Modeling and Estimation for Dry Clutch Control

Increasing demands on comfort, performance, and fuel efficiency in vehicles lead to more complex transmission solutions. One such solution is the Automated Manual Transmission (AMT). It works just like an ordinary manual transmission but the clutch and gear selection are computer controlled. In this way high efficiency can be accomplished with increased comfort and performance. To be able to control and fully utilize an AMT it is of great importance to have knowledge about how torque is transmitted in the clutch. The transmitted torque in a slipping dry clutch is therefore studied in experiments with a heavy duty truck (HDT). It is shown that material expansion with temperature can explain torque variations up to 700 Nm for the same clutch actuator position. A dynamic clutch temperature model that can describe the torque variations is developed. The dynamic model is validated in experiments, and shown to reduce the error in transmitted torque from 7 % to 3 % of the maximum engine torque compared to a static model. The clutch model is extended with lock-up/break-a-part dynamics and an extra state describing wear. The former is done using a state machine and the latter using a slow random walk for a parameter corresponding to the clutch disc thickness. An observability analysis shows that the augmented model is fully or partially observable depending on the mode of operation. In particular, by measuring the actuator position the temperature states are observable, both during slipping of the clutch and when it is fully closed. An Extended Kalman Filter (EKF) was developed and evaluated on measurement data. The estimated states converged from poor initial values, enabling prediction of the translation of the torque transmissibility curve. The computational complexity of the EKF is low and it is thus suitable for real-time applications. The clutch model is also integrated into a driveline model capable of capturing vehicle shuffle (longitudinal speed oscillations). Parameters are estimated to fit an HDT and the complete model shows good agreement with data. It is used to show that the effect of thermal expansion, even for moderate temperatures, is significant in launch control applications. An alternative use of the driveline model is also investigated here. It is found that the amplitude discretization in production road-slope sensors can excite vehicle shuffle dynamics in the model, which is not present in the real vehicle. To overcome this problem road-slope information is analyzed and it is shown that a third-order butterworth low-pass filter can attenuate the vehicle shuffle, while the shape of the road profile is maintained. All experiments in the thesis are performed using production HDTs only, i.e. production sensors only. Since all modeling, parameter estimation, observer design and validation are performed with production sensors it is straight forward to implement the results in a production HDT following the presented methodology.