FREE copy of the Uptime Elements Implementation Guide once you subscribe to Reliability Weekly

A permanent system, which covers 108 pieces of equipment, ranging from 174 to 15930 RPM, using 412 accelerometers and 48 proximity probes is discussed in this article, and the majority is on slow speed trains.

Slow speed gearboxes handle pelletizing which is the final manufacturing step in polymer extrusion. As a critical process in introducing additives or modifiers to enhance or distinguish the products by customized recipes, pelletizing is always accompanied with a heavy and fluctuating process.

Mechanical faults on slow speed pelletizing gearboxes typically have a low response. Despite their slow speed; its failure is never sweet, most of the time, it is silent without early warning and violent with extended consequences. There are a lot of myths worth discussion about this unique creature.

Consistency - The Basic Requirement for Slow Speed Gearbox Monitoring

For detection, numerous guidelines are advocated and are only limited to general rotating; they are absolute in concept and may not agree with each other. Unfortunately, no benchmark is available for slow speed gearboxes to begin with, and experts keep quiet about this topic.

Relative trending provides another option; how far and how fast the pattern is climbing which is easy and self-explained. However, consistency is essential.

Figure 1

Figure 1 demonstrates a 42 percent variation between readings taken a few inches away, which happens likely by using a portable meter during a monthly cycle. The next walk-through might be collected by a substitute, and the result unpredicable. Without consistency, it is tough to differentiate among human factors, process load, rotating speed, or mechanical defects.

The first case study occurred just after commissioning an online system; its last portrait was launched remotely before it tripped. The second case study validates the lessons learned from the first case study.

Figure 2

The First Case Study – “How High Is Too High”?

Without an existing reference, the only way to gain experience is after the fact. Figure 2 illustrates a trend on the output shaft (174 RPM) from a gearbox with three shafts, three rotating speeds and three gear mesh configurations. It is around 0.04 in/s change, ranging from damaged bearings, cracked gears, broken teeth, an egg-shaped bore housing, and loose shaft to bearing fitness, etc. As a general rule of thumb, level 0.3 in/s is roaming on the severity margin between tolerable and rough, which was five times higher than the last 0.06 in/s, and proved to be a catastrophic failure.

One popular myth for abnormal detection is: “how high is too high?” For slow speed gearboxes, the question becomes the opposite: “how low is too low?” Where should the proper threshold be: 0.05? 0.04? or 0.03 in/s? Intuition is expensive, but it can be inferred - every one thousand is counted for slow speed gearboxes. However, this only can be achieved by using a permanent system.

More than detection, time waveform and spectrum analysis are informative for diagnostics, but in their own way. Periodic spikes at 174 cpm in Figure 2 (upper left) identify that the defects originated from the output shaft.

Spectral analysis is sophisticated for gearboxes; fortunately, it is an exception here and makes it an easier precursor for the comprehensive second case study. In Figure 2, a resonance haystack (about 3000 cpm and < 0.02 in/s) is stimulated by the impacts between broken teeth. Although not related to any rotating speed or gear mesh, its 174 cpm sidebands (after zoom in) still spotlight the output shaft and match the clues from the time waveform.

To wrap up the first case study, it can be concluded that data in Figure 2 provides detection function. However, time waveform offers further insight that not only indicates the footprint (174 cpm) from the broken teeth on the output shaft, but also its impact in nature, which is valuable for severity evaluation.

Now, this learning can be deployed and elaborated in the second case study.

Figure 3

The Second Case Study - Detection at Early Stage

Figure 3 is a derivative from the first case with minor modifications: rotating speed, gear teeth and bearing models, etc. Three damages occurred: the first two are on the same bearing A (red, input shaft outboard end), and the third one on bearing B (green, intermediate shaft motor end). Each spectrum is posted next to its damaged image.

Figure 4

Figure 4 is a traditional overall trend over 20 months; scan around every 15 minutes since the online system was commissioned. Before the first ignition (bearing A inner race, November 26, 2013 and circled red), it was buried by an overwhelming data mine (variable load, stop and start up, etc.). Why is this specific date singled out? Waveform Peak-Peak is another trending, although a few times a week, do provide more sensitive warning, a jump on the same date explains the reason.

Figure 5

Figure 6

Figure 7

Figures 5 through Figure 7 illustrates comparison before and after each event with its associated time domain on the right and frequency domain on the left. 96000 cpm is sliced into ten (10x) bands with 1600 lines and 60 cpm resolution. Those bands are categorized as gear mesh and non-gear mesh, with non-gear mesh bands inserted among gear mesh without a gap. Only the most representative band trending is selected (left lower). All related information is summarized in Table 1.

Table 1

The Second Case Study – Analysis to Identify the Defects

Trending within a narrower frequency zone is sensitive to minor change for early detection and can focus on nominated targets like bearing defects as long as you know where they are. For example, higher thresholds can be assigned to gear mesh, which fluctuate with load, and false alarm can be eliminated by process variation.

Table 2 is a matrix for diagnosing. Shaft rotating and gear mesh frequency are generated as long as the machine is running, then defects are added if they occur - new, adjacent or exactly aligned with normal operation footprints (shaft rotating, gear mesh or one of their harmonics). On vertical direction, shaft speeds, gear mesh and defects are categorized into three groups (up, middle and bottom). All associated harmonics are listed along horizontal direction.

Table 2

The challenges of each event are:

1. For the first event (bearing A inner race), 600 – 4900 cpm band (Figure 5, lower left) simulates time waveform trending well (Figure 5, lower right). Both 1/2x and 1x 1200 cpm spikes in the time domain (right middle) and 1/2x and 1x 1200 cpm sidebands in the frequency domain (with inner race 10942 cpm at infant) are conspicuous and complement each other. To identify developing bearing defects from gear mesh is tough. Referring to Table 2 (bottom and blue filled), 5x inner race (54710 cpm) is 10 cpm from 4x GM2 (54720 cpm), which is also blue circled in Figure 5 spectrum. Coincidentally, 7x inner race (76594 cpm) is less than 4x resolution lines from 2x GM1 (76800 cpm) and both are marked in yellow.

2. For the 2nd event (bearing A outer race), 3900 – 7900 cpm band is assigned for 1x GM1 (5900 cpm), however, the outer race burst (6780 cpm) falls into the same zone. In Table 2 (bottom and black filled), all even 6780 cpm families (2x, 4x, 6x, 8x, 10x, 12x and 14x) are urgently next to each GM2 harmonic, which make the Figure 6 spectrum quite noisy.

3. For the 3rd event (bearing B inner race), the major peak at 2x inner race (20280 cpm) falls into a non-gear mesh band, however it is close to the 2x inner race (21884 cpm) of bearing A, both are green filled and hard to distinguish.

Lessons for Detection

No trend is equal. Traditional overall trending (colored purple) in Figure 4 is not responsive. On the other hand, time waveform P-P tracking (colored blue) is proactive. Either prediction with a three months grace period or protection at the last stage, or ideally both, just depends upon your setup.

Today, there is no isolated information island; messages can be shared within seconds, no matter when or where you are.

Predictive maintenance (PdM) incites a concern. However, if nothing happens immediately or the booming market does not allow for service, the next question an analyst might face is: How long can it last? One of the many requests by unit operations is to shorten the current cycle from monthly to biweekly to weekly and eventually to the utmost, daily, which is a definite penalty with stressful side effects for portable route checks if the detection is too early. Rather than moving people, streaming data updates gives permanent online vibration monitoring another scenario.

Lessons for Analysis

Detection just notifies personnel that something is going wrong. When more information for the reason behind it is desired, analysis is not easy on the slow speed gearboxes.

The aforementioned discussions are based on what is already known and might not actually be available in many other plants. The machine design information is complicated in nature. Accordingly, the matrix composed of permutation and combined with numerous shafts, gear mesh and bearings is amazingly intimidating. It is frustrating to make an assumption, or try to align calculated faulty frequencies with peaks in the spectrum while several defects develop simultaneously and compete with each other. Nobody can predict which component is the weak link this far in advance.

Figure 8

Back to Basics

Economics is both the propulsion and the destination for condition monitoring. Fortunately, many technologies are available (vibration, lubrication, ultrasound, infrared, etc.), however, a permanent online system is unique in its kind.

Machinery wear mechanisms (e.g., ISO15243:2004 for rolling bearings) are dynamic. Spectrum and time waveform are just two in this article, yet there are still many other tools (e.g., stress wave, envelop demodulation, etc.), that have their own strengths and all can work together to aim at the moving targets from infancy and developing to the last stage.

Information is important for decision making, however, without proper interpretation, more data means more confusion (Figure 8). Today, there is no isolated information island; messages can be shared within seconds, no matter when or where you are. However, digital devices only broadcast the repeated messages without judgment. How the system is set up to initiate the first meaningful attention is the key.

Whether it is a portable or a permanent online system, detection to do the right thing and analysis to do the thing right should always be on the list and can never be missed. Both are the core for vibration condition monitoring and also the contribution by industrious professionals.


A special thank you to Jeremy Menchaca and Scott Grantland for their review of the article.

Han-Chian Gee, P.E., is Chief Reliability Engineer in the PdM department at the Point Comfort, Texas, plant of Formosa Plastics Corporation. He is a licensed professional engineer in Texas and an ISO18436.2 Category III certified analyst.

Upcoming Events

August 8 - August 10, 2023

Maximo World 2023

View all Events
80% of 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!