A Neural Network Based Control for Semi-Active Suspension of Military Truck 5 Ton Hino GT

Abstract

The basic objectives of the suspension system include to not only support the weight of the vehicle, but also to improve ride comfort and vehicle handling. In conventional (passive) suspensions, there is a trade-off among better ride comfort or superior handling for a specific range of operational conditions, since these two requirements are contradictory. The semi active suspension on the other hand, provides a room for adjustability of the suspension system parameters (stiffness or damping coefficients) to cover much broader operational situations. The objective of this research is to control a semi-active suspension system of a heavy vehicle by designing a neural nonlinear autoregressive with exogenous input (NARX) controller. The system is studied by modelling it as a quarter truck semi-active suspension model. The modelled system is initially controlled by the help of a PID and LQR controller for a Class-D road profile. The PID based system was used to extract array of time-series input/output data set for the training of the NARX controller. A multi-layered neural network, with single hidden layer was generated and trained. The response of the systems was compared by using basic suspension system performance parameters which are; the vertical displacement and the vertical acceleration of the sprung mass (ride comfort) and the tyre deformation (tyre holding). The peak values and the RMS were taken as the performance indicators. The results showed that the designed NARX controller based semi active system, demonstrated much better performance as compared to PID, LQR and passive systems when subjected to unknown, not previously trained / tuned, road profiles. All the systems were modelled and simulated in MATLAB Simulink. The research demonstrated the efficacy of the artificial neural networks to better control the damping force of an active damping element of a semi-active suspension system which can be then be used to design much more comfortable and stable suspension systems. Finally, the NARX network controller was successfully applied to a full-truck model using TruckSim software, for analysing full truck multibody vehicle dynamics model through co-simulation. The designed NARX controller was integrated using the Simulink - TruckSim interface to set up a software-in-the-loop model and the real time performance of the system was monitored. The findings of the research were validated through the analysis of the cosimulation results.