Study on predictions of spray target position of gasoline direct injection injectors with multi-hole using physical model and machine learning

Abstract

A lot of effort is being invested in improving the performance of injectors, the core components of gasoline direct injection (GDI) engines, such as injection stability and accuracy. The purpose of this study is to establish models that can predict spray targeting according to the design parameters of GDI injector, to improve the injection accuracy and enhance the engine performance. First, this study used laser sheet beam visualization technology to measure the spray targeting images of injectors with different design parameters in different cross-sections and obtained the spray targeting coordinates through image post-processing. Then, using the experimental data, two different approaches (empirical formula and machine learning), were used to create models for predicting spray targeting, and their applicability was compared. The research results showed that both the physical model and the machine learning model had a prediction accuracy of >0.98 in terms of R2, but the physical model had a lower prediction error in terms of the root mean square error (RMSE). Further, the tendency of the target coordinate to change is proportional to 0.2 power of the injection pressure (Pinj0.2), −0.1 power of ratio of hole length to hole diameter ((L/D)−0.1), and − 1.5 power of the angle between axes of two holes (θij−1.5).