Engine combustion modeling method based on hybrid drive.

Area of Science:

• Mechanical Engineering
• Combustion Science
• Data Science

Background:

• Accurate reconstruction of in-cylinder combustion is crucial for engine monitoring.
• Traditional zero-dimensional (0-D) models face challenges in heat release rate calibration.
• Integrating big data offers potential for improved combustion prediction.

Purpose of the Study:

• To develop an enhanced 0-D physical model for accurate in-cylinder combustion reconstruction.
• To improve the calibration of the heat release rate in combustion models.
• To leverage deep learning for real-time combustion process fitting and optimization.

Main Methods:

• An improved 0-D physical model incorporating cumulative fuel mass and in-cylinder temperature was developed.
• A deep neural network, Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU), was designed for parameter calibration.
• The physical model was combined with the CNN-GRU network to reconstruct the combustion process.

Main Results:

• The improved 0-D model effectively reflects the heat release law.
• The CNN-GRU model achieved a Root Mean Square Error (RMSE) of 0.5862 for peak firing pressure (PFP) under non-calibration conditions.
• The integrated model demonstrated accuracy in simplifying and reconstructing the combustion process.

Conclusions:

• The developed 0-D physical model with an improved cumulative fuel mass approach is effective for heat release law representation.
• The CNN-GRU based prediction model shows promise for online fitting and optimization of the combustion process.
• This integrated approach offers a viable solution for timely monitoring of engine combustion states.