Proposal of a methodology for designing engine operating variables using predicted NOx emissions based on deep neural networks

The process used by engine manufacturers for the development of a new engine includes the planning and conceptual design phases, followed by the detailed design phase, in which the design target specifications are met. In the conceptual design phase, a rough specification of the target engine is presented to facilitate a detailed design and the additional cost of modification is reduced exponentially. In the conceptual design phase, however, not only is there no real engine. but there are also no 1D and 3D models present, so it is impossible to test and simulate them. Therefore, at this stage, a model that can predict emission and performance only according to the specifications and operating conditions of the engine would be very useful. Previous studies developed an EGR prediction model that can be used in the 0-D NOx prediction using a deep learning method. In this study, a NOx prediction model with high accuracy using only the operating conditions as input variables, without ECU data, was developed using deep neural networks. The developed model has high accuracy with an R-square of 0.988. The feature of this model is that all the input parameters for the deep neural network come from the operating conditions of the engine. Therefore, this model can be used in the early stages of the development of new engines when testing and simulation cannot be performed because they do not exist. The designer can set the range of the operating conditions such that they do not exceed the NOx limits at the specific operating point (specific rpm and BMEP). This variable operating design methodology is expected to be useful in the development of new engines for automobile manufacturers with various engine data.