The classification between a sequence of highly variable combustion events that have an underlying deterministic pattern and a sequence of combustion events with similar level of variability but random characteristics is important for control of combustion phasing. In the case of high cyclic variation (CV) with underlying deterministic patterns, it is possible to apply closed loop combustion control on a cyclic-basis with a fixed mean value, such as injection timing in homogeneous charge compression ignition (HCCI) or spark timing in spark ignition (SI) applications, to contract the CV. In the case of a random distribution, the high CV can be avoided by shifting operating conditions away from the unstable region via advancing or retarding the injection timing or the spark timing in the mean-sense.
Therefore, the focus of this paper is on the various methods of computing CA50 for analysing and classifying cycle-to-cycle variability. The assumptions made to establish fast and possibly on-line methods can alter the distribution of the calculated parameters from cycle-to-cycle, possibly leading to incorrect pattern interpretation and improper control action.
Finally, we apply a statistical technique named “permutation entropy” for the first time on classifying combustion patterns in HCCI and SI engine for varying operating condition. Then the various fast methods for computing CA50 feed the two statistical methods, permutation and the Shannon entropy, and their differences and similarities are highlighted.