Likelihood-Based Control of Engine Knock

Tuning and experimental evaluation of a likelihood-based engine knock controller

“…This was strengthened by the evaluation in [20] and will not be further treated here. Given the statistical independence of knock observations, the process can be completely characterized by the uni-dimensional probability density function (PDF) of the process.…”

“…The likelihood-based controller advocated in [20] addresses these issues by adjusting the spark angle only when there is sufficient statistical evidence that a control move is necessary. The magnitude of these adjustments is also scaled according to the likelihood of the observations that are actually recorded from the engine.…”

“…The reset cycle count is increased in three stages, after one, two and three spark advances with zero knock, so that the rate of spark advance increases until the first knock event occurs. With this slight modification, the algorithm presented in [19,20] becomes as shown in Algorithm 1.…”

“…5. The [20] during test at an engine speed of 1500 rpm, an intake pressure of 120 kPa and compression ratio of 12. parameter choice is the same as the final tuning in [20], p = 0.01, K Adv = 0.3, K Ret = 0.6, Likelihood threshold 0.4, and with reset cycle count after spark advance of 70, corresponding to k 0 = {0, 70, 70, 70} in Algorithm 1. Compared to the conventional controller the knock rate is slightly higher, 1.08% and the spark advance slightly more advanced by 0.14 degrees.…”

“…This controller reacts quickly to changes in knock rate while having less cyclic variability compared to the conventional controller, achieving a mean spark advance closer to the knock limit. An alternative stochastic controller using likelihood ratios was presented in [19], and further developed in [20], which has shown good performance in simulation.…”