Neurofuzzy and fuzzy control of automotive semi-active suspensions

Abstract

This paper describes a neurofuzzy (NF) adaptive controller in conjunction with a modelling neural network (MNN) applied to a vehicle semi-active suspension control. The NF controller employs a simplified fuzzy algorithm which is based on a multi-layer neural network. The modeling network is used as an estimator which can identify the vehicle dynamic model parameters and provide the NF controller with learning signals. A plant model is used initially to tune the parameters of the NF controller using a set of fuzzy control rules. A semi-active suspension system, utilised for vibration control tests, is modelled in simulation and tested with results obtained from vehicle experiments. Significant dynamic nonlinearity is inherent in the plant due to the suspension dynamics. The NF controller results are compared with a conventional fuzzy logic (FL) controller and an open loop passive suspension system. It is found that the NF controller performance is satisfactory and the resultant reduction of vehicle vibration is improved, in the presence of the dynamic complexities of the plant. The method proposed has several advantages over others presented in the literature. In order to verify simulation results, a test vehicle fitted with a semiactive suspension measurement and control system was experimented with under different vehicle speeds and road surfaces. In addition, a semi-active suspension and a passive suspension were also experimented. The test results coincide closely with those obtained from simulation. These results show that the semi-active suspension methodology exhibits good performance leading to reduction of vehicle vibration.