Title : Autoignition temperature prediction using machine learning
Abstract:
Fuels, serving as primary energy sources, encompass diverse hydrocarbons and organic compounds, with Autoignition Temperature (AiT) being a critical property indicating flammability. Accurate AiT measurement is vital for safe fuel handling. This report explores experimental methods for AiT determination, emphasizing liquid and gaseous substances. Factors influencing AiT in gas mixtures, including pressure and molecular structure, are discussed. The study delves into the phenomenon of autoignition, differentiating between "hot flame ignition" and "cold flame ignition." Additionally, alternative approaches, such as Quantitative Structure-Property Relationship (QSPR) models and Support Vector Regression (SVR), are explored for AiT prediction. A dataset of 204 pure organic compounds, categorized by functional groups, is utilized. Molecular Weight (MW) and Branching Index (BI) are key parameters. An Artificial Neural Network (ANN) model, employing the Levenberg-Marquardt algorithm, is developed. The ANN architecture comprises input, hidden, and output layers. Evaluation metrics include mean squared error (MSE) and correlation coefficient (R). The ANN models, trained with MW, BI, and functional groups, exhibit satisfactory accuracy, with the 20-layer model showing the highest correlation coefficient (R). The dataset's diversity is emphasized, suggesting the inclusion of mixtures and blends for improved accuracy. The study concludes that incorporating such complexities can enhance AiT prediction models.
Keywords: Autoignition Temperature, Artificial Neural Network, Fuel Safety, Branching Index, Levenberg-Marquardt algorithm
Audience Takeaway Notes:
- Enhancing safety protocols by accurately measuring and understanding AiT for different fuel compositions.
- Implementing experimental methods, especially in the assessment of liquid chemicals, for reliable AiT predictions.
- Addressing the complexities involved in determining AiT for gas mixtures, considering various influencing factors.
- Improving accuracy in AiT predictions, crucial for safe handling and storage of fuels.
- Offering insights into diverse fuel compositions, assisting in risk assessment and management.
- Providing alternative methodologies, such as QSPR and SVR, for AiT prediction, expanding the toolkit for researchers and industry practitioners.
