Backlash compensation by neural-network online learning

To eliminate the influences of backlash nonlinearity generally existing in servo systems, a new neural-network online learning compensation method was presented. Not basing on the identification of backlash nonlinearity, but using the online learning of neural networks, it made the output error of the system approximate to zero so that the system output could accurately follow the given input. To cooperating with this new method, the self-organizing fuzzy CMAC with Gauss basis functions (SOGFCMAC) neural network was proposed on the basis of absorbing the advantages of traditional CMAC neural networks, fuzzy logic, basis functions and self-organizing feature map (SOFM) algorithm. Finally, an actual experimental platform of servo system with low power was built to do the experimental researches. Experimental results show that the method presented in this paper can effectively get rid of the limit cycle caused by the backlash nonlinearity, and remarkably improve the system accuracy.