Since leading indicators at inception are not yet typical "wall chart" fault conditions, finding them as they happen requires a really close look at every spectra and a comparison to historical signatures to distinguish them from normal, transient conditions related to load, temperature, process rates, accelerometer data, etc. This could be done by hiring a large team of analysts dedicated to doing only that, however, the room for error is high and the associated cost and effort is extremely impractical. Alternatively, your vibration software can automatically do this for you by using trend-based narrowband envelope alarms. Setup is easier than using band alarms and the results are much more accurate. Furthermore, findings are discovered sooner, allowing your facility to drastically reduce the amount of points monitored and the frequency at which they are taken. Optimizing your vibration analysis program to find leading indicators will also allow you to relax, but not eliminate, your other predictive maintenance efforts, including oil analysis, ultrasound detection and thermography (see Figure 1). With less data to collect and analyze, a large facility containing over 2,000 machines can easily be monitored by a single analyst.

{image:1}

Figure 1: Typical overall vibration trend of a standard piece of rotating equipment

A narrowband envelope alarm is an alarm level that monitors the absolute, not overall, amplitude of every peak in a spectrum. Unlike spectral band alarms, often called power bands, there are practically no limits* to the amount of narrowband envelopes that can be displayed on a spectrum. This eliminates the problem with overall and spectral bands where low amplitude noise, such as slightly elevated noise floor, triggers the alarm. Because of this phenomenon, overall and spectral bands must be set at a higher amplitude offset to reduce the amount of false alarms. Unfortunately, this pushes the alarm limits beyond the sensitivity needed to detect leading indicators. With narrowband envelope alarms, this problem doesn't exist. The set point is unique to each frequency, so the sensitivity can be much closer and false alarms caused by broadband noise are eliminated. With a tighter offset, important forcing frequencies will never go undetected due to the higher amplitudes of the surrounding frequencies. Simply put, the narrowband envelope will stay out of alarm unless a legitimate issue presents itself.

As complex as they may sound, narrowband envelope alarms are very simple to set up. Basically, the narrowband envelope alarm is a line drawn above the outline of the spectrum collected. Most vibration software will automatically generate the alarms using an existing spectrum. If any peak breaks this line, the next time data is uploaded, the alarm is triggered, regardless of its amplitude. Hence, even very low amplitude changes at any specific frequency will be immediately detected and pointed out by the narrowband envelope alarm. To avoid these from being triggered every time the signature changes, the narrowband envelope alarms should be trend-based. This, too, can be simply done by overlaying several pieces of good data and adjusting the line drawn (i.e., the narrowband envelope) to allow for non-fault related conditions (see Figure 2). Once again, the process is simple compared to setting spectral bands or overalls. You do not have to define a frequency minimum (Fmin), frequency maximum (Fmax), specific amplitude for alert, or another for danger. All you need to do is redraw the line, which can be done by simply clicking on the line and dragging it to where it needs to be. Doing so also creates variable offsets, allowing you to compensate for the limitations of the particular type of data being collected (e.g., acceleration, displacement, or velocity). The more data you have overlaid and the more these are adjusted for different running conditions, the more accurate they become. Eventually, they will stay in the green under any previously seen condition and be triggered the first time something new shows up on the data. At that point, the analyst can see exactly what's new and decide whether it is a leading indicator of a fault or something that is normal. If it's unclear, the alarm can be adjusted to just slightly above its current amplitude at the suspected frequency and the next time the data is collected, if it stays the same, it stays green and if it worsens or anything else changes, it alerts the analyst to look at it again. This way, at the very moment the leading indicator is identifiable, the machine goes into alert status and if all is as it has been in the past, it stays clear. It sounds like a lot of work, but the software does it all, freeing up time for the analyst.

{image:2}

Figure 2: Trend-basing the narrowband envelope alarm by overlaying several fault-free spectra collected under various conditions

Figure 3a shows a historical trend of overall velocity collected from a steam turbine. The amplitude levels and spectra are constantly changing due to variances in speed, load and random white noise seen in the noise floor from steam and cooling water flowing through the turbine. At the indicated point on the trend, a leading indicator exists. If the analyst were to monitor the overall levels, he or she would be unable to detect it. Since the machine is not in alarm, it is unlikely and impractical that the analyst would spend the time analyzing the detailed data shown in the spectrum in Figure 3b. If by chance the analyst did look at the details of every piece of data collected, the primary question in Figure 3b would be, "Does this spectrum indicate a problem?" When comparing this data with the variances found in historical signatures, it is unlikely that any abnormality would stand out. Since there is no mechanical fault yet and anything that looks like a fault is at an almost null value, the diagnosis would most likely be that this machine is in excellent condition. In Figure 3c, however, the same data is displayed showing the trend-based, narrowband envelope alarm. The alarm is referenced and immediately anyone can tell that there is some strange new detail in the vibration signature. A closer look at the frequencies in question will reveal that these are the natural frequencies of the installed bearing. The low amplitude peaks seen in the lower frequency range match the ball spin frequency, the white noise is shown as an elevated noise floor and other filtered data collected from the machine (not shown) confirms a lack of lubricity. The recommendation was to drain and fill the oil in the bearing housing to "sweeten" the oil. A visual inspection of the oil drained showed the oil was dirty and discolored. After partially draining the reservoir and refilling it a few times with clean oil, data was collected again. Figure 3d shows that the data is now out of alarm. Even though the overall vibration is significantly higher than before and the spectrum has greater amplitude activity in the lower frequency range, the data has returned to within its typical historical trend. Lubrication related leading indicators are very common, especially in facilities with a less than world-class lubrication program, but they are not the only type of leading indicator that is discoverable using this alarming method.

{image:3}

Figure 3: Envelope alarms not only detect small changes in the vibration signature, but also quickly point out the change

In Figure 4a, low amplitude high frequency peaks again can be seen in a spectrum collected from the bearing housing on a split case pump. These peaks are of insignificant amplitude and line up with the sound the harmonics of the outer race of the installed bearing create. Since there are no fundamental fault frequencies, sidebands, or other indications of an issue, and the amplitudes are low, it is safe to say this is just the sound of the bearing turning. This is a true statement and there is no fault. However, further investigation shows the peaks have never been there in the five-plus years of historical data. With the tremendous amount of time saved by using the narrowband envelope alarms, the analyst can go investigate this seemingly insignificant singularity. In this case, the data was found to be repeatable and was seen in other orientations at the same location. A closer look beneath the coupling guard showed that the packing was leaking. The leak from the packing and a plugged drain had filled the bearing housing frame adapter with water. The water was drained out and the data collected afterwards returned to its normal looking signature, consistent with the five-plus year trend in which the alarm was based (Figure 4b). To satisfy curiosity, the adapter frame was allowed to refill and the data collected confirmed that the water filling the bowl-shaped frame adapter had amplified the sound (i.e., vibration) of the higher frequencies, similar to how filling a drinking glass changes the pitch of the higher frequencies as you fill it with water. Had this dirty water been left in the adapter frame, it eventually would have entered the bearing housing, causing a failure. Having this documented, if the issue recurs, the analyst can create a report stating that, according to the data, the drain on the bearing adapter is plugged. Findings like this do a lot for the credibility of the program. This type of finding may not be typical, but is a good example of how trend-based narrowband envelope alarms take analysis far beyond finding faults seen on the wall chart.

{image:4}

Figure 4: Trend-based narrowband envelope alarms free up the analyst’s time to focus efforts on seemingly insignificant abnormal conditions, detecting the first sign of a potential fault-creating situation

It takes very little imagination to envision what an actual machine fault will look like using this alarming method (Figure 5). If your narrowband envelopes are properly set up, which, again, is very simple to do, a genuine fault condition will light up terrifically. Every data point on the machine with an issue will be in red. At this point, you may be able to detect a problem with properly set overall or power band vibration alarms, but not if the machines have a wide range of normal overall amplitudes. With even a trivial fault condition, the vibration signature details all over the machine will be affected. Each narrowband envelope alarm will point out exactly what's wrong and the diagnosis will be quick and accurate. If, for some reason, the machine has to continue to run with the fault, the alarms can be quickly and easily adjusted to prompt the analyst again when the condition worsens. While finding machine faults is not the primary focus of this article, trend-based narrowband envelope alarms drastically improve that capability as well.

{image:5}

Figure 5: The narrowband around 1x allows for normal variances in the imbalance of the machine while the looseness condition caused by worn sleeve bearings below grade are detectable above in the motor before the overall velocity is affected

Finding faults early means the types of corrective actions will change. Instead of planning to shut down a machine to make repairs, corrective actions would be in the form of adding grease, tightening a bolt, changing the speed of a motor, or changing process rates. Depending on what is discovered, more times than not, a leading indicator can be addressed without shutting down the machine. The entire approach changes from a somewhat predictive but reactive one to truly preventative.

Since the problems getting fixed are not obvious, it is essential that the supporting data is shared with other departments. Data collected before the corrective action should point out the frequencies of concern. This will justify performing the corrective action. It is also equally important to show the result of the data after the corrective action has been completed. This not only proves that the issue was resolved, it also shows the rest of the facility that what is being done works and is worth doing. No matter how many machines are saved or how effective the program is, if it does not get the proper exposure, the program will not get the support it needs.

So, if this method is so effective and not new, then why isn't it being widely taught? The answer is simple. Most vibration certification companies focus on teaching the general concept of vibration analysis. They center on how to tell if a machine is good or bad, how to make sure the proper data is collected, how an analyzer works, common machine fault characteristics, standardized overall vibration limits and a grip of other things that are vital to the understanding of the craft. Most of them mention narrowband envelope alarms as well. This isn't new or exclusive technology. In an attempt to improve their vibration program, companies wanting to improve will quickly spend the money to certify their analysts to a higher level or buy more equipment. This is effective and a good practice, as well as very common. Hence, the result is a company with a very good analyst with a typically good program. It is less common for that same company to encourage its analyst to attend training from the vibration hardware and software manufacturers so he or she can learn to really optimize the company's program based on what it uses. Companies are even less likely to send their staff to conferences where analysts can network with other companies and learn from what they are doing beyond advanced early fault detection. Networking events are where these types of things are discussed, yet even the few analysts that utilize this type of technology stay at home and don't often share information outside their facility. Recognition awards are based on metrics used by the status quo, so again, they are not widely publicized. Eventually, they will make their way to certification training in a generalized capacity, but not until they are more commonly practiced in the industry. Until then, one can only learn about them at conferences and read about them in publications like this one.

The fact is, in order to have a world-class vibration program, you do not need a huge team of experts to constantly analyze large amounts of data. You do not need a lot of expensive tools, or to outsource your efforts. Additionally, you should not settle for the discovery of damaged equipment and the constant firefighting mode of taking quick action to avoid downtime just because it's what everyone else is doing. When you basically have an arrow pointing to machines that have an abnormality and then pointing out exactly what that abnormality is, it radically simplifies the whole process. All you need is to use the existing tools most programs already have. Then, simply relax and address the leading indicators when time allows, well before faults occur.

* The number of narrowband envelopes matches that of the resolution of the spectra. It takes about five fast Fourier transform (FFT) lines to build a frequency in a spectrum, depending on the resolution and frequency maximum (Fmax). If a spectrum has 6,400 lines of resolution, it will allow approximately 1,280 envelope alarms.

Richard Bierman

Richard Bierman is a Vibration Analyst for Chevron Phillips. Fifteen years of equipment monitoring experience has allowed Richard to understand the cause and effect of mechanical faults and how they relate to the vibration data, as well as recognize fault conditions not found in documents on the subject of typical vibration diagnostics. Mr. Bierman is also the SME for Lubrication Analysis and Program Optimization. 
www.cpchem.com

Upcoming Events

August 9 - August 11 2022

MaximoWorld 2022

View all Events
banner
80% of Reliabilityweb.com newsletter subscribers report finding something used to improve their jobs on a regular basis.
Subscribers get exclusive content. Just released...MRO Best Practices Special Report - a $399 value!
DOWNLOAD NOW
The Three Laws of Preventive Maintenance

The Three Laws of Preventive Maintenance

Each preventive maintenance task in an Uptime Elements developed Reliability Strategy is generated for an identifiable and explicit reason

Digital Built America: Smarter, More Sustainable and Resilient

Digital Built America: Smarter, More Sustainable and Resilient

Building back better means transitioning the current infrastructure to smarter, more sustainable forms of development to safeguard the country’s future.

DIPF Curve and RCM Failure Patterns

Predictive Maintenance Deja Vu All Over Again

Compared to total asset failures, what percentage of asset failures can be "reliably predicted" with predictive maintenance?