ASTM E2412-10 (2018) pdf free.Standard Practice for Condition Monitoring of In-Service Lubricants by Trend Analysis Using Fourier Transfor.
Distribution plots are a common tool used in thestatistical analysis of condition monitoring data.The abscissaof the plot is the test data result，and the ordinate is thefrequency at which a specific result occurs in the test population. Fig.A3.1 shows an example distribution plot. In thisexample, the data is from a population of 1910 diesel engineoils.The abscissa represents the integrated nitration result.Avertical bar is used to represent the number of oils for which the test result falls between the two abscissa values. Forexample,the tallest bar indicates that approximately 410nitration values were between 7 A/cm and 8 A/cm. To develop reliable alarm limits from statisticalanalysis of condition monitoring test data,the data must beapproximately normally distributed. The histogram shouldhave roughly a bell-shaped appearance and be free of multi-modal features. The histogram in Fig.A3.1 shows an approxi-mate normal distribution.When the FT-IR results are limited to non-negativevalues, and the median of the distribution is close to zero, thedistribution will not appear normal (see for example Fig.A3.2).While a mean and standard deviation can still be calculated, theuser should verify that alarm limits based on these statistics aredescriptive of the actual distribution. For example, only about5 % of the values should fall above the mean plus two standarddeviations.
Multimodal distributions (Fig.A3.3) and broad,flatdistributions (Fig.A3.4) should not be utilized for statisticalanalysis. Both examples are indicative of multiple sources ofthe same data,low ratio of normal data to failure data or poormeasurement precision.
Random selection of samples does not necessarily provide for normally distributed results. Sequential samples over the course of the overhaul period are more likely to yield normally distributed results.The population analyzed must include examples of all oilrelated failure modes, at typical failure rates. If too many examples of failures are included, the distribution may be broad and calculated limits may be too high. Alternatively, if failure modes are underrepresented, the distribution may be narrow, and the calculated limits may be too low.ASTM E2412 pdf download.