Backstepping intelligent control applied to a flexible-joint robot manipulator
Control designers usually require a plant model to design a controller. The problem is the controller's performance heavily depends on the accuracy of the plant model. Using adaptive control, the plant model is now allowed to have unknown parts, but the unknown parts must only appear linearly as multiplications to the known parts. Intelligent systems such as neural networks and fuzzy systems are gaining more popularity because they can be used to approximate continuous functions without knowing their mathematical structures. However, their typical offline learning is slow and limits their usefulness. ^ Our work extends the work of other researchers who incorporate some intelligent systems into the backstepping structure to create a control system that does not require the plant model and allows online adjustment of the parameters of the intelligent systems. We extend their work in a number of interesting ways. First, we solve their problem of having to know some parts of the plant by completely using the intelligent system to learn every part of the plant. Second, we extend from using intelligent systems whose adjustable parameters appear linearly to using intelligent systems whose adjustable parameters appear nonlinearly. This leads to using more sophisticated intelligent systems such as neuro-fuzzy systems and differential neural networks. Third, we incorporate a nonlinear observer to allow the control system to be designed only from the plant's output signal. ^ Our focus is on the control system consisting of a plant in strict-feedback form, the three-layer neural network as plant estimator, a nonlinear Luenberger-type observer, and a backstepping and variable-structure controller. The control law is designed from the Lyapunov's second method. ^ We apply our control system to a two-link flexible joint robot manipulator. The flexible-joint robot manipulator, despite being a complicated and difficult-to-control system, has many practical uses. Our control system enables us to control the manipulator from only link angular position signals whereas typical control designs require link and motor angular positions, velocities, accelerations, or joint torque. We perform simulation and experimentation on an actualrobot. These results confirm the practicality and effectiveness of the control system. ^
Peter H. Meckl, Purdue University.