Complexity in ISFA (in-service fluid analysis): Part XXIII

Jack Poley | TLT On Condition Monitoring July 2015

A good database for your ISFA program is the gift that keeps on giving.

 

IF YOU (AT LEAST THEORETICALLY) HEEDED MY PLEA TO ESTABLISH AS PRISTINE A DATABASE AS POSSIBLE BEFORE YOU EMBARK ON, or continue with, your ISFA program, you’re ready for next steps. And if you have your database in shape, you have accomplished the hardest chore in setting up your ISFA to perform at its best and return maximal value. Having a clean database allows you to systemize your approach to maintenance. You can confidently review very specific components within given machines in order to spot issues and successes with your CM program. Moreover you can create a lot of savings opportunities when you fully understand your equipment holdings. Let’s look at an example in terms of setting proper limits for your component types. We’ll also spot some errors along the way as a bonus.

Let’s say we are monitoring roller element bearings (for now the sub-classifications such as cylindrical- or ball-type elements are ignored for simplicity but addressed in a later article).

We received a file with 542 analyses from sumps (components) marked bearing, nothing else. My definition of a bearing is a machine that contains a shaft and one or more bearings upon which the shaft rotates. The bearing could be of a shell type (sleeve) or roller type. So immediately we are, or could be, faced with not knowing the primary bearing type, but let’s assume these were all from rolling element bearings because they were—again the power of humans to sift through incomplete information with effect. Thus, our database still may have promise. In Figure 1 are three “cuts” from the 542 sample results; ordering iron (Fe) from largest to smallest determined concentration via emission spectrometer. Also included are VIS, the other big four metals (aluminum (Al), copper (Cu), lead (Pb)) and silicon (Si) for good measure.


Figure 1. The highest 25 Fe values.

These are exceptionally high values for a roller bearing or, really, any bearing assembly with just shafts and bearings. Cutting to the chase, these are very likely gear sets—gears on shafts—accounting for the bulk of the Fe. Even if these samples were from bearings/shafts alone, they would have to be isolated from others in this group (see Figures 2 and 3), because Fe would be greatly distorted once a statistical study was made.


Figure 2. A 25-sample cut near the median value of 271.


Figure 3. The last 25 samples in the group.

By the time we get to the median value, there is hardly any Fe at all. Note, too, that there is no longer any VIS greater than ISO 220 (the first cut had only ISO 680 samples, a much thicker oil, and obviously in a totally different component type, bearing or not).

This cut shows virtually no wear metals of any type. Noticeably “thinner” oils are shown as resident in the sump(s). Clearly these are even different than those in the median group, and totally foreign to the first 25 samples.

Assumptions we can probably postulate to a great extent:

1. A number of different component subtypes is in play.
2. Further, some of these components have gears as part of their makeup (i.e., they are not isolated bearings).
3. VIS values support numbers 1 and 2 above.
4. Fe values support numbers 1 and 2 above.
5. If only one TOB (table of boundaries) was used to flag (assess severity and add a color) the various values; it could NOT have been appropriate for this wide swath of machinery.

Let’s check this out with a look at the actual TOB used to assess the test data (see Figure 4).


Figure 4. A TOB for bearings as calculated for THIS customer with ITS data.

Here are some aspects of Figure 4 with focus on Fe:

• The beginning of flagging (Normal values remain white—unflagged) occurs at the Notable (green) level; values above 251 ppm will be flagged green.
• Similarly values will be Abnormal (yellow) above 454 ppm, HIGH (orange) above 641 ppm, Severe (red) above 1065 ppm.
• Severity (coloration) steers the comment to be provided by the Intelligent Agent once rules are invoked, so that a proper sense of urgency is conveyed to the report recipient. It is the same logic a doctor of medicine will use in assessing one’s physical condition and the need for medical treatment, along with the time frame in which action should be taken.
• Figure 5 shows the hierarchical notion of severity as to data and to component (overall report and component condition as determined by comments).
• Statistical algorithms are employed to arrive at the coloration points.
o Outliers are thrown out. Fe 5/542 in the pop column indicates five values were thrown out.
o The range reflects the absence of the thrown out values, with 1671 ppm retained as the highest utilized in the stats calculations: 9999 ppm discarded, along with 4 others. You will notice in Figure 1 that the five discarded values do appear, though they weren’t used in the calculation of the TOB limit for Fe.


Figure 5. The hierarchical notion of severity as to data and to component.

This is all well and good, but it is a band aid on a big wound. We do not want this chart, with respect to Fe for sure, used to evaluate the samples a couple of hundred rows down the list of 542 results. Fe would never get flagged when it moved.

The resolution of this disparate data will involve several steps, starting with working with the plant(s) to drill down much further as to the component type—first and foremost—isolating which sumps are gearsets, then perhaps adding MFR/model (missing on most of these sample results) and so forth. In the next article, we’ll try to infer some of this because we’re not likely to get much credible assistance until we render these reports and point out the limitations we encountered. Some earnest communication is called for. 


Jack Poley is managing partner of Condition Monitoring International (CMI), Miami, consultants in fluid analysis. You can reach him at jpoley@conditionmonitoringintl.com. For more information about CMI, visit www.conditionmonitoringintl.com.