Oil analysis is performed to monitor equipment and lubricant health. Alarm limits are used to trigger an activity for either the analyst or the maintenance staff. Alarm limits are threshold values, beyond which the measurement indicates a potential unhealthy machine or lubricant condition. Wear and contamination-related parameters reflect equipment health while physical and chemical properties of the oil reflect lubricant quality.
There are at least four different ways to determine the alarm status for oil analysis reports. Each method has advantages. Whether oil analysis is performed onsite or offsite, most analysts selectively use all four for the various measured parameters.
This method relies exclusively on human judgment and is based on experience and overall interpretation. Experienced analysts review all data at one time while considering the equipment and lubricant information supplied. Often, this information is sketchy at best, or even nonexistent, when the sample arrives at the lab.
Under these circumstances, the analyst tends to deduce information from the measured parameters to compensate for the missing information, which the customer failed to send along with the sample.
For instance, a sample that has a kinematic viscosity historically around 315 cSt at 40°C with high zinc and phosphorous, is likely to be an ISO VG 320 gear oil. Unless something in the analysis or label indicates differently, the analyst will probably make this assumption and continue.
The experienced analyst has the ability to fill in missing pieces of information using logic along with available data. He also identifies any data anomalies. He relies on the combination of measurements to draw a conclusion, considering, for example, neutralization number, color and viscosity at the same time rather than individually.
Some of this method’s advantages are flexibility, ability to make the best use of incomplete information and opportunity to raise and to resolve questions. This method is particularly useful in a lab that deals with various oils and applications. Possible disadvantages of this method are inconsistency (particularly with multiple analysts) and errors due to overlooked or misinterpreted information.
Sometimes industry standards or original equipment manufacturer (OEM) standard limits are published. These limits are normally statistically derived from a large number of samples or from a controlled evaluation program. Engine manufacturers and large fleet owners often recommend alarm limits for specific equipment.
When available, these limits are probably the best ones to use. Unfortunately this information is not available for the majority of lubricated machinery found in industrial plants. Also, variation in the way individual machines or groups of machines are operated and maintained can lead to poor conclusions.
In a few cases, industry standards organizations publish suggested limits for certain parameters. The National Fluid Power Association (NFPA) has published suggested ISO 4406 cleanliness levels for various types of equipment. Table 1 reports its recommended cleanliness codes (for example, alarm limits) for hydraulic fluid power applications.
This information is also available at http://www.nfpa.com. Rules of thumb are general guidelines or starting points for setting alarm limits. Three rules of thumb for determining when to change oil due to its chemical oxidation include:
Notice that a percentage drop in BN and an absolute increase in AN or dielectric is used to determine when to change oil. It is important to keep in mind that most alarm limits are set from a baseline that may be either zero or a finite number based on an earlier new or used oil sample measurement. The alarm limits may be percentages or absolute values. They may be based on an increase in value from the baseline, a decrease or both.
Some rules of thumb simply state that less is better. In the case of water contamination, Pall Corporation reports that whatever the moisture-in-oil may be, eliminating 75 percent of it can double the component’s relative life factor.1 British Hydromechanics Research Association (BHRA) studies make similar statements about particulate contamination.
Rules of thumb should be implemented with some flexibility. The rules make sense, yet alarm limits should have other supporting rationale.
If the OEM does not supply the statistical alarm limits, then the analyst should determine the limits. After three months, plenty of data for a good statistical baseline should be available.
Commercial software can generate a statistical histogram like the one shown in Figure 1.
This graph shows all the data measured on a single parameter for one alarm limit (for example, an equipment type). The plot shows minimum to maximum values measured on the X-axis compared to the percent of samples falling below that value on the Y-axis.
The software helps identify multiple alarm levels corresponding to predefined percentages. This method typically uses a series of alarms from low alert at the 80th percentile to extreme at the 98th percentile. These values should be recalculated each year. If the histogram-based alarm levels go down over time, overall reliability is improving.
Calculated standard deviation is another useful statistical alarm method. Given at least 30 measurements of a single parameter from an alarm limit, 1-sigma, 2-sigma and 3-sigma standard deviations can be calculated (See Alarms and Limits - Field-Tested Database Techniques on page 14).
For a one-tailed distribution, there is approximately a 15 percent chance that a new measurement will be 1-sigma above the mean, a 3 percent chance that it will be 2-sigma above, and much less than 1 percent chance that it will be 3-sigma above the mean for all measured values. The sigma-limits should be used in a manner similar to histogram percentiles to set statistical alarms.
Trend-based or rate-of-change limits are the mathematical method used intuitively by analysts who don’t use alarm limits. In this case, a departure from a trend instead of an absolute value to trigger the alarm should be identified. In his presentation at the Practicing Oil Analysis ‘99 Conference and Exhibition, “Interpreting Oil Analysis Test Results,” Jack Poley suggested three practical ways to automatically accomplish trend or rate-of-change alarms.2
This method looks for abrupt changes in consecutive historical data. A 60 percent increase from 5.0 to 8.0 will trip this alarm.
This is similar to relative magnitudes except that it compares the new measurement to an average of several historical measurements.
Factoring Delta Settings
Here, the problem with small values is addressed. This method requires very large percent increases to trigger rate-of-change alarms for small values (less than 5 percent) with diminishing percentages for larger magnitude values. Using this method, the increase to trigger the alarm might be 50 percent in the range of 50 to 80 but only 37 percent in the range of 200 to 400.
The most commonly used method for rate-of-change alarming is graphical trending. Normally this requires human interpretation, so it fits under the No Predefined Limits category. In this method the analyst looks for a knee, or bend in the trend plot. A hockey-stick plot, which is straight for a length of time and then sharply rises, is good when measuring financial growth, but it is symptomatic of a serious problem for most oil parameters.
There is not a schoolbook-approved solution for setting alarm limits on oil analysis parameters. The important thing to keep in mind is that every measurement reveals something about the machine wear, lubricant system contamination or oil chemistry. Alarm limits should be used to focus attention on the values indicating possible problems.
Alarm limits need not necessarily be set for every oil analysis parameter. Some numbers on the lab report are there because the measuring instrument puts out those numbers along with others, not because they are critical to the situation. It is important to focus on the critical few parameters that reveal possible problems.
If alarm limits are not used, then it is necessary to rely on the analyst’s experience. This works well if there is a lot of variation and only a few samples. However, a computer could provide valuable assistance to the analyst by always checking at least some of the alarms before looking at the data.
Industry standard alarms or rules of thumb are very useful, particularly when the OEM supplies them. Statistical methods including histogram and standard deviation-based alarming are excellent methods for setting alarms and measuring overall changes in asset reliability. Trend-based and rate-of-change alarms are also valuable techniques that software can use to flag possible problems for the analyst.
Probably the best approach uses software-based alarms to consistently check all the data. Then the analyst can resolve inconsistencies, draw conclusions and make recommendations.
This article was originally published in CSI’s OilView News, Bulletin #88.