Adaptive Real-time Control of Nonlinear Throttle Unit

Sammanfattning: This mastert hesis has been done as a joint project between the Department of Automatic Control in Lund and Volvo Technological Development in Gothenburg. The thesis treats methods to handle nonlinearities in a throttle unit. Theapproach has been to first design a linear controller based on the results from system identification, and then to develop an adaptive updating law estimating uncertain parameters of the throttle. The implementation has been done on a PC using Matlab.

The underlaying robust linear controller design was done with two different approaches. A first attempt was pole placement on polynomial form, i.e. an RST-controller. This controller was not easy to tune, and the design of a well behaved closed loop system failed. A second attempt was made using the method of optimal control on state space form. The use of weight functions in the design made it easier to tune the controller. The closed-loop performance of the derived linear controller meets all specifications except the rise time for small steps.

For the adaptive update law three methods were tested based on gradient theory, stability theory and a state observer. The task has been to estimate the characteristics of the main nonlinearity of the throttle, the dead-zone. The first two methods cannot estimate all of the required parameters, so they were rejected. By introducing a disturbance driven by white noise as an extra state in the state observer it is possible to estimate all the needed parameters. The closed loop performance through the dead-zone is very fast and meets almost every demand in the specification. The control law fails to meet the specified rise time for small steps and the overshoot is sometimes too large. There are also some oscillations under certain circumstances when controlling close to the dead-zone.