Vehicle integrated stability control using hybrid fuzzy cmean clustering - Adaptive back propagation scheme

Most vehicles accident were caused by instability vehicle motion. The instability just occurs cause four former integration controls (Feed-forward control, H ^∞ control, Nonlinear Predictive control, Robust control) can not adapt to driving condition (vehicle, drivers character and environment), which always change their structure and parameter at high speed. This obstacle results the controlled variable of stability such as Yaw-Rate (YR) , Vehicle Side Slip (VSS) , Roll Angle (RA) cannot fulfill control targets, instability vehicle direction and then cause accident. This paper propose a new integration control design exploits combined Multi Dimension Fuzzy C-Mean Clustering (MDFC) and Adaptive Back-propagation Control (ABC). ABC consist of NN-Plant and NN-Controller. Architecture NN-Plant results from genetically optimized hybrid fuzzy neural network (gHFNN) while NN-Controller from multi-layer neural network (MLN) with single hidden layer. Instead of three former vehicle dynamics model like decoupling of linear to nonlinear plant, two dimension to three dimension plant and ESP-4WS-AS plant, which are imprecise to build a driving condition model, will be proposed a "Three in one dynamics system (TODS)" plant, which can represents the best model of interaction among vehicle dynamics, driver characters and environment. The solving methodology are arrange like follow, First step vary 2.187.000 TODS real time data's is realized by test drive a vehicle equipped with electronics stability program(ESP), four wheel steering (4 WS) and active suspension (AS), which covers 6 dimensions vector (YR from yaw-rate sensor, VSS from lateral acceleration sensor, RA from body level sensor, TDYC from ESP actuator, δr from rear steering actuator, M _AS from suspension actuator). Data's inputs to MDFC to cluster 810 centers. Second step is the training process to update the optimized architecture and parameters of NN-Plant uses all centers based on genetic algorithm (GA), LSE and BP. Third step is the training process to update optimized NN-Controller's architecture and parameters uses input reference and desired input of updated NN-Plant based on constructive back propagation (CBP). Fourth step is validating and testing of ABC use all data's of TODS. An experiment and simulation is completely setup to prove the performance of Hybrid MDFC-ABC integrated control, when is compared with four former integrated control method to control TODS. The simulation result in the form of rank shows that topmost sequence performance is Hybrid MDFC-ABC, then robust control, H-infinite control, NLPC, No-integration control and feed-forward control.