Model predictive control of an all-wheel drive system considering vehicle dynamics change by clutch engagement state

Abstract

This dissertation introduces model-based AWD vehicle controller that supports electronic AWD system advancement. Details of the suggested model-based controller can be largely divided into two parts: design of the novel AWD vehicle states estimator that is not affected by the issue of model switching, and design of the model predictive control (MPC)-based AWD vehicle controller, which employs a tire force based full-car model.

Among several parameters of the vehicle, the uncertainty of clutch friction coefficient can significantly affect AWD vehicle states estimation performance. In order to reduce the parameter uncertainty, adaptation algorithm of clutch friction coefficient was designed using wheel dynamics equation and applied to the estimator. Unlike a general clutch system, the transfer case clutch system has a structure of single input shaft and multi output shafts. In a clutch system that includes multiple output shafts, the maximum allowable torque value transmitted to sub-drive shaft is determined by not only input torque but also other factors. Therefore, it is difficult to determine the exact state transition condition between slipping and lock-up state of the clutch in the absence of external information such as the difference of vertical load and road friction coefficient between front and rear wheels. Moreover, information of relative angular velocity difference between front and rear propeller shafts that can be obtained from wheel angular velocity sensor contains noise in its signal. These two limitations of transfer case clutch system may cause a discontinuity problem in designing estimator, due to the incorrect model switching between slipping state and lock-up state, which can, as a result, deteriorate the estimation performance. To solve the aforementioned issue of the transfer case clutch in AWD system, the interacting multiple model (IMM) filter algorithm, which simultaneously considers both clutch slipping and lock-up states based on stochastics of each model was adopted to AWD vehicle states estimator design. In the proposed estimator, only in-vehicle sensors were used for practical application. The estimation performance of the suggested AWD vehicle states estimator was validated through experiments. More specifically, the effect of adaption in switching multiple model (SMM) filter, the effect of IMM filter, and the combined effect of adaptation and IMM filter were validated orderly. In addition, robustness of the suggested estimator toward uncertainty of principal vehicle parameters was validated.

The bicycle model, which is mostly applied to the design of upper-level controller for vehicle control systems in the previous studies has limitations on applying vehicle upper-level controller because this model only includes the state variables for the lateral direction and does not have direct relationship between control input and state variables of the AWD system. Therefore, tire force based planar full car model, which includes both longitudinal and lateral states and represents explicit relationship between control input and states variables of AWD system, was newly adopted to the design of AWD vehicle upper-level controller. Because of the facts that vehicle dynamic behavior includes highly nonlinear characteristic and lower-level controller of the transfer case has highly lagged response, it is difficult to apply linear system control theory to AWD vehicle controller design. Also, conventional rule-based controller tends to operate AWD system preemptively to prevent unintended wheel slip in advance. Considering these facts, the MPC methodology, which shows preemptive response through the prediction of future states and computes optimal control input while considering system constraints systemically was adopted to AWD controller design. As for the input constraint for AWD system, variable input constraint was applied to the MPC-based controller using probability of each AWD vehicle dynamic model. As for states constraints for AWD system, upper and lower limit of yaw rate and longitudinal and lateral tire forces bounded by current peak road friction coefficient were considered. Simulations were conducted before the AWD vehicle experiments to verify the feasibility and prediction steps of suggested AWD vehicle controller. Then, the suggested AWD vehicle controller was validated through experiments that includes various driving scenarios and the superiority of the suggested controller was verified by comparative studies with the rule-based controller in the experiment.