Sensors - May 2003 - Machine Condition Monitoring, Part 1

Early machine monitoring systems simply determined the need for maintenance, but modern systems assist with severity assessment, avoid false positives, and help with process optimization. Here’s a look at one enabling technology for these new systems—dynamic vibration measurement and analysis.

Vibration is often the signal of interest for machine condition monitoring (MCM). These setups are a vital part of preventive and predictive maintenance programs that seek to reduce cost and avoid unplanned downtime. Such setups include measurement hardware and software that acquire and interpret signals generated by operating machinery.

Quantifying Signal Changes
A fundamental role of analysis and signal processing of vibration for MCM is to quantify signal changes. Doing so is important because you can often relate changes in vibration to changes in machine condition. The idea is straightforward: structural changes caused by friction and wear in a machine lead to changes in its dynamic motion and hence the vibration that you can measure.

Measurable vibration is actually a composite of the vibrations that emanate from the components of the machine. The sum of the vibration from the various components constitutes the overall vibration that you measure, and this variety of sources is responsible for some apparent signal complexity when you look at how the signal varies with time.

Consider Figure 1, which shows 1 s of an acceleration signal gathered from a rotating shaft in a rolling-element bearing.

Figure 1. One way to tame the complexity of a dynamic time-domain signal is to characterize it with a single value. Common examples include average, peak, peak-to-peak, and RMS.

The signal is dynamic—its amplitude changes rapidly in time relative to the duration of the acquisition. By themselves, thưse changes don’t necessarily indicate the need for maintenance. The operation of any machine will generate some vibration whether or not it’s in need of maintenance. Analysis and signal processing for MCM assist with quantifying changes in the dynamic characteristics of the vibration.

Vibration Acquisition Requirements
Before deciding on any analysis and signal processing, you’ll need to consider some of the other elements of the measurement process. The nature of the machine you wish to monitor and the signals that it generates will indicate appropriate transducers, signal conditioning, acquisition hardware, and acquisition settings. These elements, in turn, will stipulate suitable analysis and signal processing that your MCM system can apply using software running on a standard computer.

Acquisition starts with a transducer mounted on the machine. Common transducers that measure dynamic vibration include accelerometers, velocity probes, and displacement probes. The transducers convert the physical quantities to a continuous voltage that is eventually sampled by acquisition hardware to create a set of digital samples. The sidebar lists some of the common considerations for selection of acquisition hardware.

After acquisition, you’re back to the problem of gauging changes in the dynamic characteristics of the vibration. One way of doing this is to find one value that describes the dynamic signal. For instance, you might measure the overall signal level by determining such values as peak, peak-to-peak, average, or RMS. This approach reduces the complexity of the vibration signal to a single slowly varying value that you can then compare with a predefined limit to determine if manufacturer’s specifications are exceeded and maintenance is required.

Frequency of Vibration
Although describing vibration with a single value is useful in its simplicity, it ignores much of the information contained in the signal. An alternative is to examine the frequency content of the signal (see Figure 2).

Figure 2. A power spectrum can be thought of as a recipe for how to construct a signal using a set of sinusoidal components (A). The peaks within the spectrum are a measure of the amplitude of a particular sinusoidal component at a specific frequency (B).

The top graph on Figure 2 shows an FFT-based power spectrum, which is one of a several common types of frequency-domain analyses.

When you examine the frequency content of vibration or any other dynamic signal, you see an alternative representation of your time-domain signal. Software tools with high-level functions (e.g., LabVIEW) convert signals to the time-frequency domain and analyze spectral content. For frequency analysis, the new representation is a recipe for how to construct the original signal using a set of sinusoids with varying frequency and phase. With a power spectrum, the recipe is a list of ingredients (the frequencies on the X axis) and the amounts (the amplitudes on the Y axis).

Frequency analysis relies on superposition, which is the idea that your signal is actually a sum of many sub-signals. Superposition fits nicely with vibration analysis because many of the components that sum up to make the measured signal will be the result of repetitive motion of the elements that make up our machine. This motion will create vibration components at frequencies that you can relate to the rotational speed of the machine you’re monitoring.

Vibration and Speed
Perhaps the simplest example of the association between machine action and a particular frequency component is the vibration signal generated by a misbalanced shaft. For the example, consider examining the vibration that you measure from an accelerometer attached to a mounting point in which the shaft rotates. With each rotation, the accelerometer will experience a periodic cycle of motion related to the force generated by the misbalance. This motion will contribute to the level of the component of the measured vibration at a frequency equal to the rotational speed of the shaft.

Interaction of articulated machine components can also generate vibration components at specific frequencies. A pair of meshed gears is a good example. When you attach an accelerometer to a mounting point on the gearbox that contains such a pair, the vibration you measure will be the sum of frequency components that result from actions and interactions of the gears, as well as other elements in the gearbox. With a rotational speed of F_rot, you might identify several frequencies of interest for gears with, for example, 29 and 17 teeth:

Many other associations between frequencies and machine components exist. For instance, bearing manufacturers publish tables of data that tie operating speed with elements of the bearing models that they offer. Turbines, blowers, and other machines with rotating blades will show a blade-pass frequency component at multiples of the number of fan blades. Goldman includes a listing of other common associations.

Comparing Frequencies
If you know something about the design of the machine you’re monitoring, you can associate what you see at specific frequencies with elements of your machine. For MCM, this extra information lets you focus your monitoring on specific machine elements by trending the individual signal levels at the frequencies of your elements of interest. The technique is more robust than trending an overall signal level because you can focus on a selected subset of the frequency components of your signal and ignore others.

Trending with frequency analysis starts the definition of a baseline, or signature, that specifies the minimums and maximums that you expect for each of the frequency components of interest. By defining this set of limits when your machine is in good working order, you have a basis for comparison for either continuous or periodic monitoring. You can also set up a baseline for later comparison with other types of frequency-domain analysis.

Figure 3 shows displays created in LabVIEW based on high-level sound and vibration analysis tools, including fractional-octave analysis.

Figure 3. Fractional-octave analysis breaks the frequency domain into a set of constant bandwidth octaves or fractions of octaves. You can apply the tool to monitoring by establishing high and low limits.

Fractional-octave analysis is an alternative that examines the frequency content of a signal by dividing the spectrum into well-defined regions called octaves or fractional octaves. The results are commonly presented on a logarithmic frequency scale as a bar graph, with bar heights that correspond to the energy contained in each octave band. Because the spacing between the upper and lower frequencies of each octave band is set so that they have a 2:1 ratio, the distribution of bands appears linear despite their presentation on a logarithmic frequency scale.

Standard Comparison
The scheme proves useful for several reasons. For one, there are IEC and ANSI standards that define how to perform fractionaloctave analysis. As such, if you work with a standards-compliant fractional-octave tool, you can expect consistent/repeatable reTults that you might compare with other measurements taken under the same standard.

Octave analysis is also useful for monitoring because it simplifies comparison. With a standard power spectrum, it can prove difficult to completely characterize every frequency component of the vibration that your machine generates. With a monitoring system that examines specific frequency components of a power spectrum, there’s a chance that an important frequency component will be among those that aren’t being monitored. By grouping frequencies into bands, octave analysis enables monitoring that watches more of the frequency domain.

At the same time, fractional-octave analysis still offers more frequency discrimination than a level measurement that gauges the aggregate level of all signal components. Your comparison is still a function of signal-frequency content, and you can program your monitoring system to ignore frequency bands associated with noise or pay extra attention to frequency bands associated with vital machine components.

For example, if you’re monitoring a machine that’s near a loading bay, you might have the MCM system ignore frequency bands associated with the sound and vibration of trucks or loading. You also could have your system flag warnings when frequency bands associated with bearing frequencies change. By ignoring noise, and watching only signals of interest, your MCM system will be less prone to false positive warnings.

For Further Reading
Goldman, Steve. 1999. Vibration Spectrum Analysis: A Practical Approach, 2nd ed., New York: Industrial Press, Inc., ISBN 0-8311-3088-1.

Common Selection Considerations for Dynamic Vibration Acquisition Hardware

The dynamic range characterizes how well you can distinguish a low-amplitude signal in the presence of high-amplitude signals. A high-dynamic range ensures that you won’t miss low-amplitude components that are sometimes key to a successful analysis.

Anti-aliasing filters should be applied prior to sampling to avoid signal aliases that result from sampling. Anti-aliasing filters pass low-frequency signal components and block high-frequency components. So they will also help to reduce errors caused by the presence of wideband noise.

The sampling rate determines the maximum frequency that you can examine from your samples. The sampling frequency is usually driven by the maximum rotational speed of the machine and the highest order of interest. As a rule of thumb:
min. sampling rate = 5 × max. rotational speed × max. order of interest.

Simultaneous sampling ensures that you are gathering useful cross-channel phase information and lets you calculate orbit plots, which are a common type of analysis applied for MCM.

Multiple input gains let you take advantage of the full dynamic range of the analyzer on a per-channel basis.

Voltage-mode excitation is a common requirement for accelerometers and microphones. Vendors of these transducers have assigned a variety of trade names for such excitation, including ICP, Isotron, and Deltratron.