Highly time-resolved measurements of cylinder pressure acquired simultaneously from three pressure transducers were used to investigate the nature of knocking combustion and to identify biases that the pressure measurements induce.
Computer simulations of the wave equation were performed using the combustion chamber geometry in order to (1) understand the potential for bias in the result by the choice of transducer location, (2) investigate whether observed wave patterns could be explained (but we later figured out that was just a filtering and measurement artifact), and (3) choose the location to place the new transducer. Transducer 2 was installed at the identified location.
It was shown by investigating the magnitude squared coherence ( MSC) between the three transducer signals that frequency content above approximately 40 kHz did not arise from a common source, i.e. it was noise. The major source of noise at frequencies higher than 40 kHz is the natural frequency of the transducer that is excited by the impulsive knock event; even if the natural frequency is above the sampling frequency it can affect the measurements by aliasing. Furthermore, the knock-induced transducer noise was determined to be excessive such that the contaminated data lacks fidelity even with low pass filtering (digital or analog) even for moderate knock levels. Based on this analysis, it is recommended that 40 kHz is the appropriate cutoff frequency for low-pass filtering the pressure signal. While it is acknowledged that these results were acquired in a single engine, the principle source of noise that one needs to be concerned with is the transducer?s natural frequency and aliasing effects associated with it, so it is anticipated that the current results will give a good guideline for other engine geometries. Knowing this, one can isolate the knock event from noise more accurately. Window selection process for knock analysis (and more generally, combustion analysis) was performed and a Tukey window with 25% flat-top mid-segment is recommended.
The time series analysis of the pressure signals helped identify the contaminated portion of the pressure trace and provided insight into the nature of knocking combustion. Four time windows were identified for a knock event: (1) pressure rise due to flame propagation; (2) a rapid but resolved pressure rise that is ?20 to 50 ?s in duration; (3) a transducer shock period that is highly contaminated by noise that is estimated to last ?200 to 500 ?s; and (4) a sustained ringing period that is well resolved. Data during the transducer shock period needs to be eliminated from all measures of knock because they do not represent the cylinder pressure ? even when filtered. The commonly used maximum amplitude of pressure oscillation is heavily influenced by the transducer-induced noise.
New procedures for knock onset and knock intensity characterization are proposed. The knock intensity metric uses the exponential decay envelope of the sustained oscillations to estimate the magnitude of the initial knock event. This metric was shown to correlate well between the different transducers. The knock onset detection method does not depend on any explicit filtering method, instead it defines the KO time as the first occurrence of deviation from an extrapolation of the pre-knock pressure trace.
The evidence for presence of a developing-detonation event was investigated using the pressure rise duration for the transducers with the earliest and latest KO. The data suggests there is no evidence for a supersonic detonation-like event, because if such an event occurred within the pressure rise time of the earliest transducer, then the pressure rise of the latest transducer would be cut short by the detonation, which was not the case. Furthermore, the autoignition propagation velocity was estimated to be well below the speed of sound based on the pressure rise amplitude, the hot spot radius, and properties of burned and unburned gases.
Additionally, a comparison of pressure-based and accelerometer based knock intensity measurement was performed, primarily, using MSC as a tool in the frequency domain. It was shown that the interdependence between the pressure signal and the accelerometer signals increases with an increase in KI and that the MSC between pressure and acceleration differs significantly by accelerometer location. Therefore, MSC can be a guiding tool for accelerometer location optimization.