At the onset, defects in rolling element bearings can be inconspicuous. But bearings run hard, and minor flaws can rapidly escalate. With the right training and equipment, however, you can spot those easy-to-miss symptoms before they do serious harm.

Vibrations, anyone?

Bearing fault detection has came a long way since the days when mechanics used to check for flaws by placing a screwdriver tip on the bearing cap while pressing the handle against the bone structure behind their ear.

Portable vibration measuring equipment allowed a more technological approach, and troubleshooters could now see the vibrations they used to listen for. Not surprisingly, however, there were still quite a few bugs to work out.

Some people began to connect bearing failure to sensor resonance; flaws in the bearing tend to generate energy that can excite the natural frequency of accelerometers. Operators would single out a bad bearing based on such a response. Sometimes they were right and sometimes they were wrong. If, for example, other parts of the machinery created high-frequency vibrations and caused a similar accelerometer reaction, a "blameless" bearing could get replaced.

The next step in the progression was to take velocity measurements of spinning bearings and look for specific frequencies generated by individual bearing elements. This improved accuracy, but even the best technicians would miss things when very slow-rotating bearings were involved.

Today, thanks to better instruments and special functions like enveloping algorithms, bearing analysis accuracy is nearly 100%. (Enveloping algorithms are formulas that sum the harmonics in a signal, and suppress the noise.) A few bad bearings still get misdiagnosed, but they are rare, and there are ways to find even these.

Get analytical

Velocity measurements indicate the speed of vibrations that result from bearing flaws and roughness. Velocity, rather than acceleration, is best when low-frequency vibrations are expected, typically below 60 Hz. A common way of obtaining velocity is by integrating the output of an accelerometer. The measurement is further enhanced by data collectors with enough dynamic range to pick up faint signals, such as those generated by rolling elements passing over flaws in the races, or by defects in a ball or cage.

One problem, however, is that when higher frequencies are a concern, velocity levels don't usually climb very high, even with very large, deliberately induced flaws. When bearing elements vibrate and displace because of defects, they don't generally get a lot of speed going, even though the changes in speed can be abrupt.

In one case, a substantial flaw only triggered a velocity response of 0.0027 ips (at around 107 Hz) at a shaft speed of 1,800 rpm – although far from invisible, it wasn't enough to get caretakers anxious about performing maintenance. Now consider the same bearing with a shaft speed of 180 rpm, as on a paper machine. The amplitude will read a mere 0.00027 ips for a bearing that should be replaced. Clearly something else is needed to evaluate the condition of rolling element bearings.

Extra refined

The acceleration signal for a bearing with a known flaw in the outer race reveals a conspicuous amount of energy in the 3,400 to 3,800 Hz frequency band, due to the test accelerometer resonating from contact with the bad bearing. An "old-fashioned" way of evaluating bearings actually relies on sensor resonance to produce such an amplitude spike. This can also be misinterpreted, for bad bearings aren't always the culprit; high frequency disturbances elsewhere in the machinery could induce a similar accelerometer response.

In the last few years, the mathematical signal processing method known as enveloping has been incorporated into portable data collectors. This computational technique, initially developed in Europe back in the 1970's, used to be limited to lab analysis.

Enveloping uses high-pass filters to collect the signals generated by rolling bearings with flawed rings, cages or rollers. Mathematical manipulation enhances the repetitive signal components, while suppressing random signals. The process then sums the energy, passes it through a Fast Fourier Transform (FFT), demodulates it, and presents it in the selected frequency range as an enveloped acceleration (gE) spectrum.

By using the enveloped signals coupled with demodulation, the resultant FFT can have a starting point of zero without any electronic "ski slope" as seen in velocity measurements, and without the roll-off at the low end of an accelerometer's range.

Enveloped acceleration really stands out as a dynamic signal analysis technique for accurately assessing slow-turning bearings. A gear box shaft output housing turning at 8.3 rpm is measured using velocity, plain acceleration, and enveloped acceleration. The corresponding traces show the progression in signal processing techniques.

Using an enveloped Fourier transform brings out very low signals, but the transform itself also provides positive evidence of bearing damage. A rotating bearing free of defects translates into a sine wave. With damage present, the sine wave is clipped or truncated, giving it a Fourier transform consisting of the fundamental frequency plus harmonics. Harmonics in the FFT of a bearing indicate trouble.

Enveloped acceleration is used to analyze a mill vessel with an extremely low cycle rate of 0.5 rpm. At this speed, the frequency spectrum will not provide the necessary information, so the time domain is used. A harmonic marker defines the time interval between energy pulses, and establishes the frequency, which turns out the same as the ball pass frequency of the outer race. Compared to a new, similar bearing's time waveform, (not shown), the used bearing has an enveloped acceleration amplitude nine times greater. The lack of energy in the first 10 seconds is due to the race being deformed and not always being contacted by the rollers. Although the large amount of energy in the frequency measurement suggests trouble (a new or "perfect" bearing would give a very low, flat signal), specific bearing frequencies are difficult to sort out.

Using the enveloping method, it's possible to evaluate bearings rotating very slowly, down to 0.5 rpm. A side benefit is that it can also detect looseness. If the rotating components of a machine are loose, or if the machine itself is loose on its base, the sine wave is again clipped, and shaft speed harmonics crop up in the FFT display.

Field report

One field problem frequently encountered with bearings is contamination of the lubrication system. This can often be detected using a technique that collects signals with an acoustical emissions sensor and processes them with a special algorithm. It lets the engineer "hear," through a visual representation, the acoustic output from rolling elements passing over and crushing contaminants. This signal processing method, called Spectral Emitted Energy (SEE), is also helpful in detecting bearings with insufficient lubrication, where there is slight metal-to-metal contact.

Here's an example of a flaw induced into the bearing of a dc motor. The motor speed was then increased in increments from 50 to 3,600 rpm with enveloped acceleration measurements taken at various intervals. The plot shows the influence shaft speed has on enveloped acceleration amplitude.

In one situation with a new fan, the bearing became hot enough that it was uncomfortable to touch after operating for just ten minutes. With the bearing running hot, SEE analysis showed the even harmonics of the ball spin frequency at amplitudes standing well above the noise floor. (Why the system displayed only the even harmonics is still a mystery.)

>After replacing the grease pack and purging the system, the fan was run for an hour before collecting another set of SEE data. The ball spin frequencies had smoothed out completely, and the only notable signals were the ball pass frequencies of the outer race. But these were of a low level, and therefore of little concern, likely due to a small amount of contamination embedded in the outer race. Another reading using enveloped acceleration also showed very low amplitudes. A follow-up lab test of the grease revealed minute particles that have yet to be identified.

Robert M. Jones is Principal Applications Engineer with SKF Condition Monitoring, San Diego.