Applying a multivariable and adaptive control to a turbocharger test bench

Abstract

Turbochargers have been widely used in internal combustion engines to enhance their power output and decrease greenhouse gas emissions. To investigate turbochargers dynamics and their interaction with internal combustion engines, standardized experimental tests are usually performed to acquire their thermodynamic and rotordynamic behaviors. In our setup, which includes a fixed-size turbocharger with a combustion chamber simulating automotive engines, the turbine’s inlet temperature and the compressor speed must be controlled to achieve transient regimes. Therefore, we conducted experimental tests to demonstrate that our setup exhibits multiple-input, multiple-output (MIMO) system characteristics with two inputs and two outputs. Both the valves that regulate the amount of fuel powering the combustion chamber and control the air that speeds up the compressor affect the turbine’s inlet temperature and compressor speed. Additionally, turbochargers work at various operating points and may face distinct thermodynamic characteristics. Therefore, we applied steps to these valves for different compressor speeds, generating step responses with significant differences, indicating that adaptive control is also recommended. Thus, we proposed designing a multivariable controller to regulate the turbocharger speed and the exhaust air temperature simultaneously, in order to achieve enough accuracy to create turbocharger operating maps. Multivariable control design techniques were used to define the controller. We also proposed the use of a machine learning model that recognizes the plant’s operating point based on the turbocharger shaft speed data gathered during the tests. In this work, we compared the results of the single-input single-output (SISO) controller already in place at the plant with the controller obtained through the proposed methodology. We evaluated the step response through simulations for the turbocharger axis speed and the temperature at the turbine inlet, considering variations in the reference values of these variables. Results indicated that the controller from the methodology presented in this work can perform decoupling for the multivariable system and can be adjusted according to the plant’s operating point. This led to less oscillatory responses and smaller errors compared to the reference values. Compared to the SISO controller, MIMO yielded a significant reduction in speed errors of nearly 70% and 85% for operating points at 50 krpm and 115 krpm, and also presented lower temperature mean absolute errors for temperature setpoints.