Neural network-based controller designs for active vehicle suspension systems
Date
2014-02-17
Authors
Dahunsi, Olurotimi Akintunde
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Vehicle suspension design necessitates achieving complex compromise between various
performance objectives. Active vehicle suspension systems (AVSS) outperforms all the other
suspension types in this regard but at the cost of higher bandwidth and power consumption as
well as, physical space constraint. This limitations have however not hindered research on
AVSS as some of the automobile manufacturers have started introducing AVSS in their
products thereby prompting improvement of its current level of performance.
The challenges of AVSS design centres around the inherent nonlinearities and uncertainties.
This explains the recent interests in the introduction of intelligent control techniques to AVSS
design. Neuro-adaptive controllers designed in this work are able to leverage on the
combination of the strengths in chosen nonlinear techniques (that is, feedback linearisation
and model predictive control) and model-based neural networks, thus avoiding the traditional
need for linearisation.
The design of an indirect adaptive, neural network-based model predictive control (NNMPC)
for a 7DOF nonlinear full-vehicle suspension design has been presented in this thesis. Its
performance was benchmarked against that of PID controller in the presence of both random
and deterministic road disturbance inputs. Improved system control was achieved by
stabilising the actuator dynamics using PID control sub-loops. The performance of the
NNMPC was superior, though the control process was slower because of the internal
optimisation routine of the NNMPC.
Control voltage, actuator force and actuator spool-valve displacement are bounded within the
specified limits. They are also well regulated (except at the instance of disturbance, however
steady state was restored within about 0.5s) without chattering. The result presented
improved ride comfort, handling and road holding without violating the suspension travel
limits. The weighted RMS body acceleration values for the vehicle sprung mass were
evaluated based on international standards. Frequency domain analysis also showed that the
AVSS was relatively insensitive to changes in the physical parameters between 7-80Hz.