Machine-Learning Approach to Behavioral Identification of Hybrid Propulsion System and Component 2022-01-0229

Accurate determination of driveshaft torque is desired for robust control, calibration, and diagnosis of propulsion system behaviors. The real-time knowledge of driveshaft torque is also valuable for vehicle motion controls. However, online identification of driveshaft torque is difficult during transient drive conditions because of its coupling with vehicle mass, road grade, and drive resistance as well as the presence of numerous noise factors. A physical torque sensor such as a strain-gauge or magneto-elastic type is considered impractical for volume production vehicles because of packaging requirements, unit cost, and manufacturing investment. This paper describes a novel online method, referred to as Virtual Torque Sensor (VTS), for estimating driveshaft torque based on Machine-Learning (ML) approach. VTS maps a signal from Inertial Measurement Unit (IMU) and vehicle speed to driveshaft torque. The unique advantage is that VTS does not explicitly rely on the first principles unlike other estimation methods. A robust mapping framework implicitly accounts for road grade, while compensating the effects of vehicle mass and drive resistance. Mapping coefficients are automatically and adaptively learned during selective drive conditions and continuously updated by means of Kalman filtering. VTS is implemented in a test vehicle with a P2 hybrid electric propulsion system for the assessment of robustness and sensitivity to drive conditions. The accurate estimate of driveshaft torque from VTS is utilized to determine the characteristics of a wet clutch which is employed for cranking an internal combustion engine during EV-HEV mode transition. VTS demonstrates a ML-based data-driven solution to the accurate determination of driveshaft torque and wet clutch behaviors. VTS framework can be readily extended to broader applications, including battery electric vehicle, with additional capabilities such as wheel torque estimation during braking and steering.