Identifying low-speed pre-ignition events
An abnormal combustion phenomenon, known as low-speed pre-ignition (LSPI), has arisen from the use of downsized gasoline engines to improve fuel economy and to comply with global CO2 legislation. In this fourth feature on LSPI, Anne Young, Infineum Lubricant Development Technologist, talks about the statistical methodology developed for the ILSAC GF-6 Ford engine to more accurately identify LSPI events.
Engine downsizing is gaining popularity with OEMs as an effective method to improve fuel economy and reduce CO2 emissions. However, increased boost pressure is needed to compensate for the lost power and torque output from the reduced engine displacement. This has led to the occurrence of a disruptive abnormal combustion event, known as LSPI. This phenomenon occurs prior to the spark being triggered and is often followed by heavy knock that can lead to severe engine damage.
Understanding the mechanisms and causes of LSPI is a key enabler in the continued use of smaller, boosted engines to achieve future fuel economy and emissions requirements.
Lubricant and fuel properties have been implicated as significant influencers of LSPI frequency and intensity. Data collection and rigorous statistical analysis are needed to accurately quantify the effects of these properties. The upcoming ILSAC GF-6 specification will include an LSPI test using a Ford engine. Building on its previous experience with a General Motors (GM) engine, Infineum developed a new statistical method to identify LSPI events in collaboration with Ford. These efforts have yielded a versatile statistical toolset that can be easily used to analyse engines from a broad range of OEMs, without requiring advanced statistical programs and experience.
The type and quality of fuels and lubricants have been found to influence the occurrence of LSPI events. A previous article in the Insight LSPI series reported on the work being carried out by Infineum, using a stationary GM engine, to examine the impact of different formulations. This work implemented a methodology for studying LSPI, and used a rigorous statistical approach to assess the data. Using this method, consistent results were obtained from the GM engine. Click here to read the full article.
In addition to the GM engine test, a stationary Ford engine test has been developed for inclusion in the ILSAC GF-6 engine oil standard.
This was necessary because the methods, which had proven to be effective in the GM engine, identified engine cycles as having LSPI events that did not physically appear to be LSPI. Investigation indicated that the cause was due to the parameter distributions departing from a normal distribution. Consequently, a new method was required to quantify the symmetry (skewness) and flatness (kurtosis) of the distribution in the Ford engine test. This was then used to expand on the methodology and further explore the statistical analysis methods, enabling a better understanding of the degree to which the lubricant can affect LSPI.
LSPI events are defined as outliers of peak pressure (PP) and crank angle location of 2% mass fraction burned (MFB02) data. The number of standard deviations beyond which an event can be considered an outlier are calculated and these outliers are identified as possible LSPI events.
The test procedure used to identify LSPI events in the GM engine included six segments of 25,000 engine cycles at constant speed and load conditions. Outliers for PP and MFB02 metrics were identified as possible LSPI events. Determination of these outliers was an iterative process of calculating the mean and standard deviation of the two metrics for each segment in each cylinder of the engine and determining cycles with parameters exceeding n standard deviations from the mean. If outliers existed, those outliers were omitted and the process was repeated; otherwise, the process was deemed complete. The number of standard deviations, n, which was used as a limit for determining outliers was calculated to be 4.7 using Grubbs’ test for outliers for 25,000 cycles.
The Ford procedure includes four individual iterations of 170,000 engine cycles each. Initially, the only modification to the GM procedure for analysing the Ford engine data was to utilise PP and MFB02 trigger points of five standard deviations.
A consistent observation of the LSPI cycles in early testing with the Ford engine, (not observed in the GM engine) was the occurrence of events within a few tenths of a standard deviation beyond the trigger points for both metrics. This was especially the case when considering the parameters individually. It is possible that these are real LSPI events, but it is more likely that the distribution is not sufficiently characterised by a normal distribution curve.
Some results that are not abnormal may be flagged as outliers (false positive) and other true outliers may not be identified (false negative).
Reducing false results
By applying more rigorous statistical methods to the data from the Ford test, the departure from normality could be identified. Fleishman’s cubic transformation method was developed to transform normally distributed random numbers (or other distributions) into a distribution with the desired skewness and kurtosis. This method was applied to estimate trigger levels for PP and MFB02 distributions with skewness and kurtosis with non-normal distribution. Of four possible solutions, three did not give meaningful results. The remaining solution had a monotonically increasing region that corresponded to the operating conditions of the Ford engine test.
This method had two limits to its robustness of application. Firstly, it entailed simultaneously solving three equations, which added complexity, and in extreme cases, resulted in regions in which a legitimate solution could not be obtained. Secondly, the method itself had limitations in its use. Fleishman states that solutions for the entire skewness and kurtosis space cannot be obtained using this method for large departures from normal operation, such as LSPI events and extraneous readings due to failed transducers. Therefore, steps were added to remove all obvious sensor errors and omit any obvious LSPI cycles.
The principles of this method can also be used to reduce the frequency of false positives and negatives in other cases of departure from normality. While the impact of using this method varied, in general the reduction was in the order of 10% to 20%.
In our view, this is an exciting development since this method of analysis is not exclusive to PP and MFB02, but can be applied to any parameter that departs from normality in terms of skewness or kurtosis.
In the upcoming ILSAC GF-6 engine oil standard, the Ford engine test will apply this statistical analysis method to more accurately identify LSPI events.
Note: This work was initiated after the selection of parameters and is not necessarily an endorsement of PP and MFB02.
Reprinted with Permission from SAE International
Original paper: Controlling Low-Speed Pre-ignition in Modern Automotive Equipment: Defining Approaches to and Methods for Analyzing Data in New Studies of Lubricant and Fuel-Related Effects (Part 2); 2016-01-0716 copyright 2016.
Visit the SAE website to access the paper here