Learning Based Approaches to Engine Mapping and Calibration Optimization

Abstract

In this chapter we consider a class of optimization problems arising in the process of automotive engine mapping and calibration. Fast optimization algorithms applicable to high fidelity simulation models or experimental engines can reduce the time, effort and costs required for calibration. Our approach to these problems is based on iterations between local model identification and calibration parameter (set-points and actuator settings) improvements based on the learned surrogate model. Several approaches to algorithm implementation are considered. In the first approach, the surrogate model is defined in a linear incremental form and its identification reduces to Jacobian Learning. Then quadratic programming is applied to adjust the calibration parameters. In the second approach, we consider a predictor-corrector algorithm that estimates the change in the minimizer based on changing operating conditions before correcting it. Case studies are described that illustrate the applications of algorithms. © Springer International Publishing Switzerland 2014.