Physics-Informed Data-Driven Modeling for Engine Volumetric Efficiency Estimation

作     者：Qian Li Fan Guo Kang Song Hui Xie Shengkai Zhou Hailang Sang 

作者机构：State Key Laboratory of Engines Tianjin University Tjianjin 300200 China Guangxi Yuchai Machinery Group Co. Ltd. 

出 版 物：《IFAC-PapersOnLine》 

年 卷 期：2024年第58卷第29期

页      面：403-408页

主　　题：Volumetric Efficiency Estimation Physics-Informed Data-Driven Model Feedforward Neural Network Physical Model 

摘      要：Accurate volumetric efficiency modeling is crucial for enhancing engine performance regarding fuel consumption and emissions, but it is challenging due to the variability of the intake process and valve strategies. Therefore, this paper proposes a physics-informed data-driven volumetric efficiency modeling method (PDM). Firstly, this paper constructs a model based on the simplified first law of physics to capture the main trends of volumetric efficiency changes. To improve the accuracy of the estimation, a PDM is proposed. This method includes a physical loss term and a data loss term. These loss terms are fused into a single fusion loss to train the neural network parameters, effectively merging the physical model with the neural network. The high correlation coefficient (R 2 = 0.958) between the PDM s volumetric efficiency estimates and the measured data demonstrates the robustness of the method.