A functional approach to nonstationary signal analysis with automotive applications

Motivated by the strong need for the improvement of road vehicle suspensions, this thesis is concerned with theoretical and applied problems pertaining to the proper development and operation of fully active suspension systems.;Towards this end, a typical quarter-car hydraulic active suspension configuration is studied, and a non-linear computer-simulation mathematical model of it developed. This model captures the dynamics of the actuator and its associated hydraulic components in detail. These components include suspension bushing, pump accumulator, and power and bypass valves. The models of these components capture such physical characteristics as non-linear pressure-flow relationships, fluid compressibility, pump and valve non-linearities, leakages, and seal friction. Simulation results are in substantial qualitative agreement with experimental measurements. The developed model is suitable for analysis, design, control law optimization, and diagnostic strategies development.;As part of the vehicle control strategy, the problem of on-board active suspension power demand prediction is considered. Using both simulated and experimental data, the power demand signal is found to be a nonstationary stochastic process, with stationarity approximately achieved in case an averaged version of it is considered.;This approximately stationary case is studied first, and two different types of prediction schemes, referred to as direct and indirect, are developed based on novel and fast (microcomputer-suitable) signal estimation/prediction techniques that use the power demand signal history alone. The good performance characteristics of both types of schemes are verified.;In addressing the nonstationary case, a broad and basic study on the estimation and prediction of nonstationary stochastic signals is undertaken. Time-varying AutoRegressive Moving Average (TARMA) signal representations, with coefficients being explicit functions of time, are considered. Certain fundamental properties of these models are examined, and a novel and fast estimation and prediction method is developed. By offering a computational complexity that is at least two to three orders of magnitude smaller than that of the alternative Prediction Error approach, overcoming non-linear search procedures and difficulties associated with local extrema, and requiring no initial guess parameter values, this method is the first TARMA approach suitable for engineering applications and unintended operation.;The effectiveness of the nonstationary prediction method is finally demonstrated by using experimental vehicle data.