This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Machine Learning-Based Turbine Vane Position Estimation for Advanced Engine Airpath Control
ISSN: 1946-3936, e-ISSN: 1946-3944
Published July 29, 2021 by SAE International in United States
Citation: Kamath, R., Venkobarao, V., Kopold, R., and Subramaniam, C., "Machine Learning-Based Turbine Vane Position Estimation for Advanced Engine Airpath Control," SAE Int. J. Engines 14(6):833-851, 2021, https://doi.org/10.4271/03-14-06-0050.
In an engine airpath system, modeling the nonlinearities associated with turbocharger vane position estimation is challenging as most influencing factors are not clearly known. State-of-the-art models are predominantly data driven, which makes the accuracy skewed toward the clusters with data richness. The data for these models are derived from the experiments in a dynamometer (dyno) or vehicle. The measurement data is extrapolated for the entire operating zone to derive the control models, which are inaccurate and slow in transient response. With emission norms becoming stringent, the models need to be more accurate with an improved transient response.
The authors derive the synthetic data using “adaptive synthetic data generation” for the one engine operating zone. This improves the data richness of the dataset, which can be used for data-driven models, which is state of the art. The authors propose artificial intelligence (AI)-based model instead of data-driven models to improve the transient response of the system. The study shows that the use of AI models reduces the complexity of modeling and predicting turbine vane position without compromising the accuracy during transients and stability during steady states. The AI-based models are derived using supervised and unsupervised learning techniques. The most influencing parameters are extracted using feature extraction methods hybridized with expert decision-making. The AI-based model with an electromechanical time constant of the turbocharger was modeled using a “nonlinear autoregressive network with exogenous inputs” (NARX) and deep neural networks (DNN) methods. The comparative study with respect to the accuracy of both the models is done with “Worldwide Harmonized Light Vehicles Test Cycle” (WLTC) and “Real Driving Emission” (RDE) tests.