Effects of Data Quality Reduction on Feedback Metrics for Advanced Combustion Control

Advances in engine controls and sensor technology are making advanced, direct, high-speed control of engine combustion more feasible. Control of combustion rate and phasing in low-temperature combustion regimes and active control of cyclic variability in dilute SI combustion are being pursued in laboratory environments with high-quality data acquisition systems, using metrics calculated from in-cylinder pressure. In order to implement these advanced combustion controls in production, lower-quality data will need to be tolerated even if indicated pressure sensors become available. This paper examines the effects of several data quality issues, including phase shifting (incorrect TDC location), reduced data resolution, pressure pegging errors, and random noise on calculated combustion metrics that are used for control feedback. Symbolic data analysis is an effective technique for identifying underlying patterns in noisy data, and has been applied to cyclic variability of dilute SI combustion, identifying deterministic effects that underlie the stochastic variations that are present. These techniques form a basis for current attempts to implement active next-cycle control of cyclic variability for dilution limit extension, and their effectiveness is evaluated in the presence of the reduced-quality data. The results indicate that IMEP and cumulative heat release are the most sensitive combustion metrics, showing the strongest trends with TDC offset and encoder resolution. While these combustion metrics show sensitivities to data quality issues, the symbol sequence statistics were always able to identify the correct deterministic trajectories in cycle-to-cycle variations, proving that production sensors provide data of sufficient quality for control methods based on symbol sequence statistics.