Variable Geometry Turbine Turbocharger Optimization Using Machine Learning for On-Highway Fuel Cell Applications

Abstract

<div>Turbocharger design involves adjustment of various geometric parameters to improve the performance and suit mechanical constraints, depending on the application-specific requirements. In designing the turbine stage, these parameters are optimized to maximize durability and efficiencies at the required operating points. For a heavy-duty class eight truck, “road load” and “rated power” are generally considered the two most important operating points. The objective of this article is to improve the efficiencies of these two operating points.</div> <div>The common challenge in the development of a turbine wheel design is the large number and interdependence of parameters to optimize. For example, increasing the blade thickness improves structural strength but reduces the mass flow capacity, thus influencing its performance. It is general practice to optimize the wheel geometry using iterative CFD analysis. However, running simulations for every single change in geometry involves significant computation time and does not guarantee the global optimum.</div> <div>This study proposes a machine learning (ML)-driven optimization process for variable turbine geometry (VTG) wheels with two target operating points. Initial CFD data is collected and used to train/validate a ML model for each operating point. A multi-objective optimization algorithm utilizes these ML models to provide a list of ideal turbine wheel designs, and the best candidates are validated in CFD for accuracy. The results are a balanced maximization in efficiency for both operating points, with an improvement in the pareto front by 1.4% points over CFD optimization alone.</div>