Development of a tool for the preliminary design of large engine prechambers using machine learning approaches. (5th June 2021)

• Record Type: Journal Article
• Authors: Posch, S.; Winter, H.; Zelenka, J.; Pirker, G.; Wimmer, A.
• Is Part Of: Applied thermal engineering. Volume 191(2021)
• Publication Date: 2021-06-05
• Subjects: Machine learning -- Internal combustion engine -- 3D CFD -- Prechamber -- Large gas engine; Heat engineering -- Periodicals; Heating -- Equipment and supplies -- Periodicals; Periodicals; 621.40205
• DOI: 10.1016/j.applthermaleng.2021.116774
• Languages: English
• ISSNs: 1359-4311

Abstract

Highlights: Preliminary design of large engine prechambers using machine learning methods. Application and subsequent validation of several machine learning methods. Model trained by condensed CFD simulations data. Model satisfactorily predicts trends in prechamber behavior. Abstract: The design of internal combustion engine prechambers requires the consideration of a wide range of parameters. Especially in large gas engines, the design of the prechamber has a significant influence on the combustion process and thus on engine performance and emissions. Since the testing of design parameter combinations on engine test beds is both time-consuming and expensive and even complex 3D computational fluid dynamics (CFD) simulations of the high number of combinations are computationally expensive, this paper presents a novel approach, condensing complex CFD simulations of the prechamber and the engine combustion chamber into simple simulations. The short computational time of the simplified simulations permits the analysis of a wide range of prechamber parameter variations. Based on the simulation results, a model to predict the prechamber behavior is developed using machine learning methods. The parameters considered are maximum impulse, pressure difference between the prechamber and main combustion chamber and their associated crank angles, and NOx level. Several approaches are applied and subsequently validated their prediction accuracy is assessed. In the present case, an artificial neural network (ANN) performed best. The model is tested on geometry and equivalence ratio variations of a real large gas engine prechamber. A comparison of the values predicted with the trained ANN are compared to CFD results generated by a baseline simulation setup. The results indicate that the model is able to satisfactorily predict trends for impulse, pressure difference and NOx level. Absolute values of maximum impulse and pressure difference crank angle can be predicted with a deviation less than 10%.