Application of Multi-Scale CNN and Thermodynamic Model to Analyze the Airflow Features in Combustion Chamber of Internal Combustion Engine

The multi-scale features of airflow in combustion chamber are extracted by multi-scale CNN (Convolutional Neural Network), and the influence on combustion process is quantified by thermodynamic model. The purpose of this paper is to optimize the efficiency and emission of internal combustion engine. The features of micro-vortex, mesoscale turbulence, and large-scale circulation of airflow are extracted by designing parallel multi-scale network architecture, and the thermodynamic model is constructed based on finite volume method to quantify the airflow-combustion coupling process. The quantitative response relationship between airflow features and combustion performance is established by combining bidirectional LSTM (Long Short-Term Memory) and PSO (Particle Swarm Optimization) algorithms. The method has low error in extracting airflow features; the correlation coefficient of cross-operating condition prediction is above 0.85; the total effective power is increased by about 33.0%. The research results provide a new path for data-driven and physical model collaborative optimization for efficient combustion of internal combustion engines.