Detection of difference of fuel composition and combustion process in a spark ignition engine using anomaly score based on principal component analysis of ion current signal

抄録

Use of alternative fuels, such as e-fuel , will be more common in automobile industries. However, diversity in fuel composition influences combustion characteristics of an engine, requiring highly-robust control for stable and high-efficient operation. For that purpose, real-time monitoring of combustion is necessary. Ion current sensor has attracted attention for its low cost and potential in catching information about combustion. The objective of this study is to extract information about fuel composition from ion current signal. Experiments were conducted on a gasoline engine bench with a spark plug modified to detect ion current. To simulate the diversity in fuel composition, different flow rates of CH₄ and CO₂ gas were introduced into intake manifold while gasoline was directly injected into cylinders. To analyze ion current data, which is characterized by noise and high cycle-to-cycle variation, principal component analysis (PCA) was employed. Principal components were extracted from the condition under the base fuel setting, and ion current history for other fuel settings were reconstructed using the principal components. To quantify the deviation in ion current signal caused by fuel variation, anomaly score was defined as the reconstruction error. Results showed that as gasoline injection amount decreased, CH₄ and CO₂ flow rates increased or equivalence ratio deviated from stoichiometric condition, anomaly scores tended to be higher. This demonstrates that the conditions deviating further from the base setting lead to high anomaly scores. Moreover, anomaly scores were higher under conditions with CO₂ introduction than those with CH₄, which implies that inactive gas has larger impact on ion current behavior than combustible gas. Finally, to evaluate the effect of calculation process, calculation sections of mean anomaly scores and number of principal components used for reconstruction were varied. It was clarified that adjustment of calculation process could have large effect on anomaly scores.