A neural network based method is developed that can learn the underlying physics of hydraulic turbocharger (a radial pump coupled with a radial turbine) from a set of sparse experimental data and can predict the performance of a new turbocharger design for any given set of previously unseen operating conditions and geometric parameters. The novelty of the algorithm is that it learns the underlying physical mechanisms from a very sparse data spanning a broad range of flow rates and geometrical size brackets and uses these deeper common features recognized through a “mass-learning process” to predict the full performance curves for any given single geometry. The deep learning algorithm is able to accurately predict the key performance parameters like total efficiency of the turbocharger, its operating speed, pressure rise provided by the radial pump of the turbocharger and the shaft power produced by the radial turbine of the turbocharger for any given input combination of pump and turbine flow rates, differential pressure across the turbine and a limited set of geometrical parameters of pump and turbine impellers and volutes. Lastly, a novel method for fast inverse design of turbomachinery using a physics trained neural network and a constrained optimization algorithms is developed. The algorithm uses Nelder-Mead and Interior Point methods to find the global minimum of turbocharger design objective function in multi-dimensional space. The newly developed method is found to be very efficient in optimizing turbomachinery design problems with both equality and inequality constraints. The inverse design algorithm is able to successfully recommend an optimal combination of geometrical parameters like pump blade exit angle, pump impeller diameter, blade width, eye diameter, turbine nozzle diameter and rotational speed for a given target efficiency and head rise requirements. The preliminary results from this study indicate that it has a great potential to minimize the need for expensive 3D CFD based methods for the design of turbomachinery.

This content is only available via PDF.
You do not currently have access to this content.