A Deviation-Based Centroid Displacement Method for Combustion Parameters Acquisition

Abstract

<div class="section abstract"><div class="htmlview paragraph">The absence of combustion information continues to be one of the key obstacles to the intelligent development of engines. Currently, the cost of integrating cylinder pressure sensors remains too high, prompting attention to methods for extracting combustion information from existing sensing data. Mean-value combustion models for engines are unable to capture changes of combustion parameters. Furthermore, the methods of reconstructing combustion information using sensor signals mainly depend on the working state of the sensors, and the reliability of reconstructed values is directly influenced by sensor malfunctions. Due to the concentration of operating conditions of hybrid vehicles, the reliability of priori calibration map has increased. Therefore, a combustion information reconstruction method based on priori calibration information and the fused feature deviations of existing sensing signals is proposed and named the "Deviation-based Centroid Displacement Method" (DCDM). The method based on priori calibration information, extract features of crankshaft transient angular velocity and knock signals. Using the parameter identification method, it acquires transient values of combustion parameters reconstructed based on various signal features. The fused deviation between transient values and calibration values is calculated using the Kalman filter and employed to adjust the priori values, realizing the computation of transient combustion parameters. A test platform for reconstructing combustion information is established in conjunction with an engine bench. The DCDM model is verified under 11 operating conditions, with the maximum error between the CA10, CA50 and CA90 computed by the DCDM model and experimental values being less than 2 °CA and the average error being less than 1 °CA, indicating high accuracy of the model. The Minkowski distance is less than 0.7, and the model distance is less than 0.3, demonstrating a good real-time performance and consistency of changes.</div></div>