Akademik Veri Yönetim Sistemi
INTERNATIONAL JOURNAL OF ENGINE RESEARCH, cilt.24, sa.2, ss.567-578, 2023 (SCI-Expanded)
This study aims to determine the optimal injection strategy by predicting the performance and exhaust emission parameters of a four-cylinder CRDI engine under several operating conditions. The experimental determination procedure is challenging and expensive calibration task since it requires a high number of tests. Many studies have focused on a limited level of parameters. In this study, design of experiments technique and deep neural network (DNN) modeling are used together. The experimental data set for the model is created using Taguchi L16 and L32 orthogonal arrays. The DNN model is developed to predict BSFC , N O x , HC, and CO emissions with speed, torque, injection timings and fuel quantities of each injection called as pilot1, pilot2, main, and post. In this way, it has become possible to evaluate the effects of a larger number of operating parameters in a wide range than the literature. The developed DNN model predicts the BSFC , N O-x , HC, and CO with R-2 0.939, 0.943, 0.963, and 0.966, respectively. Additionally, RMSE and MAE values for the model are between 0.024 and 0.048. The proposed method compared with the conventional look-up table method performs better in reducing the complexity and cost of experiments and exploration of the effects of injection parameters on engine emission and performance characteristics in a wide engine operating range. In conclusion, until 2300 rpm at specified torque (90 Nm), it is found that 70% of fuel quantity should inject in main injection to minimize BSFC and NOx emissions. The post injection quantity should be increased by reducing the amount of main injection from this operating condition on. Furthermore, it is observed that the ratios of pilot injection durations do not change with increasing engine speed, but quantity of first pilot injection is more than that of second pilot injection.