Automotive Turbocharger Rotor Optimization Using Machine Learning Technique 2022-01-0216
Turbochargers are widely employed in internal combustion engines, in both, diesel and gasoline vehicle, to boost the power without any extra fuel usage. Turbocharger comes in different sizes based upon the boost pressure to increase. Capacity of turbocharger are available in great range in the market which are designed to match the requirement. From structural point of view, key component of an automotive turbocharger is rotor. This rotor consists of compressor wheel, turbine wheel, shaft and bearing (journal/ball) mainly. In industries, design & development of turbocharger rotor for its dynamic characteristics is done using virtual engineering technique (Computer Aided Engineering). Multibody dynamic (MBD) analysis simulation is one of the best approaches which is used to study the rotor in great details. In this current MBD procedure fluid-structure interaction problem is solved by modelling oil film in the journal bearing and solving it using “Reynolds equation”. Shaft displacement is provided to oil film which eventually output the pressure development in the bearing. This pressure then acts upon the shaft and modal transient analysis is performed for the structure analysis.
As current approach is a quite complex, time require to complete the simulation is in days. Multiple simulation is required to study the design sensitivity and reach to an optimum design of turbocharger rotor. So, any essential design study takes this huge time to carry out, hence it is one of the challenges in the development cycle.
Over the period of last few years, a lot of such simulation has happened for variety of turbocharger. It provides a huge data set which could be utilized to find a pattern or prediction of rotor dynamic characteristics. One of such effort has been made and a deep learning-based rotor dynamics model has been developed. This model further with non-linear Optimization technique is very promising for optimizing the rotor design parameters. This method not only predicts the outcomes with given design parameter but also minimize the outcomes by providing the most optimized set of design parameters within hours.