For many years, I've heard lab analysts advise their clients to not pay much attention to specific numerical data values (compared against a benchmark or limit) but focus instead on how the data is "trending". Indeed, trending is one of the most important and effective nomographic techniques used by diagnosticians to extract meaning and significance from time-based data. However, when this method is over applied or simply used alone, a large amount of critical information may go unnoticed or be wrongly dismissed.
What is Data Trending?
The best way to trend oil analysis data is to follow its movement visually using a standard trend plot (Figure 1). Trending can quickly reveal the rate-of-change over time (slope on the plot) associated with a series of monotonic data points that might reveal a reportable condition. It can sometimes be concluded that if the rate-of-change is normal and constant (linear trend slope) that the lubricant and machine conditions are equally normal and acceptable. However, abnormal or unhealthy conditions do not always produce steep trend lines.
It is true that the use of data trending (versus level limits) overcomes certain problems or complexities that have plagued the oil analysis field for years. This works best when all other variables are locked down such as:
samples taken from the same test port using the same method
oil and machine service life (in hours) are known
makeup rates are known
when machine operating conditions and environment are constant
oil type and formulation remain fixed
exact same laboratory test instrument and procedure are used
When this regiment is followed, then trending can correct for differences that are often outside of the control of the laboratory. For instance, when you use trend analysis and follow the six points above, the following trend-corrupting conditions would not occur.
Data reproducibility discrepancies between identical instruments. There is often data disagreement between two instruments of the same type used for the same procedure. This trend corruption might be noticed if you send samples randomly to different laboratories using the same instruments. No problem if the instrument used is always the same in the trended data series.
Differences in instrument types used by different laboratories. There are many diverse instrument types that target the same exact property. For instance, ferrous density can be measured by several different proprietary instruments, but the data produced (including the units of measure) are often not the same. No problem if the instrument type and brand are the same in the trended data series.
Differences in test procedure and calibration method between laboratories. ASTM procedures that report the same value often should not be used interchangeably. For instance, there are three different ASTM procedures for measuring base number. Additionally, many standardized methods are modified by laboratories to improve productivity and reduce cost. No problem if the same suitable test procedure is used in the trended data series.
Machine wear rate and operating environment differences. Identical machine types often produce sharply different wear rates, preventing the data between machines from having any mutual statistical significance in setting alarms. Some machines are "high readers" while others are "low readers" compared to statistical averages. Nominal readers, however, follow statistical averages for a machine group. No problem if the trended data is compared only to historical data from the same machine in the same operating environment.
Dangers of Relying Only on the Trend Line
Regardless of the many valuable features in using trend analysis, there are of course a few important caveats and key applications in which the approach does not apply well alone. These limitations can be corrected by coupling trending with several other data analysis strategies. Table 1 lists common data trending limitations and possible remedies.
Despite the mentioned limitations, trend analysis is an intrinsic part of data interpretation strategy in the analysis of in-service lubricants. When combined with other alarming tactics, it can recognize such things as bad samples, an oil filter going into bypass, additive depletion, a new forcing function (abnormal wear) and wrong oil in use. While computer software is extremely helpful, it is hard to improve upon the ability to detect discernable oil analysis trends by simply plotting the data graphically and using our eyes.
Table 1. Data Trending Limitations