Neural networks-based adaptive control of uncertain nonlinear systems with unknown input constraints

In this work, we solve the adaptive actuator backlash compensation control problem of uncertain nonlinear systems. A new generalized backlash model is first proposed, which takes both the actuator perturbation and unidentifiable coupling into account, and hence captures the practical backlash behavior more accurately. Nevertheless, such a model makes the adaptive control design difficult, where the most challenging one is that the unrecognizable coupling makes traditional compensation structure no more feasible. To address this issue, we propose an adaptive compensation control structure synthesizing neural networks learning and novel smooth backlash inverse model. With the established compensator and the iterative control design of compensator input, an adaptive neural controller is subsequently proposed to guarantee that all signals of the closed-loop system are bounded, and the tracking error converges to residual of zero asympotically. Simulation results are given to verify the effectiveness of the proposed control scheme.