacturing Research Institution of SME 2010, NAMRI/SME
T2 - 38th Annual North American Manufacturing Research Conference, NAMRC 38
Y2 - 25 May 2010 through 28 May 2010
ER -