In nonlinear system identification, black-box and grey-box models are widely used. Such methods require no or only little physical or formal information. In this paper, the nonlinear system identification of the vertical dynamics of a vehicle is addressed, which serves as benchmark problem for the proposed algorithm. Vehicle models are used for a variety of purposes, for instance tuning of control algorithms in a computer simulation environment helps to reduce expensive in-vehicle tuning effort. Such models can be from different classes like multi-body physical models or neural network models. In the present contribution, the nonlinear identification of the vertical dynamics of a half-car vehicle model using only noisy measurements of road excitations as input and the suspension deflections as output is investigated. The integration of prior knowledge by means of equality constrained GTLS (generalised total least quares) parameter optimisation was presented. As was shown, prior knowledge of the nature of the system reduces the complexity of the model and parameter estimation itself. Moreover, the GTLS algorithm yields consistent parameter estimates when some or all input channels are subject to noise and covers both LS and TLS as special cases. Together with a well-tried LMN (local model network) approach, the CGTLS algorithm has an outstanding capability for nonlinear, complex system identification tasks. The efficiency of the proposed concepts was validated by modelling the vertical dynamics of a vehicle. For the sake of transparency and reproducibility of the results, a significantly nonlinear but yet elementary model of a car was considered the "real" vehicle. Indeed, the full power of the proposed MIMO (multiple-input and multiple-output) model with the CGTLS algorithm will appear in problems with still more input and output quantities, and if either the knowledge about the interior structure of the system is limited or if it is very complex and therefore difficult to model; in vehicle vertical dynamics, the very problem of a sophisticated 3D-suspension with only approximately known stiffnesses and damping parameters might be such a topic. Another problem in vehicle dynamics which offers a very promising application of the meth ods presented above is the computational prediction of the driveability of cars, which means the qualitative assessment of the vehicle's response to the driver's input. There, many input quantities such as longitudinal acceleration, engine speed, vehicle speed, pedal positions, etc., and several output quantities like kick, surge, engine speed fluctuations, response delay, etc., must be interrelated, taking moreover different driving modes into account: this problem is a subject of current research. However, it shall finally be emphasized that the field of applications of the procedures suggested in this paper is by no means restricted to vehicle dynamics, but they rather cover a large variety of complex identification tasks in Solid Mechanics, Fluid Mechanics and Electrics, and also combinations thereof.
An advanced algorithm for partitioning and parameter estimation in local model networks and its application to vehicle vertical dynamics
Acta Mechanica ; 223 , 8 ; 1693-1706
2012
14 Seiten, 12 Bilder, 1 Tabelle, 24 Quellen
Aufsatz (Zeitschrift)
Englisch
Parameter Estimation of Vehicle Handling Model Using Genetic Algorithm
SAE Technical Papers | 2002
|Parameter estimation of vehicle handling model using genetic algorithm
Kraftfahrwesen | 2002
|Advanced Natural Parameter Methods for Vehicle Dynamics
British Library Conference Proceedings | 1996
|Parameter Estimation (Identification) of Vehicle Dynamics
Springer Verlag | 2021
|Vehicle parameter estimation using nested RLS algorithm
IEEE | 2013
|