Automating risk-aware calibration of internal combustion engines
Maarten Vlaswinkel defended his PhD thesis at the Department of Mechanical Engineering on September 11th.
The automotive industry strives to develop cleaner, more fuel-efficient, and more sustainable combustion engines to reduce the emission of harmful greenhouse gases and other pollutants in the transportation sector. However, innovative development methods are needed to minimize the increase in development time and costs for combustion engines that meet these sustainable demands. Maarten Vlaswinkel focuses on a development method to automate engine calibration within his PhD research.
Engine calibration is an important part of the development process of sustainable combustion engines. During this phase, a calibration engineer selects actuator settings and controller parameters for all possible combinations of torque and engine speed demands. The goal is to optimize the fuel-efficiency of the engine, while keeping emissions and other undesirable behaviors within regulator or design specifications. Automating this process could accelerate development and reduce costs, but it鈥檚 a complex task. That鈥檚 because calibration engineers develop an intuition regarding which settings are suitable and how far an engine can be pushed to achieve the desired results. Therefore, Maarten Vlaswinkel developed methods that can learn the engine鈥檚 behavior and predict which settings will yield desirable outcomes.
Learning and optimizing the engine鈥檚 behavior
The developed methodology relies solely on measured engine data. It鈥檚 based on a probabilistic model and optimization technique, and it captures the engine's behavior, while the optimization algorithm selects new settings that are likely to improve the fuel efficiency and unlikely to violate emission limits or cause other undesirable behaviors. The resulting automated calibration process is carried out iteratively, continuously learning more about the engine鈥檚 behavior and refining the selected settings.
Demonstrating the method
Vlaswinkel demonstrated this method on a simulated internal combustion engine. The results show that it successfully identifies the settings that maximize fuel efficiency, while consistently respecting emission limits and avoiding other undesirable behaviors throughout the calibration process. This approach represents a significant step toward the automation of engine calibration.
Title of PhD thesis: . Supervisor: Prof. Frank Willems and Dr. Duarte Guerreiro Tom茅 Antunes.