One goal of modern power train control systems in heavy trucks is to damp driveline oscillations using appropriate controllers. Modern control algorithms like state-space controllers are based on a state-space model, which should accurately characterize the real process behavior. Otherwise, optimal control can not be guaranteed. These state-space models include a huge number of parameters, which have to be identified by an identification process. However, existing driveline models contain two serious problems: an increasing offset over time between measured and simulated data and an inadequate detection of the longitudinal dynamics of the truck. Therefore, this article deals with two goals: to optimize the offline identification process for the special use in driveline systems and to establish an online adaptation of the model parameters to guarantee an optimal model fit.
The considered models of the driveline consists of two or more inertia masses, which are connected by spring-damper-elements. The identification of the resulting model parameters is realized by an adaptive algorithm, which minimizes a weighted cost function. Due to the existing problems, the weighting of the cost functions is extended by dividing the measured data in oscillation phases and scaling the weighting coefficients. A way to identify eigenfrequencies of measured data as well as a new identification routine solving the offset problem is established in this article. Moreover, a new way of benchmarking the identified state-space models is developed.
Dealing with the problem of changing parameters over time, a useful online-identification process using advanced recursive least squares algorithms is implemented.
The results of both methods - the offline and the online identification - are verified with measured data of a heavy truck and show a very good performance.