Modeling, Control, and Adaptation for Shift Quality Control of Automatic Transmissions 2019-01-1129

The parameters determining shift quality control in automatic transmissions are determined as part of the calibration of the transmission control. The resulting control system typically has three components: feedforward control, where the control output is determined before a gearshift; feedback control, where the control output is determined during the gearshift based on sensed feedback; and learning control (adaptation), where the feedforward or feedback controller parameters are modified after the current gearshift has ended and before the next similar gearshift begins. Gearshifts involving the same ratio change are referred to here as similar gearshifts, though such gearshifts may involve differences in other variables such as vehicle speed or engine torque. In most automatic transmissions, gearshifts are controlled by hydraulic clutches, and operating conditions for these clutches may vary widely, requiring a dedicated transmission controller involving significant calibration effort.

In the current work, novel model-based methods are used to accomplish feedforward control of gearshifts, involving offline calibration of fill and torque phase control parameters and learning control of the fill phase. Towards this end, a physics-based model of the oncoming clutch involved in an upshift of a production automatic transmission was developed and experimentally validated against test bench experiments for a wide variety of inputs and operating conditions. The resulting model is used to generate a feedforward controller, offline model-based calibration algorithm, and a learning controller that corrects for clutch under-fill and over-fill. The effectiveness of the resulting controller is validated by simulation studies using the experimentally validated transmission hydraulic system model, in conjunction with a powertrain model. In particular, it is demonstrated that the learning controller corrects for initial under- or over-fill error in two to three gearshifts. Convergence and robustness properties, and transient performance of the learning controller are also discussed.