Research On Model-Based Online Learning Algorithms Of Friction Torque For Diesel Engines

In the existing framework of torque based control, the control strategy calculate target indicating torque and required injection quantity according to the demand of the effective torque from the driver and estimated friction torque. As an important component of torque calculation, the accurate estimation of the friction torque is very important for the realization of effective accurate engine torque. But actual engine friction torque is a changing parameter during the life cycle of engine due to the mechanical wear, aging, and other issues, presenting obvious time-varying characteristics. But the traditional friction torque are usually estimated using look-up table method, of which the data was calibrated before the engine leaving the factory. In order to solve these problems, this study presents a model-based online learning algorithms of friction torque for diesel engines in order to accurately estimate engine friction torque throughout the life cycle of the engine.The proposed learning algorithm includes three parts: simplified basic friction torque model, friction torque observation algorithm and model parameter identification. Firstly, the basic friction torque model of rotational speed and oil temperature is established based on the friction torque mechanism, replacing friction torque of MAP in ECU. Through parameter calibration, the model can reflect engine friction torque characteristics under the whole working conditions, which is the foundation of the proposed learning algorithm. Then, the friction torque observation algorithm is proposed based on the idling and stopping condition. In the idle condition, the target indication torque in controller was used to obtain friction torque in the same speed and under different temperatures. In stopping condition, the relationship between friction torque and angular acceleration was used to establish stopping condition observation algorithm. Steepest tracking differentiator was adopted preferentially to estimate angular acceleration to obtain friction torque in different speeds and under the same temperature. Finally the results observed from idle and engine stop processes are used to correct the temperature coefficient and speed coefficient respectively through recursive least squares method, which realizes the online learning of engine friction.Algorithm in this study was proposed for the application in actual ECU control strategy. Basic software was developed on the platform of self-developed high-pressure common rail diesel engine controller(TJU-CR-ECU). Then online learning algorithms of engine friction torque were developed on this basis, and verified through HIL platform and bench test. The results show that the algorithm can be applied to the ECU online operation, realizing the accurate estimation of friction torque in controller. Through this algorithm, friction torque of updated model and data from drag test are basically the same, and the deviation is about 2~7Nm.