A Fourier series neural network based friction compensator for a servo system with friction is proposed. The downward-bend low-velocity friction is modeled by an unknown nonlinear function, which is further resembled by the neural network with linear outputs to render the adaptive control applicable. Then a robust adaptive control approach is incorporated for achieving the practical tracking stability of the error dynamics. Since the Fourier series coefficients can be calculated explicitly and used in initializing the weights of the networks, therefore, it leads to faster convergence than existing similar designs.
