Synthesis of mechanical linkages using artificial neural networks and optimization

The synthesis of mechanical linkages to generate particular curves in space is a difficult problem. The use of artificial neural networks (ANNs) in combination with optimization is explored in order to synthesize an appropriate mechanical linkage that generates a user-specified curve. The problem of selecting the parameters describing a four-bar, planar linkage is examined, such that the linkage generates a coupler curve which optimally approximates a user-specified curve. In general, no exact inversion is possible, so the best fit is sought. The power spectrum of a curvature plot of the curve is used to provide a scale- and orientation-independent input representation. To perform the inversion a radial basis function ANN is trained to retrieve a tuple of linkage parameters, given a user-specified description of a curve. The network´s output provides an approximate design which is fine-tuned via gradient-based numerical optimization techniques. It is demonstrated that ANNs, in conjunction with optimization, are capable of inverse modeling of multidimensional highly nonlinear systems