How to identify failures using spectral analysis

Vibration analysis is one of the most widespread and effective techniques for detecting potential failures in rotating assets and is part of predictive maintenance, as can be seen in the well-known DIPF curve (figure 1), which presents the different phases that an asset goes through, from its design to the end of its useful life.

The DIPF curve is a generalization of the PF curve, which also considers the design and commissioning phase of the equipment.

There are several technology options for collecting vibration data on assets. In this paper we will discuss data collection using wireless sensors fixedly installed on critical components.

This type of measurement device generally allows the collection of two types of data: global or continuous data, and spectral data.

Global data allows basic monitoring of a machine’s operating levels, quickly pointing out any deviation in the expected pattern. An example of such a case is shown below.

With the use of Dynamox’s wireless sensors, the client could clearly notice the evolution in the vibration of an engine, with a significant increase in the RMS speed levels in the axial shaft of the sensor installed in the coupled side of this equipment.

Although important for quick detection of deviations and with a very visual ability to show when something is out of normal operation at a monitored point, this type of chart is usually not enough to point out what the failure mode is, i.e. what the root cause of the problem is.

To achieve this level of detail, spectral data collection and analysis comes into play. This type of data usually allows an assertive diagnosis to be made about the type of failure and its severity. In this way, the most appropriate corrective actions can be planned and executed by the maintenance team.

How do failures appear in Spectral Analysis?

Spectral analysis consists of investigating signals in the frequency domain through the Fourier Transform applied to the signal originally collected in the time domain (waveform).

The waveform itself (signal in time) contains relevant information for analyzing the condition of a component and can be used by analysts.

The waveform is represented by a sum of sines and cosines of different frequencies and amplitudes and, after processing, gives rise to the spectrum of this signal.

The presence of dominant frequencies in the spectrum, or the concentration of energy in frequency bands, can be indicative of the presence of failures. Each type of failure manifests itself in a different way on the analysis graphs and modifies the spectral signature of the machine.

Below we will point out some examples, starting with an unbalance failure present in an engine monitored with Dynamox’s monitoring system.

This is an engine that operates close to 3600 RPM and presented an evolution in vibration levels in the month of August 2021. When carefully analyzing the waveform and spectrum obtained from the sensor installed on the coupled side, it was possible to check the root cause.

This is the original waveform obtained, in time, transformed to velocity. When converting to the frequency domain, one can see the high dominance of vibration at the rotational frequency of the asset, in this case 3492 RPM, or 58.20 Hz, as can be seen in the image below.

The dominance and high levels of vibration at the rotational speed of the machine (in its first harmonic) is precisely the form in which unbalance usually presents itself.

This type of fault is more present in the vertical and horizontal shafts, and less active in the axial shaft. This example shows exactly this case, where there is a dominance of horizontal vibration (blue), followed by vertical vibration (yellow), and virtually no vibration in the axial shaft (pink).

Furthermore, by analyzing the history of this machine, one can see the evolution of the unbalance over time through the spectral cascade of RMS velocity.

Let’s look at other examples:

The following figure shows the spectrum at the bearing housing monitoring point of an electric motor. A peak at approximately 40 Hz, corresponding to 2x the rotational frequency of the shaft, can be seen prominently.

The presence of the 1 x RPM frequency is also noted, although at a lower intensity than the 2 x RPM frequency. This feature in a spectrum indicates a parallel misalignment on the drive shaft.

This results in stresses from the motor’s magnetic field, evidenced in the spectrum by excitation at the rotor (at 1080 Hz) and stator (at 1500 Hz) groove frequencies, along with sidebands at the rotational frequency of the shaft.

Another fairly common failure is the occurrence of cracked teeth in gearboxes. In this case, the severity of the failure can be determined based on the presence of harmonics of the gear frequency, as shown in the figure below.

In addition, the cracked teeth generate modulations in the gearing frequency relative to the rotation of the defective gear. Thus, another symptom will be the presence of sidebands of the damaged gear rotation with many harmonics.

The Harmonics marker allows you to precisely identify the location of these harmonics, with the fundamental frequency at 343.75 Hz.

The identification of faulty gearing can be done based on the spacing of the sidebands around the gearing frequency.

The Sideband marker allows you to identify that the sidebands are spaced at 5.08 Hz, which corresponds to the rotational speed of the output shaft, and therefore corresponds to an output gear failure.

What to do when the fault does not appear in the spectrum?

In some cases, it is possible that the fault is not as visible in the waveform or spectrum, which requires more advanced signal processing tools.

For example, bearing failure frequencies may be masked by other machine components, making diagnosis difficult. See, for example, the waveform of a motor bearing housing in the following figure.

The spectrum reveals the presence of several peaks and harmonics at different frequencies, but the application of the bearing failure frequency marker reveals very low amplitudes at these frequencies.

In this context, it is possible to perform signal demodulation using the Envelope technique, which consists in obtaining the envelope of the waveform after applying a bandpass filter. Different filter frequencies are available in the Web Platform to ensure the best failure diagnosis.

As shown in the figure below, the spectrum of this envelope reveals the presence of harmonics in the failure frequency of the bearing outer race, requiring the monitoring of the evolution of this type of failure. More information on bearing fault detection can be found in this article.

The envelope technique is available on the Web Platform, along with a bank of failure frequencies (BPFO, BPFI, BSF and FTF) of almost 70,000 bearings of different makes and models, allowing frequency markers to be added to the spectrum to facilitate interpretation of the results.

How to set up spectral data collection?

Successful spectral analysis is only possible with proper spectral setup. This impacts not only the quality of the collected signal, but also the ability of the spectrum to provide relevant clues for asset diagnosis.

The dynamic range should be selected based on the expected amplitude levels in the asset.

Low dynamic range values can result in saturated signals, so that the evolution of the failure will not be captured.

On the other hand, high dynamic range values in assets that, even in the presence of failures, do not reach such vibration levels, can result in signals with low resolution, i.e., little ability to distinguish different amplitude levels, thus impairing the quality of the spectrum.

More details about the optimal dynamic range setting for your Dynalogger (Dynamox sensors) can be found in this article.

The sampling frequency determines the maximum identifiable frequency in the spectrum: the Nyquist frequency.

Thus, for assets that have failure modes that manifest themselves at high frequency, the highest possible sampling frequency is desirable.

In other cases, the most critical failure modes may occur at lower frequencies, so a lower sampling frequency is still adequate. This allows the collection time to be increased and consequently increases the spectral resolution, i.e. the ability to distinguish between distinct frequencies in the spectrum.

The combination of sampling frequency and collection duration determines the number of lines in the spectrum, as seen in this article.

DynaLoggers

Dynamox’s family of wireless vibration and temperature sensors, the Dynaloggers, is prepared to meet different demands and machine types. More technical information on the Dynaloggers is available in the product datasheets.

Dynamox offers consulting services for the purchase of the appropriate wireless sensor to perform spectral analysis and fault identification in specific equipment and provides customer support for the best use of the Web Platform and its features.

Contact us today and request a quotation.

Dynamox’s sensors, the Dynaloggers, have the possibility to collect global levels, to monitor the equipment’s conditions, and also collect spectral data, which is sent to the Web Platform, using an App or in an automated process via Gateway.

Spectral data collection is configurable for each Dynalogger. In this case you need to select which axes will be monitored, the sampling frequency, the dynamic range of the signal, and the collection duration.

With the data availability on the Platform, the user can view the requested spectral analyses of each monitored asset. The collections performed become available along with the continuous monitoring of speed, acceleration, and temperature on each asset’s page, as seen in Figure 1.

In this way, when observing any increasing trend in the continuous monitoring, the user can quickly access the spectral analyses corresponding to this period to diagnose the asset’s health, identify a possible failure, and plan maintenance.

This analysis can be done for any type of equipment, such as motors, gearboxes, bearings, pumps, among others.