Model Predictive Control (MPC) is an enabling technology in applications requiring controlling physical processes in an optimized way under constraints on inputs and outputs. However, in MPC closed-loop performance is pushed to the limits only if the plant under control is accurately modeled; otherwise, robust architectures need to be employed, at the price of reduced performance due to worst-case conservative assumptions. In this paper, instead of adapting the controller to handle uncertainty, we adapt the learning procedure so that the prediction model is selected to provide the best closed-loop performance. More specifically, we apply for the first time the above “identification for control” rationale to hierarchical MPC using data-driven methods and Bayesian optimization.
30 May 2013 in International Journal of Adaptive Control and Signal Processing
In this study, a data-driven technique is proposed to deal with multivariable fixed-order controller design. The method is based on the virtual reference feedback tuning (VRFT) philosophy and thus does not require any model of the plant. Unlike standard VRFT for SISO systems, extended instrumental variables and variance weighting are used to counteract the effect of noise and achieve consistent controller estimate with a single set of input–output data. The proposed strategy is numerically compared to other existing direct and indirect techniques on a benchmark simulation example. The effectiveness of the method is finally tested on the airpath control of a real diesel engine.
The spread of active braking controllers on vehicles with significant mechanical differences and on low-cost products asks for control design approaches which offer easy and fast calibration and re-tuning capabilities. This task is made difficult by the use of model-based control approaches which heavily rely on specific vehicle dynamics descriptions. To address these issues, this brief paper proposes a data-driven approach to active braking control design, grounded on the virtual reference feedback tuning (VRFT) approach complemented with a data-driven nonlinear compensator. The effectiveness of the proposed approach is assessed both on a full-fledged multibody simulator and on a tire-in-the-loop experimental facility.
In electric vehicles with in-wheel motors, torque vectoring is an effective technology to reach high-level performance in sport driving conditions. To vary the torque at each wheel while guaranteeing proper safety constraints, Model Predictive Control (MPC) has been shown to be the most suited strategy. However, MPC requires simple models and significant tuning effort to be really effective. In this paper, we propose a simple predictive greybox model of the vehicle dynamics and an efficient calibration strategy to tune the MPC cost with little additional experimental effort. Results on a full-fledged multi-body simulator complete the paper.