Automotive Turbocharger rotor optimization using Machine Learning Technique
2022-01-0245
03/29/2022
- Event
- Content
- Turbochargers are widely employed in in both, diesel and gasoline engine vehicles to boost the power output while retaining the fuel economy. From structural point of view, key component of an automotive turbocharger is rotor and primarily consists of compressor wheel, turbine wheel, shaft and bearing. Typically, design & development of turbocharger rotor is done using Computer Aided Engineering(CAE) and the dynamic characteristics are evaluated using Multibody dynamic (MBD) analysis. The current MBD approach solves the fluid-structure interaction problem by modelling oil film in the journal bearing in conjunction with modal transient analysis of the shaft. This process is quite complex and demands huge computational resources with simulations running for days. Over the period of last few years, lot of such simulations were conducted for different turbocharger applications and a huge database is created. Taking advantage of the availability of data, a deep learning-based approach is proposed to find a pattern or prediction of rotor dynamics characteristics. The model uses non-linear regression algorithm based functions with geometric and operating parameters of rotor as inputs to determine outputs like, bearing load and compressor and turbine tip displacement. This deep learning model augmented with a multi objective non-linear Optimization technique helps optimizing the rotor design parameters for the selected application. The genetic algorithm based optimization process determines the optimized rotor design parameters by minimizing the objective function(load and displacement of compressor and turbine) subjected to constraints like, oil film thickness and packaging limits. The overall process runs in few minutes resulting in significant cost and time saving in the turbocharger development cycle.
- Citation
- Shrivastava, S., Sinha, A., Ray, S., Du, I. et al., "Automotive Turbocharger rotor optimization using Machine Learning Technique ," SAE Technical Paper 2022-01-0245, 2022, .