Shape optimization for path synthesis of crank-rocker mechanisms using a wavelet-based neural network

Some recent developments in path generation have been based on neural network mechanism databases, which instantaneously provide an approximate solution of the synthesis problem. We describe a way to reduce the design space, ensuring that the neural network always yields a consistent crank-rocker mechanism with optimal transmission angle. Moreover, instead of the usual strategy of using Fourier coefficients, we propose a new method based on wavelet descriptors to represent the shape of the path, where the points do not need to be sampled at a constant time interval. Numerical results demonstrate the superiority of this wavelet-based neural network over the Fourier-based network in finding the optimal mechanism. They also show the accuracy of the proposed approach in providing near optimal crank-rocker mechanism solutions for path generation.