Application of Reactive Response Surface Machine Learning System to Multi-Point Optimization of a Centrifugal Pump Impeller using ADT’s TURBOdesign and Simcenter STAR-CCM+

Application of Reactive Response Surface Machine Learning System to Multi-Point Optimization of a Centrifugal Pump Impeller using ADT’s TURBOdesign and Simcenter STAR-CCM+

In today’s aerospace business, there are tougher requirements than ever before, and one must use the latest in simulation technology to meet those requirements as well as beat competition. This includes several components in the air as well as on the ground that can be improved. From the propulsion of the jet engine to fans for cooling electrical components and HVAC systems, to fuel and water pumps.

Machine learning and AI can play a major role in helping to achieve optimal design and increased productivity of turbomachinery applications, especially when it comes to hard to solve problems which mainly involve multi-objective, multi-point or multi-disciplinary optimization. 

But application of machine learning (ML) in design optimization, where each data set evaluation is computationally very expensive, is only feasible if accurate ML models can be generated with very small data sets. To achieve this objective, three key areas need to be addressed:

1. Improvement in machine learning models that enable high accuracy with small data sets.
2. Improvement in parametrization of the blade geometry, a key long neglected aspect to achieve high accuracy small data set models.
3. Improvement in automation in evaluating key performance parameters
   
   

In the TURBOdesign Suite 2024.2 release, all the above three functionalities were addressed by coupling 3D inverse design parameterization with unique functionalities of the new Reactive Response Surface (RRS) and automatic CFD integration with Simcenter STAR-CCM+, enabling an important foundation for application of machine learning in turbomachinery design.

In this webinar, we will show the application of the new RRS with automatic Simcenter STAR-CCM+ integration to multiple operating point optimization of a centrifugal pump impeller. We will show in the details the setting up of the case in 3D inverse design method and  automatic CFD calculation in Simcenter STAR-CCM+ for multiple operating points.

Furthermore the set up of input parameters including meridional shape and blade loading parameters are described. The results of the optimization are presented and also the process of verification of the results and its accuracy is demonstrated.