Error compensation for high precision reference encoder based on RBF neural networks

The angle errors of high precision reference encoder deeply influence the precision of small size absolute photoelectric encoder's detection equipment. There are many factors that influence encoder's errors which can't be described by mathematical model. So in this paper, a method using RBF networks was proposed to amend errors. Firstly, the method used several kinds of polyhedron to detect the reference encoder with collimator, so that it could get more points of errors, and the errors were composed in one coordinate curve. Secondly, the RBF networks were built by using the result of error detection, in order to make their outputs closed to the true angle. Lastly, the reference encoder applied in small size absolute photoelectric encoder's detection equipment was compensated by the circuit of compensation. The experiments show that the circuit is fast, convenient and not influenced by the complexity of networks. The precision of encoder is improved 2 times, and this method improves the precision of detection equipment.