Stochastic Policies for Online Computation Triggering in Powertrain Control

With the rapid growth in the amount of computations that need to be performed by modern electronic control units and in the complexity of the algorithms, there is a pressing need to develop approaches to reduce chronometric loading in automotive vehicles. This paper presents a cyberphysical systems framework for the development of stochastic optimal decision policies that trigger the computations online. For a specific case study, we consider online triggering of the computations involved in obtaining a linearized model in a setting when such a linearized model is required by control or estimation algorithms. The objective is to define a policy for triggering the linearization that balances the average model accuracy (or expected closed loop performance) with the average computational cost. The problem is formulated as a stochastic optimal control problem and solved using stochastic dynamic programming (SDP). The approach is described, then illustrated with three examples, a pendulum, a turbocharged diesel engine, and a turbocharged spark ignition engine, that illustrate the trade-off between the computational cost and expected linearized model accuracy.