Identification of an LPV vehicle model based on experimental data

A physically parameterized continuous-time velocity-scheduled LPVstate-space model of a heavy-truck is identified from measurementdata. The aim is to develop a model for controller which steersthe vehicle by braking either the one or the other front wheel. Itcan be applied in many vehicles, where the sole possibility toautomate the steering in emergency situations, like e.g.unintended lane departure, is the application of the electronicbrake system. Such steering controllers usually require theprediction of the yaw rate and the steering angle on everypossible velocity. This problem defines the requirements for themodel. Four different order model structures are derived from a certainphysical description. Assuming state and output noise, all of themare identified in parameter-varying observer form using predictionerror method. The quadratic criterion function is composed frommeasurement data of several different experiments. Eachexperiments are carried out on constant velocities but the cost isconstituted from different velocity experiments. That structure is selected for controller design which has thebest cost on test data out of those the poles of which are in thecontrol bandwidth. The poles are defined on constant velocity. Theresulted nominal model consists of the feedback connection of theyaw dynamics with one state-variable and the steering systemdynamics with two states and of a first order actuator dynamicswith time-delay. The predicted outputs show a good fit to themeasurements.