| Summary: | Experiments to determine the performance and characteristics of an internal combustion engine related to turbocharger is costly, time intense andcomplex. Besides, it is very difficult to develop an accurate mathematic regression model for engine performance because it involves many independent and dependent variables. Using artificial neural network (ANN), a machine learning model are used to predict the vehicle's engine performance parameters. ANN model aims to minimize the cost and time required for an engine tuning experiment. However, the experimental data from engine tuning experiment is usually small because of its high cost and time intensive. Therefore, this work intent to reduce the optimization process for an ANN model with a small dataset. Optimization of an ANN is critical to prevent the model trapped within a local optimum solution. Systematic methods utilizing modified grid search and random search are suggested in tuning the model's hyperparameters. Moreover, the importance of each optimization steps is ranked based on the analysis of the tuning result. The most optimizedmodel has achieved an overall mean absolute error (MAE) of 4.92% for 7 output parameters. Finally, future works are suggested that can be applied such as building a model for each of the output parameter.
|
|---|