Please contact clientservices@accessintel.com or call 1-888-707-5814 if you are unable to login.
Not a member? Sign up
December 1, 2022 | By Chad Pasho, Mechanical Solutions Inc.
By allowing maintenance professionals to visualize vibration on entire pump systems, video vibration analysis and motion magnification can help to rapidly diagnose vibration issues
Video vibration analysis, and motion magnification in particular, is democratizing complex system-level vibration testing by making it simple and straightforward for maintenance professionals to measure and visualize vibration of entire pump systems. This comprehensive technique enables rapid diagnosis of machine or piping vibration issues in minutes, and the enhanced “motion magnification” video helps decision makers genuinely understand what is going wrong with the pump and pump system, facilitating buy-in for corrective action. And because the technique is fast and cost-effective to use, any bad-actor pump qualifies for analysis, no matter how critical. This ability can help reduce the backlog of chronic vibration issues plaguing the maintenance and operations staff. With the right tools, training and personnel, video-vibration-analysis technology can become a key component in your pump-troubleshooting toolbox.
Pumps are at the heart of many processes in the chemical process industries (CPI), moving fluids from one process to another. In nearly all instances, the pumping involves a dynamic rotating element of some kind that needs to be coupled to a static, stationary place in the facility. The combination of dynamic rotating components and static stationary components can often lead to clashes between the two — resulting in what engineers affectionately call “vibration problems.”
While some of these vibration problems arise explicitly within the domain of the pump itself, many are actually due to system-level issues, such as misalignment, or problems with the foundation or piping. In these instances, it may not be the pump itself that is causing the issue, but how the pump is interacting with the rest of the members of the team. For example, with misalignment, the pump and its driver are not properly positioned relative to one another during operation. With foundation issues, the dynamic pump may not be adequately coupled back to the static earth, either through structural issues with the pump base (stiffness, for example), structural issues with the foundation (such as cracking or voids in the baseplate), or the joining of the two (for example, loose or insufficient bolting). And with piping, it is not safe to assume that a pipe flange magically connects to a pump flange with no effect whatsoever on the forces being applied to one another, particularly as thermal changes occur and the forces of the pumped fluid join the fray.
Properly diagnosing these pump-system vibration issues at the system level is often challenging because a) it can be technically difficult to know where to start looking within such a complex, interactive arrangement, and b) it can be organizationally difficult, both in terms of internal maintenance staff and external vendors, to sort out who is responsible to make things right. In terms of knowing where to start looking for the cause of the pump-vibration issue, sometimes there are “rule-outs” that can narrow the list of potential suspects, but it is often best to approach the situation with a curious, detective-like mindset, looking at the overall system’s big picture and not ruling-out any prospective contributors prematurely. As for the organizational difficulties of sorting out responsibility, given the complexities of interaction, there is a contractual disincentive for vendors to take ownership of the problem. It is often going to be the de-facto responsibility of the end-user, or their appointed independent third party, to step back and look at how all the pump-system equipment is working together (or not working together as the case may be).
Historically, gaining a system-level diagnostic perspective for pump vibration problems has been possible, but it typically involves a lot of time and resources. Accelerometers are temporarily roved throughout the pump system, similar to the way a physician would listen to various points of a patient’s chest with a stethoscope. The vibration measurements obtained in this way are then mapped onto a wire-frame model of the pump system. The collated vibration data are subsequently synchronized and the motion exaggerated onto the model to visualize how the pump system’s components are interacting with one another. This technique, aptly called “operating deflection shape” (ODS), can often identify the root-cause offender, because it translates point-specific vibration information into a contextualized shape of the deflection for interpreting the data.
While effective, this system-level technique requires both time and money — typically tens of thousands of dollars and days to weeks of data collection and post-processing time. Unless this is done frequently enough that maintenance personnel are able to keep sharp on the techniques involved, it is best left to external experts. The potential need for outside help turns every troubleshooting exercise into an explicit budgetary return-on-investment exercise as well, consequently leaving many chronic problems to fester below the investment threshold. Even if the problem “cost” passes the investment threshold, there is a persistent trade-off when collecting the data: since collecting data takes time (and therefore, the time equivalent of money), the pressure to get just enough data to make a diagnosis may not necessarily result in getting the data necessary to make the correct diagnosis.
What if the same system-level analysis could be achieved simply by internal personnel with diagnostic information available in minutes instead of weeks, without the pressure of incremental data collection yielding an uncertain return on investment? Enter video vibration analysis, which uses high-speed, high-resolution video to capture a comprehensive, system-level matrix of vibration data in seconds, then post-processes that information in minutes, and presents accurate vibration information and motion-magnified video for system-level diagnostic evaluation (Figure 1).
FIGURE 1. High-resolution video can capture a system-level matrix of vibration data in seconds
These systems work by taking high-speed, high-resolution video, and then extracting vibration data based on the change in light intensity on a per-pixel basis. This change in light intensity at the pixel level is then referenced to real-world dimensions, and tiny amounts of motion are quantified as displacement data. The primary benefit of this technique is that these displacement data are comprehensive, capable of capturing the entire pumping system’s vibration data simultaneously. As an added bonus in certain situations, it helps that it is a non-contact sensor, meaning that high temperatures and inaccessible equipment do not necessarily present a problem — as long as you can see it, you can measure it.
In addition to your data set, you now also have a picture of the system that you can enhance in various ways. Amplitude and phase data can be plotted onto the image to communicate location-specific data at the pixel level. And, the ultimate data enhancement is to take the actual motion that is present and magnify it, presenting a video with exaggerated motion of the entire system, similar to the ODS accelerometer-based approach, albeit without all the time-consuming hassle. Not only is this highly effective for root-cause diagnosis on system-level issues, it also becomes a very effective tool for not-quite-as-technical decision makers in the organization to understand what the problems and potential corrective actions are.
On the surface, video vibration analysis sounds great, but a reasonable question to ask is “what’s the catch?” Any new technology can be as useless as snake oil without understanding the fundamentals that will enable it to be successfully deployed to meet the expectations that you and your organization have for it. The following are several factors that have been found to be critical to ensuring productive and successful use of the technology.
Toy versus tool. Beware the “as-seen-on-TV” rush to get the latest diagnostic gadget, only to leave it on the shelf unused over time because of training or support issues. Video vibration analysis does indeed qualify as new technology, and does require personnel to acquire proficiency in some new skills and techniques. It is doubtful that most maintenance teams include professional videographers, so it is important to make sure to budget resources for training. A sufficient critical-mass of skilled and teachable users and internal customers will ensure that the new capability will be adopted by your organization.
In addition to standard software support, it is also important to consider “engineering support.” Given the novelty of the technology, it helps to have mentors available to help determine the following: first, whether or not you have collected good data, and most importantly, what to do if you didn’t; and second, how to interpret the data for successful diagnosis and problem resolution. If your organization has experience collecting and interpreting ODS data, you have a head start, but this head start shouldn’t be confused with the finish line. There is a lot to learn about how to best use this new technology, and in the words of American author and motivational speaker Zig Ziglar, “Some of us learn from other people’s mistakes, and the rest of us have to be other people.”
Capex versus service. Given its novelty, and the commitments required to truly make video vibration analysis technology successful in your organization, each company should evaluate whether it makes sense to acquire, staff and maintain the technology in-house, or if it is preferable to rely on third-party service providers. Sometimes organizations opt for an incremental approach, trialing video vibration analysis via a third party, before committing to purchasing the technology outright. If your organization does engage with third parties, either for evaluation purposes or for permanent supply of the capability as a service, do not assume that any organization with a high-speed camera and some software knows what they are doing. As discussed previously, this is a tool that requires training and experience to utilize properly, and if your organization prefers to outsource, make certain that the training and experience is indeed present in the third party. As an additional consideration, proficiency with a diagnostic tool does not automatically make one a skilled problem solver. In order to properly diagnose and solve problems using video vibration analysis, people are still required to understand and interpret the data, and conceive of solutions, so make sure you are hiring the right people, not just time with a widget.
If you do conclude the technology belongs in your organization, or even if you opt to rely on service providers, you will need to do some homework to understand the various capabilities available.
One of the most important concepts, and possibly the least explained in the industry, is the detection threshold. Perhaps you have heard complaints about the technology not working except for very high vibration levels, or have experienced this issue yourself. This is likely due to the detection threshold. In short, video vibration analysis has a minimum level of displacement required before it can detect the signal (Figure 2). Unlike the piezo-electric accelerometer, which can effectively measure nearly zero vibration, the pixel of the camera sensor has a noise baseline that makes a measurement of near-zero impractical (more on that later). Thus, is becomes critical to understand what that minimum level of displacement required is, and equally important, what the displacement depends on, because it’s not absolute — it’s relative.
FIGURE 2. Any velocity values above the detection threshold lines will be accurately detected and magnified. Understanding the detection threshold as it relates to frequency is very important for diagnosing lower-displacement issues
Given that this is an optical measurement, and that a pixel has specific dimensions of its own, each pixel is subsequently covering a given amount of space on whatever the camera is viewing. As a basic example, consider a four-pixel camera (two pixels wide by two pixels high). If this four-pixel camera was observing a 1-in. × 1-in. square, then each pixel is observing a 0.5-in. × 0.5-in. area for motion, and its detection threshold scales accordingly for this area. Now if the same four-pixel camera then observed a 1-ft × 1-ft square, each pixel would then be observing a 6-in. × 6-in. area for motion, and the detection threshold would be scaled higher to accommodate the larger area. Since there are typically millions of pixels in play, the detection threshold is most effectively considered as a function of the pixels in aggregate, or the “field of view.” Rather than determining what each pixel is measuring to figure out the detection threshold, their combined dimensions can be easily determined. For example, a 10-ft-wide field of view (meaning the camera is viewing something 10-ft wide from horizontal edge to edge) has a given detection threshold when considering all the horizontal pixels together (for example, 1,920 pixels wide). In this example, each pixel is observing a 0.0625-in. × 0.0625-in. square, with the detection threshold being a function of that area.
The detection threshold is considerably less than the pixel dimension, but is not zero. The technical reason this detection threshold exists is due to the noise in the sensor. Recall that the fundamental measurement is variation in light intensity. The noise in the sensor (one of the millions of camera light detectors) effectively produces a relatively constant baseline level of light intensity variation. Consequently, actual change in light intensity due to movement within the level of the noise floor is not distinguishable from the noise itself, and therefore is not accurately measurable, or magnifiable for visual motion interpretation.
Keep in mind that this detection threshold is a displacement value, not a direct velocity or acceleration measurement. While there is no frequency dependency to the displacement value, there is a practical frequency relationship. As the frequency of vibration increases, the velocity required to reach a given displacement also increases (as the frequency of oscillation increases, there is less time for the pump to move from point A to B). Typically, vibration velocity is the measurement of interest when determining machinery health, so a fixed-displacement “detection threshold” will equate to an increasing level of velocity “detection threshold” as the frequency increases. When this concept is applied, as the frequency increases, displacement levels are lower for a given velocity, and it becomes all the more significant to understand the displacement-based detection threshold to ensure adequate measurement and magnification at higher frequencies.
If you have followed the logic thus far, you may be thinking, “why not just reduce the field of view to see lower levels of vibration, even at higher frequencies?” This is indeed possible, to the point of detecting nanometer levels of displacement, but there is a trade-off. As discussed earlier, the primary advantage of this technology is the comprehensive nature of the measurement, whereby the user gets to “see” and measure all the components of the system together at once. By narrowing the focus to achieve lower levels of displacement detection, the user has resorted back to a small measurement area, which behaves similarly to the accelerometer, and thus forfeits the underlying value of the technology.
With this fundamental understanding of detection threshold in hand, note that there are two types of data realized, and potentially two different detection thresholds. Depending on the vendor, the vibration measurement may or may not have a different threshold than the magnification threshold. It is important to do your homework, and ask for data to support the claims, since your mileage may vary if the quoted measurements are not provided in an appropriate context (for example, what is the field of view for a given detection threshold?).
Also, with regard to the magnified motion capability, there are differing approaches to enhance the signal, producing the magnified motion videos. The two major methods are optical flow and sub-pixel magnification. Primarily in an effort to speed up the processing of all the pixel-level data, optical flow takes the millions of pixels collected and averages them together into regions, and applies significant gain, along with interpolation across regions, to effectively recreate the wire-frame model of ODS. If the averaged region size is sufficiently small (and the user sufficiently patient), reasonable enhanced video results can be achieved, but the user must appreciate that the interpolation combined with noise can leave much to the imagination, and finer details can be lost in the process. An analogy would be drawing with a paint brush versus a mechanical pencil.
On the other hand, sub-pixel magnification maintains a pixel-level analysis to keep the hard-won resolution in the diagnostic assessment. This makes detecting issues requiring detail possible that might be missed or misrepresented by optical flow, such as foundation cracking, loose fasteners, piping supports, and similar issues, while still being able to magnify overall motion issues, such as misalignment, that may be also displayed with optical flow. The motion is much more precise than optical flow, and effective at lower levels of displacement.
Regarding the accuracy of the technology, if all you need is a single-point vibration measurement, it is still going to be faster, easier and more accurate to obtain that single point of data with an accelerometer for the foreseeable future. The advantage of using video for vibration measurement is the comprehensive nature of the measurement, effectively taking millions of measurements all at once. But how accurate are the vibration data that the video vibration analysis provides? Is it just a visualization tool, or can you measure the vibration as well?
FIGURE 3. A moderate level of light and ligh-dark contrast are required for accurate displacement measurements
With the above discussion of detection threshold in mind (since accuracy will not be applicable if the signal is below the noise floor), there are several practical elements to consider to improve the accuracy. Since the measurement is one of changing light intensity, it is important to make sure there is a “Goldilocks” level of light — not too bright, not too dark. Either end of the spectrum ends up clipping the vibration signal, resulting in inaccurate measurements. Relatedly, since the technology is looking for a change in light, contrast is required to make a measurement. Given a featureless, smooth, single-color surface, there will be no accurate measurement possible. In situations where targets can be applied to provide a point of contrast, the camera and software can accurately derive displacement measurements (Figure 3). Once there is sufficient lighting and contrast, it is still up to the user to provide a dimensional reference of some sort, either the distance between the camera and target, or a reference length within the image itself. Either way, the system needs a way to correlate the pixel size back to real-world size. Any lack of accuracy in this step will directly correspond to a lack of accuracy with the calculated measurement.
And again, any vendor claims should be substantiated with data to support them, and their relevance should be clearly understandable by the user. For example, if the system has a certain accuracy, but it is for 0.1 in. peak-to-peak of displacement motion with a 1-ft field of view, it is not very relevant to the real-world pumping system conditions for which the user will likely be seeking accurate data. Understanding the accuracy, particularly as the system approaches the noise floor, and over the entirety of the frequency span, will help ensure users know what to expect when deploying the technology.
In addition to magnifying the motion present as a result of operating forces within the system, like that achieved in an Operating Deflection Shape, there is another critical diagnostic piece of information when dealing with pumping systems: At what frequency is the vibration occurring, and what are the natural frequencies? Every system has several inherent natural frequencies. When these natural frequencies are matched (or “excited”) by a forcing frequency in the system, such as 1x operating speed, they resonate. This resonance automatically amplifies the vibration level, and generally makes for a very bad day. When your pumping system “sings” in this manner, it is not music to the ear, but rather a headache.
These natural frequencies can be detected by intentionally impacting the object and then measuring the frequencies at which it naturally vibrates. While often done with an accelerometer, or several if the shape of the motion is being characterized, this can now be achieved quickly and comprehensively using video. Since the vibration after impact is generally very short in duration, several impacts can be “averaged” together, which reduces the noise floor, thus lowering the detection threshold. This lower detection threshold makes natural-frequency characterization using video possible, even while equipment is operating.
In addition to natural frequency characterization, is it useful to analyze trends in low-level vibration over time to understand how the condition is changing. If the detection threshold is too high, then the user must wait until the vibration levels are problematic to the point of requiring action before employing the technology. But the aforementioned averaging capability can also be applied using a keyphasor or similar signal to trigger multiple videos for averaging. This reduces the noise floor, and lowers the detection threshold, so that vibration issues can be monitored and understood well before they require intervention.
This can be useful for early-stage diagnosis of impending problems that have a certain characteristic frequency, since the video pixel intensity variation detected in this manner can be analyzed by fast-Fourier transform (FFT), and then the vibration-versus-frequency spectrum at each pixel can be determined by the motion magnification system. Various vibration levels, or changes in those levels, at key frequencies, can be excellent tell-tales for machinery diagnosis and prognosis, as discussed in the ISO 13373 series of standards.
Video testing simplifies the measurement and visualization of vibration on pump systems, enabling rapid diagnosis of vibration issues.
Chad Pasho is the director of business development for Mechanical Solutions, Inc. (MSI; 11 Apollo Drive, Whippany, NJ 07981; Email: chad.pasho@mechsol.com). Since joining MSI in 2014, Pasho has focused on rotating machinery diagnostics and development, and several high-speed video characterization products, including VibVue for characterizing vibration and motion magnification. Pasho’s previous experience includes technical and management roles at Schlumberger, Octave Communications (now Polycom), and NextEra Energy. He has an MBA from the Olin Graduate School of Business at Babson College, and a B.S.Ch.E. from Worcester Polytechnic Institute.
Join today to get all of CHEMICAL ENGINEERING
Chemical Engineering news and content delivered every week
© Access Intelligence, LLC. All rights reserved. | Privacy Policy | Diversity Inclusion & Equity