Condition-based monitoring is regularly checking the condition of your machines and looking for changes in the data collected. Vibration analysis is an analysis method for looking for and identifying the machine’s faults in vibration data. CM should and mostly will include vibration analysis of your machine. But you may want to take other data, and maybe you should be using other collection methods, such as thermometers or ultrasound.
Most condition monitoring systems use vibration to collect the data. It’s cost-effective and easy to do so, and produces good data. Vibration analysis is done at a higher resolution than condition monitoring data. Today’s fast and powerful vibration analysis tools can allow us to collect both condition monitoring and vibration analysis data together, when they are set up correctly.
One problem that the analyst has to face is, is the actual machine worth collecting data from? Is it important enough? And how much data is needed to identify any machinery faults? You should have done a criticality study to answer that question. I worked with a cement manufacturer in the UK and called a fault on a small machine that it had classed as non-critical. The problem was a small bearing that would have cost £50 to replace and about £500 to change, as it was a few hours’ work. But they didn’t actually fix it, and it failed, leading to 16 hours of downtime costing £10,000 an hour. That’s quite a big number. And you don’t know how that loss of production could have affected their sales. These failures do add up.
Armed with an array of display formats, data collection techniques and frequent sample selections, the analyst can get caught up with the process of data collection rather than the systematic solutions to the problems that are happening. Do you have all of the correct information? Some of the basic steps are often overlooked. The most important thing is don’t forget the obvious. And here that is: confirm the running speed. Analysis cannot be done without correctly knowing the running speed of your machines.
Condition monitoring should be an adaptive process. The analytics needs to be able to determine what piece of data to collect based on the types of data already taken. If one takes an overall velocity reading and it reaches an alarm, then the analyst should be taking some spectral data to try to find out what caused that alarm. Because of the variation, it’s difficult to lay out all of the data types that you want to collect from every machine. Not all data can be collected at the same time. Sometimes vibration analysis doesn’t fit into a pre-programmed route of condition monitoring.
For that reason, you need to be able to take advantage of all of the measurement capabilities available. You should have a good knowledge of your machines and the types of data that can be collected from them. It’s no good taking ultra high-frequency ultrasound data from a bearing that’s fluid-filled, because there’s very little ultrasonic going on there.
Condition monitoring should be just looking for changes in that machine, so that you can then go back using vibration analysis techniques and see what’s happening. The data that you collect that’s required for full vibration analysis takes too long. The best condition monitoring surveys – and I have to stress this – have the best alarms, where all exceptions are issues. The analyst needs to do more analytical work if an exception is breached. That said, any alarms should be accurate. You don’t want false alarms and wasted time on machines that are ‘normal’.
Setting alarms is very difficult. Fortunately, there are lots of guides – international standards guides and API guides to help. How you set the alarms is usually based on historical data or reference charts. Some, such as unbalance and misalignment, are easily sorted, and some need more detective work. To get the best solution pinned down as efficiently as possible, the analyst needs to know if additional data is needed to solve the problem, and if so, what type of data.
the importance of records
You should really record every failure that you have. And the data that you’ve collected about the failure can also help determine whether that particular machine failed for a reason that’s common. It may be that you know that that machine is not rated properly for the job that it’s doing, or that it needs some sort of modification. Root cause analysis and finding out why a machine failed is the proper approach here. You can’tspend all of the time doing that, but simple statistics can always help you and point you in the right direction.
There are four main principles. First is detection: finding patterns in data of changes that indicate that something has gone on. The overall measurements are a summation of the total vibration acting in a particular direction at a given instance in time. They are plotted on a trend, and any change in the levels can be easily identified. This goes beyond just vibration. Infrared thermography, which deals with changes in temperature, is very cheap these days. If you’re doing oil analysis, the objective is to look at the contamination levels or the additive packages and whether they are breaking down; you may need more of it. Ultrasound and acoustic emissions relate to noise levels.
Second is analysis. Having identified that a problem exists, the next step is to identify the specific problem, or problems, for correction. This can be performed through vibration spectral analysis, on velocity, acceleration, bearing envelope spectra or even ultrasound spectra. Spectral analysis involves processing the analogue vibration signal through a digital converter and then through FFT (fast Fourier transform) circuitry. This allows us to see specific amplitudes at the individual frequencies which can identify any machinery problems. If you collect the correct data, you will see the machines’ faults in the spectra: peaks at 1x are due to unbalance; peaks at 2x and/or 3x are due to misalignment; peaks at line frequency and 2x line frequency are due to electrical problems; non-synchronous peaks are due to bearing faults.
If the detection and analysis stages enable detection of a problem, that process has given time to plan for a repair. That’s what condition monitoring is for; it’s not just to see what the current case is, it’s also there to predict when something is going to fail, because you’re seeing it getting worse. By avoiding a breakdown, you can plan the repair and plan the correct tools or materials– spare motors or the right bearings – and even the correct personnel to do the job – does it need a precision balance or alignment? The third stage is correction. By planning the repair, you are keeping downtime is to an absolute minimum, which means you keep uptime at a maximum. That’s what we’re all after, as uptime ensures production of revenue.
Fourth, verification, is the most important; confirming that the work that’s actually been carried out has fixed the problem. Condition monitoring might be telling you to change the motor, but you need to verify that. After any correction work, new readings should be obtained to ensure that all defects have been eliminated, and to establish new baseline characteristics. You may even need to change the alarm levels to suit the new data and the new machine, because the old machine was already part-worn from when you started collecting data from it. Then, if a similar fault occurs, you should be able to see it earlier.
Experience has shown that the machine operator and maintenance personnel who work with the machine every day often provide a good insight to any changes in the machine. The break room is a great place to find out what’s going on. They’re seeing it or hearing it or smelling it before it actually happens.
Ask them these questions: When did the problem start? Was there a sudden increase in vibration, or a gradual increase? Did the machine always run rough? Has it ever run acceptably? Has there been a change in the performance expectations? What sort of changes were made? Were they structural, or was a repair carried out? Were there any electrical changes? Have the pulleys or gearbox been changed with replacements that are a slightly different size? Were they any changes in the machine load or changes in the product that affect the machine load?
You’ll also need to get the machine details; its physical makeup. Ideally this is completed while it is in good working order. Machine details can include digital images or sketches, with accurate input and output speeds. If there are gearboxes, you need to know how many teeth are on each gear.
Recording the sampling locations is usually done in the software. You need to have a standard recording method so it’s easily identified for others. A widely-accepted method of measurement identification is to start with the outboard bearing of the driver (motor/diesel engine/steam turbine) and identify this as bearing number one. (Don’t forget the direction of data collection: horizontal, vertical, axial. Best practice says to measure all three, although in the real world we can’t always do that.) Then you work through the drivetrain until you reach the outermost bearing of the driven unit (pump, fan, generator, etc.), advancing the bearing number or designation for each measurement location.
All of this information is put into the database so you can put frequency markers over data to easily identify problems. Best practice is to have this information before collecting any data, so that the criticality of the machine and the best measurement setups for the type, speed or load of the machine are known.
Analysts are usually eager to jump in and start collecting data without getting a general feel for the machine’s condition. But before starting, look around the machine. Are there any obvious faults or defects that could contribute to the machine’s current condition? You are looking for loose, worn, broken parts, leaking seals, foundation/structure cracks, and excessive lubrication issues.
Then we recommend doing a quick assessment of the machine’s overall condition by taking a series of overall vibration readings. It’s easy to put these against an alarm specification for the machine type based on the international ISO or API standards. They are quick and easy to take, and are usually sampled in all directions. This will allow the analyst to determine the area of the machine with the highest vibration levels. In general, the area with the highest vibration levels are usually the ones closest to the source of the problem. When time is limited, the analyst can focus most attention on just the suspected problem area or machine.
You should also have some sort of overall bearing condition measurements. We use the shock pulse method, but there are other forms of acceleration high frequency detection. These readings should be made on all rolling element bearings. They can help focus attention or quickly eliminate the bearings as an issue.
Best practice says to collect vibration spectrum data (velocity, acceleration, envelope and even displacement, if required at slow speed) in each plane and at each bearing point to provide a complete picture of the frequencies and vibration amplitudes affecting the machine. Avoid taking data off loosely-supported structures like fan shrouds; they will not transmit all of the data but will shake at their own frequencies.
An example of this was a recent measurement of a motor. The horizontal reading was a couple of mm/sec, but the vertical reading on this motor was 20mm/sec. It’s unusual that one direction is massively higher than another. We worked out that there was a structural problem: the motor down-bolts were not fitted correctly. So check the foundations, the mounting block and piping. Check all of the mating interfaces, the background vibration and the floor vibration.
In taking measurements, you also need to make sure you’re taking the correct spectral data. Route-based condition monitoring spectral data is usually optimised for speed. This may mean that the frequency fmax and the lines of resolution of the spectra are not correct for accurate vibration analysis. Select the correct frequency range and resolution needed to accurately identify all of the vibration frequencies.
Finally, phase is a fundamental tool of vibration analysis that is often overlooked. It is essential for differentiating a number of vibration problems, including imbalance, foundation issues, misalignment and even bent shafts. It is usually collected by a triggered input such as a laser tachometer or photocell. The analyst should always record the phase information to help in the diagnostics. The exact number of degrees is generally not needed for evaluation of vibration problems; we allow +/- 30°. Using a clock face drawing at each measurement point on the bearing face would be adequate for finding a fault. For example, if you’re looking for a bent shaft, the phase angle should be the same for all points in the direction that you’re measuring it; then you know the machine is moving in the same direction. As soon as the results are different, there’s something wrong.
BOX: SUMMARY The most important thing is to confirm the running speed Take overall data; collect as much as possible, because it gives you an indication Collect bearing information Check vibration levels at points other than the bearings:the analysis side is If possible, take temperature readings or thermal images. It’s simple to take a Take the correct spectral measurements. There may be different ones for Have a good look around. You can hear what’s going on. You can smell. Is there When you have successes, show off your findings. Keep records. Some people Try to work out cost savings for your finds. It may cost £x to change a bearing,
This paper is an edited version of a webinar presented on 11 November published by Accelix, an industrial internet of things software platform that is a subsidiary of test and monitoring corporation Fluke. It is available to watch for free via www.is.gd/ujunem.