A Neural Fuzzy Approach for Controlling Active Vehicle Suspension Systems

Abstract

In this work, a novel neural fuzzy (NF) control scheme is introduced to design a fully active suspension system. The fuzzy part of the controller handles uncertainties, whereas the neural part learns from past events and tunes the controller to optimize the performance of the suspension system. The two-degree-of-freedom quarter-car model is used to illustrate the control strategy and to evaluate the performance parameters for sinusoidal and random road inputs. A sinusoid road surface description is first used to obtain an initial design of the NF controller. The acceleration of the sprung mass is compared with that of an ideal skyhook model to produce an error signal, e(t); this error signal, as well as Δe(t) are employed as inputs to the controller. Results obtained for this type of road input indicate that the NF active system has significant advantages over the linear quadratic regulator (LQR) active suspension system. In order to get a broader view, more realistic road descriptions and practical control laws were used. The performance parameters were computed when the road surface was presented as a random road input. The control law of the NF active system was modified to achieve a novel non-linear control (NC) strategy. This control law requires only measurement of the body acceleration and the road input displacement, and hence, it can be realized easily in practice when compared with all other control laws, including the LQR one. For a wide range of road surfaces, results show that the performance capability of this novel system is much better than that of the LQR active system. For example, the improvements, under a medium-quality road surface and a 30-m/s vehicle speed, achieved a reduction in the rms values of the ISO weighted body acceleration and the dynamic tyre load by 17% and 20% respectively.