Error compensation of photoelectric encoder based on improved BP neural network

A new method to correct and compensate the error of a photoelectric encoder was presented by using the neural network. A modeling method based on the Back Propagation (BP) was set up, in which the output follows the test value of high precision instrument and the input was the angle of sample points. The connecting weights of hidden layer and output layer were modified according to the steepest descent method. Momentum term was introduced to neural network to avoid oscillation, variable step length was suggested to accelerate study speed and avoid local optimum. Experiments showed that the precision of measuring system was improved greatly by using the BP model as error compensation, and the effect of nonlinear errors on the system was also reduced.