New Tools for Vibration Condition Monitoring Chris Engdahl, GE Energy, Measurement & Control Solutions

5th Reliability, Asset Management & Safety (RAMS) Conference 2011

Introduction:

Machinery vibration monitoring is no longer simply a proportional measurement for operator surveillance or machine protection, but a fundamental input to a mature predictive maintenance strategy. When correctly applied, vibration monitoring will accurately detect root cause failure modes, enabling cost-effective maintenance planning, optimizing asset availability and profitability.

Conventional vibration condition monitoring techniques have, for the better part of the last 30 years, been based on spectrum analysis and simple variations of it. Advanced diagnostic methods have been well documented, but typically reserved for experimental and academic applications using offline tools like MathWorks’ MATLAB. Recent advances in online instrumentation capabilities, however, has enabled such methods to be implemented in real time, unlocking new possibilities in managing a far broader range of failure modes in complex rotating machinery.

A great example of where predictive monitoring demands advanced diagnostic techniques is modern wind turbine monitoring. From low speed main bearings, through multi-stage planetary gears, to a medium speed generator, the combined protection and monitoring requirements present a real challenge for a single online monitoring system.

This paper introduces a number of advanced vibration monitoring techniques which are now being effectively applied online in real time. These methods include:

Dynamic Energy Index (DEI) Kurtosis Skewness

Cumulative Impulse Acceleration Enveloping (AE)

Sideband Energy Ratio (SER) High Pass Filtering

The theory behind these measurement types is explained, showing typical examples of how these signal processing methods have been applied. Initial applications and examples are shown from the wind turbine industry, where these techniques are ideally suited and have been proven to enable the accurate and early detection needed for effective maintenance optimization.

Dynamic Energy Index (DEI)

Complex gearboxes have a large number of components and can produce a rich mixture of operating and fault frequencies. A good example is the planetary gearboxes employed in the wind power industry. These gearboxes typically have several increasing stages: one or more planetary stages followed by one or two conventional stages. Each stage has a characteristic set of mesh frequencies and harmonics that is different from what is produced in other stages. Each shaft in the gearbox has two or three rolling element bearings, each producing its own set of fault frequencies. Finally, impacting events (due to rolling element bearing faults or broken gear teeth) can excite gearbox casing structural resonances. These structural frequencies tend to be well above those produced by bearing faults or meshes, typically 4 to 10 kHz or higher.

The Dynamic Energy Index (DEI) was developed to provide a relatively simple set of five numbers that characterize the vibration energy in spectral bands that roughly correspond to bearing fault frequencies, mesh frequencies in the different stages, and structural frequencies (Table 1). The lower four DEI band boundaries are adjusted with input shaft speed. The gearbox structure does not change with speed, so the highest DEI band boundaries are not adjusted for speed.

In gearboxes, bearing and mesh loads are a function of the torque being delivered through the gearbox. High torque can produce high loads, which in turn will produce higher vibration levels. To compensate for this affect, the DEI spectral energy calculations in the monitor are divided by torque to help reduce the influence of torque on the DEI values. Finally, a DEI multiplier is used to produce a number that is easier to use. DEI values have no physical meaning in and of themselves; they are used to trend vibration levels over time.

DEI Variable Full Speed Freq. Range

Description

DEI Bearing 0-25 Hz Bearing fault frequencies

DEI Planetary 28-130 Hz Planetary stage mesh frequencies

DEI Intermediate 145-570 Hz Intermediate stage mesh frequencies

DEI High 590-3300 Hz High-speed stage mesh frequencies

DEI Structural 4 -10 kHz Gearbox structural frequencies

To summarize, DEI is calculated through a number of steps. For each DEI band,

• Sum the squares of the individual bin amplitudes • Divide the sum by the generator torque • Multiply by a factor to produce a more useful number

Modern vibration monitors allow each DEI band value to be independently alarmed and trended, with alarms set independently for different machine operating modes (ie: different power ranges). In the Bently Nevada ADAPT.Wind monitor, for example, it is possible to set a total of 25 DEI alarms, 5 for each DEI frequency band.

On a typical gearbox DEI is calculated separately for each accelerometer mounted on the gearbox (commonly one each for planetary gears, intermediate shaft, and high-speed shaft.)

Kurtosis

Kurtosis is a measure of the “peakedness” or flatness of the statistical distribution of vibration amplitude points in a waveform record relative to a normal distribution (bell curve). The name originally comes from the Greek word kyrtosis, which refers to convexity or curving.

Technically, Kurtosis is defined as the “Fourth Moment of the probability function normalized by the square of the variance of the signal.” It is a very useful parameter for vibration analysis, being more sensitive than the traditional “Crest Factor” measurement for detecting impulse events in the vibration signal. Modern online vibration monitors are able to calculate and trend this value online, in real-time, a process historically requiring considerable offline computational power.

Figure 1. Typical DEI bands and descriptions (example only). The first four band boundaries are adjusted for turbine speed.

Kurtosis is low (1.5) for a pure sine wave, and higher for real-world vibration signals. A machine in good condition would typically exhibit a Kurtosis below 5.0. Higher kurtosis means more of the variance is due to infrequent but very extreme deviations from the norm – such as large impulse spikes.

The gearbox with the damaged bearing has a higher value of kurtosis, but even 14 is a fairly low for real-world failure indications. Some severely damaged gearboxes actually generate kurtosis values in the hundreds!

Skewness

Like Kurtosis, Skewness is also a parameter derived from statistical evaluation of the acquired timebase waveform record. Mathematically it is defined as the Third Statistical Moment of the data set.

Skewness indicates the symmetry of the vibration waveform points about a neutral axis. In the example above, it was shown that a mechanical impulse (due to bearing or gear damage) causes a vibration resonance response, ringing down normally. In that example, the Skewness value would be close to zero.

Figure 2. Kurtosis examples from two identical gearboxes. First plot shows a healthy gearbox, The second indicates a damaged bearing.

Figure 3. Example of a Vibration waveform with High Skewness. In this case, electrostatic discharge produced a Skewness value of around -2.0.

In the example shown, the accelerometer is responding to electrostatic discharge through an unearthed bearing (in this case, it was the result of a problem with shaft grounding brushes). High Direct vibration, High Kurtosis, combined with high Skewness is a conclusive indication of electrostatic discharge on susceptible machine types.

For machines which are known to be susceptible to electrostatic discharge related failure modes, Skewness measurement is a very effective technique to generate corrective action notifications to avoid serious and costly machine damage.

Cumulative Impulse

Cumulative Impulse is a set of measurements that detect and trend the passage of debris particles through planetary gearboxes. Planetary gearboxes have traditionally been challenging to monitor as the sun gear, planet gears, and planet bearings have to transmit defect signals through multiple parts and connections to arrive at the outer fixed ring gear where the accelerometer is mounted. This complex signal path can significantly attenuate vibration signals. In addition, during operation, the planets move about the sun, constantly changing the location of the teeth engagement and changing the vibration transmission path. This motion modulates the amplitude of the defect signals.

When damage begins to occur in the planetary stage, metallic debris well be shed from these parts. Usually this wear debris will pass through planetary stage mesh, causing secondary damage.

The debris particle causes a vibration impulse when the debris particle gets trapped between meshing teeth. This produces a vibration impulse that is transmitted through the entire planetary gear set to the outer ring gear. The result is an impulse of the gearbox casing followed by a decaying vibration at the structural natural frequency.

This characteristic impulse-response signal is detected by the vibration monitor and is reported as simple trending variables. The monitor studies the planetary accelerometer waveform in real time, looking for impulse response events. For events above a detection threshold, it reports the amplitude of the impulse, and calculates the number of impulse events that have occurred in the past hour.

Cumulative Impulse (CI) information in three variables: Cumulative Impulse Count, Cumulative Impulse Energy, and Cumulative Impulse Rate.

• Cumulative Impulse Count is the most fundamental of the measurements. Each time a new impulse is detected, the count is increased by one. Trending this measure over time reveals how many particles above the detection threshold have appeared in the planetary stage mesh. As the planetary stage progresses toward failure, the number of debris particle detections should increase more rapidly.

• Cumulative Impulse Energy is a measure of both the number and the amplitude of the accelerometer impulse signals. The impulse signal amplitude is related to the size of the particle, so larger particles will produce higher vibration impulse amplitudes. These amplitudes are accumulated to produce Cumulative Impulse Energy. When this measure is

Figure 4. Vibration impulse caused by trapped debris

trended over time, the slope is a measure of both the number and size of particles that have appeared in the planetary mesh, giving a clear indication of severity of damage in the gearbox planetary stage..

• Cumulative Impulse Rate measures the number of impulse events that have occurred over the past hour. It is derived from the slope of the Cumulative Impulse Count plot curve. A high Cumulative Impulse Rate indicates that a relatively large number of debris particles have been detected in the mesh in the past hour. Cumulative Impulse Rate is well suited for alarming so that operators can be alerted to sudden changes in debris particle detection rates.

To reduce the chance of false alarms, the Cumulative Impulse system only detects debris particles that produce a signal above a minimum threshold. Thus, it is possible that relatively small particles will not be counted. In the earliest stages of deterioration, the planetary stage may shed particles too small to be detected but that may be detected by other means, for example, inductive oil debris detectors or filter inspection. As damage progresses, larger particles will be generated, the Cumulative Impulse system detection threshold will be exceeded, and the Cumulative Impulse variables will begin to show changes.

Acceleration Enveloping (AE)

Acceleration Enveloping (AE) is a signal processing technique that reveals the presence of periodic “impact-like” activity that is buried in a complicated timebase waveform. AE is already used extensively in the “walk-around” Condition Monitoring community to identify and diagnose rolling element bearing degradation. It is also known as High Frequency Resonance Demodulation (HFRD).

When a rolling element bearing has a defect, it tends to produce a periodic impact-like event, every time the rolling element (ball or roller)l encounters the spall. Thus, we expect to see one such event per ball pass, and a series of such impacts will occur at the corresponding Inner Race or Outer Race Ball Pass Frequency.

These are impulse type shock events which excite the bearing support structure just as if a hammer blow were applied to the structure. Once excited, the structural resonances will freely vibrate at its natural frequency (typically several kHz), and the vibration will decay over time. With each ball pass impact event, the structure is excited again, repeating the process.

Figure 5. Impulse signal processing for Cumulative Impulse parameter calculation.

The signal processing objective is to extract the impact repetition frequency from the high frequency carrier signal. To do this, the acceleration signal is first filtered using either a high pass filter. The low frequency corner of the filter is set to exclude any frequencies not associated with structural “ringing”. All rotor frequencies and REB ball pass frequencies are normally excluded.

After filtering, the signal is full wave rectified and then sent through a low pass filter to remove the high frequency carrier component. What remains is an enveloped triangular waveform signal that represents the amplitude of the original vibration signal over time, or the amplitude envelope. This signal is then subjected to a Fast Fourier Transform, and a spectrum of the envelope signal is produced and displayed.

AE depends on the existence of high frequency, poorly damped, structural resonances for its use. Accelerometers must be mounted carefully and correctly. Poor mounting of accelerometers can reduce the mounted resonant frequency of the accelerometer and unintentionally exclude important structural carrier frequencies. Even paint on the accelerometer interface can destroy the high frequency signal.

Trending and alarming on the Enveloped Acceleration amplitude is very effective in detecting bearing faults, however good baseline information is needed to establish a reliable detection threshold. There is no standard amplitude threshold value, as this is specific to machine design and sensor mounting arrangement.

DefectRepetitionFrequency

StructuralRinging

Impact-LikeEvent

Figure 6. Bearing elements generate a localized impact force and structural resonance response each time the damaged region is engaged.

Figure 7. Diagram of the Acceleration Enveloping signal processing method.

Sideband Energy Ratio (SER)

Sideband Energy Ratio (SER) is calculated from high resolution spectrum data. Each spectrum is typically created from timebase waveform data generated by an accelerometer sensor and collected by the monitoring system. Waveforms are synchronously sampled so that the sampling frequency tracks changes in speed. This sampling technique produces narrow spectral lines of speed dependent frequencies, like gear mesh frequencies and associated sidebands, for variable speed machines and is essential to accurately calculate SER.

To understand SER, first sidebands must be understood. Sidebands appear in a spectrum around a center frequency and generally occur as a result of an amplitude modulation of that center frequency.

A damaged gear tooth within a gearbox can cause this phenomenon. As shown in the example above, the damaged tooth will produce an amplitude modulation of its associated gear mesh frequency each time it passes through the mesh. That amplitude modulation occurs once per revolution of the shaft that the damaged gear is mounted on. When viewed in a spectrum, this amplitude modulation shows up as a series of spectrum lines at evenly spaced frequencies on either side of the central gear mesh frequency.

Once the spectrum is generated, the SER algorithm sums the amplitudes of the first six sideband peaks on each side of the center mesh frequency and divides by the amplitude of the center mesh frequency.

Figure 9. Borescope image of damaged gear (Left), and vibration waveform image (Right) Note the impulse generated each revolution of the intermediate shaft.

Figure 8. Examples of amplitude modulation for gearbox vibration (Waveform & Spectrum). Plot on Left may indicate unbalance, plot on Right shows typical gearing problem.

SER=∑ Sideband Amplitude

� 6i=1

Center mesh frequency amplitude

SER is sensitive to the sideband amplitudes relative to the center mesh frequency. For a healthy gear mesh any sidebands will have small amplitude compared to the center mesh frequency or may be nonexistent resulting in a low SER. The SER value is typically less than 1 for a healthy gear mesh. As damage develops on a gear tooth that passes through the gear mesh, the sidebands will increase in amplitude as well as number resulting in a larger SER value. The SER value is calculated for the first 3 harmonics of each fundamental gear mesh frequency

High Pass Filtering

While many of the monitoring and diagnostic parameters discussed in this paper already provide excellent detection of specific root-cause failure modes, there is often a need for the vibration analyst to access further diagnostic information to verify the suspected problem. The timebase waveform is often used to validate the nature of a fault prior to committing to a corrective action plan.

A common problem with gearbox monitoring, however, is that, for data acquisition A/D processing, the dynamic range is almost entirely consumed by gearmesh frequencies and their harmonics, resulting in poor signal-to-noise resolution for the higher frequency content which would reveal the actual impacting events in the time waveform.

Figure 10. Spectrum of the above vibration waveform showing SER calculation for the affected shaft. High SER value (above 1.0) is a clear indication of gear damage.

Figure 11. Typical gearbox spectrum showing gearmesh content as the dominant frequency components, relegating high frequency resonance response behaviour into the noise floor.

By applying a high-pass filter to remove all vibration components below 5kHz, a second vibration waveform can be acquired, with good resolution for the repetitive events related to mechanical defects. Whilst not leveraging specialized signal processing or computational techniques, this method does rely on the analogue filtering and A/D acquisition being done directly at the monitor itself, while display and interpretation of the waveform shape is performed using the condition monitoring software system.

As with the raw waveform, statistical analysis may be carried on the high-pass filtered waveform to detect the presence of impacting or other fault types using Kurtosis, Skewness and associated calculations, within the monitor itself. Visual inspection of the high-pass filtered waveform can confirm the suspected problem, as shown in the below example where three periodic impacts represent three damaged teeth on a high speed gearbox pinion.

Summary

Modern machinery vibration condition monitoring techniques are able to support complex, yet accurate and reliable failure mode detection techniques, which can operate in real-time onboard the field-mounted monitor hardware. This not only ensures best-in-class predictive maintenance can be applied effectively, but also greatly simplifies the use of such systems. Fault detection can be largely automated using the new diagnostic parameters to trigger investigation and repair actions, freeing up experienced analysts for value-added activities.

Figure 12. Comparison of acquired vibration waveforms for Raw signal and High-Pass Filtered signal. The latter waveform shows the impacting event with very good resolution.

Figure 13. Analysis of the High-Pass filtered waveform clearly shows three impacts per revolution. Three damaged teeth on the High Speed pinion gear were found.

As these capabilities enter more mainstream machinery monitoring applications, end-users should be encouraged to adopt the techniques to better deploy their predictive maintenance strategies. In the wind energy sector, this is particularly relevant, where lead times for replacement gearboxes, bearings and even mobile cranes, can often be up to 6 months and long-term maintenance planning information translates into immediate financial benefit.

References:

“Wind Turbine Signal Processing” PPT presentation file. 2010 Charlie Hatch (Jan 2010 GE Energy FAE technology briefing)

“Detection of Wind Turbine Gear Tooth Defects Using Sideband Energy Ratio™”, 2011 J. Hanna (Lead Engineer/Technologist. GE Energy Bently Nevada) Charlie Hatch (Principal Engineer/Technologist, GE Energy Bently Nevada) Matt Kalb (Lead Engineer/Technologist, GE Energy Bently Nevada) Adam Weiss (Senior Product Manager, GE Energy Bently Nevada) H. Luo (Senior Technical Manager, GE Energy Bently Nevada)

“A Successful Application Of Wind Turbine Condition Monitoring Using Acceleration Enveloping”. 2009

Charlie Hatch (Principal Engineer/Technologist, GE Energy Bently Nevada) Adam Weiss (Senior Product Manager, GE Energy Bently Nevada)

“Acceleration Enveloping Principles” Application note. Bently Nevada Corporation, 2002 Charlie Hatch (Principal Engineer/Technologist, GE Energy Bently Nevada)

“Cumulative Impulse” Application note. GE Energy MCS, 2011 Charlie Hatch (Principal Engineer/Technologist, GE Energy Bently Nevada)

“Back to Basics #6 - Data Acquisition (Rev1)”, GE Energy MCS Internal training presentation, 2010 Clyde Dooley (Field Application Engineer, GE Energy Bently Nevada) Alberto Jahn (Field Application Engineer, GE Energy Bently Nevada) Petri Nohynek (Field Application Engineer, GE Energy Bently Nevada)