Browsing by Author "Liu, Licheng"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Predicting the homogeneous ignition delay time of gasoline fuel blends using machine learning
    (2023-10-09) Liu, Licheng; Toldy, Arpad; Insinööritieteiden korkeakoulu; Santasalo-Aarnio, Annukka
    The transportation sector contributes 25% of energy consumption worldwide in 2022. Some optimization approaches and strategies in the transportation sector need to be done to decrease carbon emissions globally. According to EU Commission, electric vehicles will replace CO2-emitting cars by 2035 in Europe, while during the vehicles transition period, traditional vehicles will dominate the market for a certain amount of time. Therefore, optimizing the combustion engine and finding renewable gasoline fuel surrogates have become essential in recent years. The ignition delay time (IDT) is a crucial parameter for optimizing internal combustion engines, particularly from the energy efficiency perspective and environmental impacts. Investigation of IDT can contribute to the selection of gasoline fuel blends and the development of renewable gasoline fuel surrogates. However, the ignition delay time presents challenges due to its multiple magnitudes range, non-linear relationships, and negative temperature coefficient region. Additionally, chemical kinetic simulations demand significant computational resources due to complex reactions and components. This study aims to use machine learning to predict the IDT of toluene primary reference fuel. The model developed in this study will overcome the limitations of the experimental method and chemical kinetic simulation as an alternative. Two linear sequence and Sobol sequence datasets were constructed based on the chemical kinetic simulation data obtained from Cantera. This study validated, trained, and tested four machine learning algorithms, including random forest, extra tree, support vector regressor, and gradient boosting. The results show that using the Sobol sequence with better sample filling properties, the gradient boosting model is the most accurate and effective model to predict the IDT of toluene primary reference fuel. Notably, the validation comparison results among experimental values from other literature, predictions in this machine learning model, and simulation results from the chemical kinetic model indicate that the main challenge to predict IDT lies in the chemical kinetic mechanism limitations rather than machine learning approaches weakness in this study.