Abstract: An approach using BP neural network is presented to synthesize the dimensions of a planar mechanisms, especially for the case in which the number of the prescribed positions is larger than the maximum a linkage can generates precisely. The planar displacements are mapped onto a curve in a space of mapping and two-dimensional Fourier transform is employed to find its Fourier descriptors. By kinematic simulations of crank-rocker linkages with random value of dimensions the samples are obtained, and the neural networks are trained with the Fourier descriptors as the inputs and the dimensions of the linkage as outputs. An example is given to demonstrate the approach, and its advantages over others are discussed.

Key words: BP neural networks, Kinematic mapping, Motion generation of planar mechanism