VVT Control

From autolab

Jump to: navigation, search
Predictive engine model
Predictive engine model
Sample fuel consumption map of an engine with VVT
Sample fuel consumption map of an engine with VVT

Project Title: Control of VVT Valvetrains - Development of Engine Controls and Calibrations for High Degree-of-Freedom Systems Using a Simulation based Methodology

Contents

[edit] Researchers

  • Dennis Assanis
  • Zoran Filipi
  • Tae-Kyung Lee

[edit] Sponsors

  • Chrysler Corporation

[edit] Background

Continuously growing demands in fuel economy, emissions and performance, are stimulating development and application of new powertrain technologies such as ETC, EGR, VVT, CVT etc. As a result, the number of independent control variables increases significantly. This makes the experiment-based calibration impractical, since the number of tests increases exponentially also.

[edit] Abstract

In this project, we propose to use simulation-based algorithm to reduce the dependence on hardware tests, and to ultimately reduce the time and cost of developing controls and calibrations for a new powertrain design. The methodology uses a high-fidelity engine simulation tool for developing a series of neural network surrogate models that are simpler and faster. Finally, the surrogate models are exploited in an optimization framework for determining the best combination of control settings for any given operating conditions.

[edit] Scope of work

  • A high-fidelity simulation tool is developed based on the commercial gas-dynamics code (WAVE) and an in-house combustion model based on physical principles (SIS). This co-simulation tool is capable of predicting fuel consumption, torque generation and pollutant emissions as function of actuator set-points.
  • A 2.4L DCX SI engine has been set up at the W.E. Lay Automotive Laboratory, University of Michigan, in order to provide experimental measurements for model calibration and validation.
  • Depending on driver command, engine calibration tasks are formulated as a series of nonlinear optimization problems. However, conducting optimization directly on the high-fidelity simulation tool is not practical due to prohibitively high computation cost. Instead, we propose to use artificial neural networks (ANN) as surrogate models. Full simulations are performed for representative operating points, and results are used to train ANNs. Then, optimization is efficiently conducted on fast ANN models.
  • Initially, the methodology has been demonstrated on a conventional four-cylinder SI engine, while on-going studies include cases with increased number of control variables, such as independent intake and exhaust cam phasing.
Personal tools