Hybrid data-driven and mechanism-modeling approaches for online injection rate sensing of dual-fuel co-direct injector under carbon neutral background

This study proposes an innovative concept for online sensing of the injection process. By using the injector inlet as the pressure signal detection point, contactless testing of the injector is achieved, theoretically requiring no structural changes to commercial engines or standard injectors. For the pilot diesel injection, the pressure signal is abstracted as the evolution and transmission of Riemann waves, and the relationship between pressure and mass flow rate is established. Additionally, correction and decoupling methods for system interference and superimposed waves are proposed to extract Riemann single-wave components caused by injection. For the main fuel gas injection process, a data-driven injection mass neural network prediction model is constructed based on the pressure drop characteristic, which achieves quantitative sensing of the injection characteristics. Moreover, concerning gas pressure fluctuation, the interference phenomenon in the first-order differential pressure signal is identified. It is found that gas pressure waves propagate as second-order micro-pressure in the HPDI system, and the columns of pressure waves do not interfere with each other. A qualitative reconstruction method for the injection rate curve is developed based on 1D mechanistic models. In conclusion, this study uses hybrid datadriven and mechanism-modeling approaches to achieve online sensing. By comparing with offline testing methods, it exhibits high accuracy, with errors in the majority of injection parameters remaining within 5.5%.

Automated summary

This study proposes an innovative concept for online sensing of the injection process. By using the injector inlet as the pressure signal detection point, contactless testing of the injector is achieved. For the pilot diesel injection, the pressure signal is abstracted as the evolution and transmission of Riemann waves, and the relationship between pressure and mass flow rate is established. A data-driven injection mass neural network prediction model is constructed for the main fuel gas injection process, which achieves quantitative sensing of the injection characteristics.