How does sensor performance support condition monitoring solutions?

To realize these new capabilities and reap the benefits of condition monitoring, new solutions must be accurate, reliable, and robust so that real-time monitoring can extend beyond basic detection of potential equipment failures to provide insightful and actionable information. The performance of next-generation technologies combined with system-level insights are enabling a deeper understanding of the applications and requirements needed to address these challenges.

Vibration is one of the key elements of machine diagnostics and has been reliably used to monitor the most critical equipment in a variety of industrial applications. There is a large amount of literature to support the various diagnostic and predictive capabilities required to implement advanced vibration monitoring solutions. However, there is not much literature on the relationship between vibration sensor performance parameters such as bandwidth and noise density and the fault diagnostic capabilities of the end application. This article describes the main types of machine faults in industrial automation applications and identifies the key performance parameters of vibration sensors that are related to specific faults.

Several common fault types and their characteristics are highlighted below to provide insight into some of the key system requirements that must be considered when developing a condition monitoring solution. These fault types include, but are not limited to, imbalance, misalignment, gear faults, and rolling element bearing defects.

Imbalance is the uneven distribution of mass that causes loads to shift the center of mass away from the center of rotation. System imbalance can be attributed to improper installation (such as coupling eccentricity), system design errors, component failure, or even the accumulation of debris or other contaminants. For example, the cooling fan built into most induction motors can become unbalanced due to uneven accumulation of dust and grease or damaged fan blades.

An unbalanced system will generate excessive vibrations that can mechanically couple to other components within the system, such as bearings, couplings, and loads, potentially causing accelerated degradation of components that are in good working condition.

Increased overall system vibration may indicate a potential fault caused by an unbalanced system, but the root cause of the increased vibration needs to be diagnosed through frequency domain analysis. An unbalanced system produces a signal at the system's rotation rate (usually referred to as 1×) with an amplitude proportional to the square of the rotation rate, F = m×w2. The 1× component is usually always present in the frequency domain, so an unbalanced system can be identified by measuring the amplitude of 1x and harmonics. If the amplitude of 1× is higher than the baseline measurement and the harmonics are much smaller than 1×, an unbalanced system is likely present. Horizontal and vertical phase-shifted vibration components may also appear in an unbalanced system.

The noise must be low to reduce the impact of the sensor and enable detection of small signals generated by unbalanced systems. This is very important for the sensor, signal conditioning, and acquisition platform.

In order to detect small imbalances, the acquisition system needs to have a high enough resolution to extract the signal (especially the baseline signal).

Sufficient bandwidth is also required to capture sufficient information (beyond just the rotation rate) to improve the accuracy and reliability of the diagnostics. The 1× harmonic can be affected by other system faults, such as misalignment or mechanical looseness, so analyzing harmonics of the rotation rate (or 1× frequency) can help distinguish between system noise and other potential faults. For slow rotating machines, the fundamental rotation rate can be well below 10 rpm, which means the low frequency response of the sensor is critical to capturing the fundamental rotation rate. Analog Devices’ MEMS sensor technology can detect signals as low as DC and is able to measure slower rotating equipment while also measuring wide bandwidths to obtain higher frequency content typically associated with bearing and gearbox defects.

Figure 1. An increase in the magnitude of the rotation rate or 1X frequency may indicate an unbalanced system.

As the name implies, system misalignment occurs when two rotating shafts are misaligned. Figure 2 shows an ideal system where the alignment starts with the motor, then the shafts, couplings, and on to the load (in this case, the pump).

Figure 2. Ideal alignment system

Misalignment can occur in parallel, angular, or a combination of the two (see Figure 3). When two axes are misaligned in the horizontal or vertical direction, it is called parallel misalignment. When one axis is at an angle to the other, it is called angular misalignment.

Figure 3. Examples of different misalignments, including (a) angular, (b) parallel, or a combination of both.

Misalignment errors can affect the larger system by forcing a component to operate under stress or loads higher than it was originally designed to handle, ultimately potentially leading to premature failure.

Misalignment errors typically appear as the second harmonic of the system rotation rate, referred to as 2×. The 2x component is not necessarily present in the frequency response, but when it is, its amplitude relationship to 1x can be used to determine if a misalignment is present. Added alignment errors can excite harmonics up to 10×, depending on the type of misalignment, measurement location, and orientation information. Figure 4 highlights the characteristics associated with a potential misalignment fault.

Figure 4. A growing 2× harmonic plus growing higher harmonics indicate a possible misalignment.

To detect small misalignments, low noise and sufficiently high resolution are required. The machine type, system and process requirements, and rotation rate determine the permissible misalignment tolerance.

Sufficient bandwidth is also required to capture sufficient frequency range to improve the accuracy and reliability of the diagnosis. The 1× harmonic can be affected by other system faults, such as misalignment, so analyzing the harmonics of the 1× frequency can help distinguish other system faults. This is especially true for higher speed machines. For example, machines (such as machine tools) that rotate at speeds above 10,000 rpm typically require high-quality information above 2 kHz in order to accurately and reliably detect unbalance.

The system phase is combined with the directional vibration information to further improve the diagnosis of misalignment errors. Measuring the vibration at different points on the machine and determining the difference between the phase measurements or across the system can provide insight into whether the misalignment is angular, parallel, or a combination of both types of misalignment.

Rolling element bearing defects are often artifacts of mechanically induced stresses or lubrication issues that create small cracks or defects within the mechanical components of the bearing, resulting in increased vibration. Figure 5 provides some examples of rolling element bearings and shows several defects that can occur.

Figure 5. (Top) Example of rolling element bearing and (bottom) lubrication and discharge current defects

Rolling element bearings are used on nearly all types of rotating machinery, from large turbines to slow-rotating motors, from relatively simple pumps and fans to high-speed CNC spindles. Bearing defects can be a sign of lubrication contamination (fig. 5), improper mounting, high-frequency discharge currents (fig. 5), or increased system loads. Failures can result in catastrophic system damage and have significant impacts on other system components.

There are a variety of techniques available to diagnose bearing faults, and due to the physics behind bearing design, the defect frequency for each bearing can be calculated based on the bearing geometry, rotational speed, and defect type, which can help diagnose the fault. The bearing defect frequencies are shown in Figure 6.

Analysis of vibration data for a specific machine or system often relies on a combination of time and frequency domain analysis. Time domain analysis can be used to detect trends in overall increases in system vibration levels. However, this analysis contains very little diagnostic information. Frequency domain analysis can increase diagnostic insight, but determining the fault frequency can be complicated by the influence of other system vibrations.

For early diagnosis of bearing defects, using harmonics of the defect frequency can identify early or emerging faults, allowing them to be monitored and maintained before catastrophic failure occurs. To detect, diagnose, and understand the system impact of bearing faults, techniques such as envelope detection (as shown in Figure 7) combined with spectral analysis in the frequency domain can often provide more insightful information.

Low noise and sufficiently high resolution are critical for early bearing defect detection. When a defect first appears, the amplitude of the defect signature is usually low. Due to design tolerances, the inherent mechanical slip in the bearing will spread the amplitude information to multiple bins in the bearing frequency response, further reducing the vibration amplitude, thus requiring low noise to detect the signal earlier.

Bandwidth is critical for early detection of bearing defects. During rotation, each time a defect is struck, a pulse containing high frequency content is generated (see Figure 7). Monitoring harmonics of the bearing defect frequency (rather than rotation rate) can reveal these early faults. Due to the relationship between bearing defect frequency and rotation rate, these early features can occur in the thousands of hertz range and extend beyond the 10 kHz to 20 kHz range. Even for low speed equipment, the inherent nature of bearing defects requires wide bandwidth for early detection of defects to avoid the effects of system resonances and system noise that affect the lower frequency bands.

Dynamic range is also important for bearing defect monitoring because system loading and defects can affect the vibration experienced by the system. Increased loads result in increased forces acting on the bearing and defects. Bearing defects can also create shocks that excite structural resonances, amplifying the vibrations experienced by the system and sensor. As machines ramp up and down in stop/start situations or during normal operation, the varying speeds create potential opportunities for system resonance excitation, resulting in higher amplitude vibrations4. Saturation of the sensor can result in loss of information, misdiagnosis, and in the case of some technologies, even damage to the sensor element.

Figure 6. Bearing defect frequency depends on bearing type, geometry and rotation rate.

Figure 7. Techniques such as envelope detection can extract early bearing defect signatures from wide bandwidth vibration data.

Gear failures usually occur in the gear mechanism teeth pitch due to fatigue, spalling or pitting. They manifest as cracks in the tooth root or metal removal on the tooth surface. Causes include wear, overload, poor lubrication and backlash, and occasionally improper installation or manufacturing defects5.

Gears are the primary elements of power transmission in many industrial applications and are subject to considerable stresses and loads. The health of the gears is critical to the proper functioning of the entire mechanical system. A well-known example from the renewable energy sector is that the single largest contributor to wind turbine downtime (and corresponding revenue loss) is the failure of the multi-stage gearbox in the primary powertrain5. Similar considerations apply to industrial applications.

Detection of gear faults is challenging due to the difficulty of mounting vibration sensors close to the fault and the presence of considerable background noise caused by the multiple mechanical excitations within the system. This is especially true in more complex gearbox systems where there may be multiple rotational frequencies, gear ratios, and meshing frequencies6. Therefore, detecting gear faults may require the use of a variety of complementary methods, including acoustic emission analysis, current signature analysis, and oil residue analysis.

In terms of vibration analysis, accelerometers are usually mounted on the gearbox housing and the dominant vibration mode is axial vibration7. The frequency of the vibration characteristic produced by a healthy gear is the so-called gear mesh frequency, which is equal to the product of the shaft frequency and the number of gear teeth. There are also usually some modulated sidebands related to manufacturing and assembly tolerances. These conditions for a healthy gear are shown in Figure 8. When a localized fault such as a tooth crack occurs, the vibration signal in each revolution will include the mechanical response of the system to short-duration impulses of relatively low energy levels. This is usually a low amplitude broadband signal and is generally considered to be non-periodic and non-static7,8.

Figure 8. Spectrum of a healthy gear, with crankshaft speed of ~1000 rpm, gear speed of ~290 rpm, and 24 gear teeth.

Due to these characteristics, standard frequency domain techniques alone cannot accurately identify gear faults. Spectral analysis may not be able to detect early gear faults because the impact energy is contained in the sideband modulation, which may also contain energy from other gear pairs and mechanical components. Time domain techniques (such as time synchronous averaging) or mixed domain methods (such as wavelet analysis and envelope demodulation) are generally more appropriate9.

Generally speaking, wide bandwidth is very important for gear fault detection because the number of gear teeth is a multiplier in the frequency domain. Even for relatively low-speed systems, the required detection frequency range quickly rises to the multi-kHz region. In addition, localized faults further extend the bandwidth requirements.

Resolution and low noise are critical for a number of reasons. It is difficult to mount a vibration sensor close to a specific fault area, which means that the mechanical system may attenuate the vibration signal to a high degree, so being able to detect low energy signals is critical. In addition, because the signal is not a static periodic signal, standard FFT techniques to extract low amplitude signals from a high noise floor cannot be relied upon, and the noise floor of the sensor itself must be low. This is especially true in a gearbox environment where multiple vibration signatures from different components are mixed. In addition to these considerations, early detection is important not only for asset protection reasons, but also for signal conditioning reasons. It has been shown that the vibration severity of a single tooth fracture fault situation may be higher than that of a fault situation with two or more teeth fractured, which means that detection at an early stage may be relatively easier.

While common, imbalance, misalignment, rolling element bearing defects, and gear tooth faults are just a few of the many fault types that high-performance vibration sensors can detect and diagnose. Higher sensor performance, combined with proper system-level considerations, is enabling a new generation of condition monitoring solutions that provide greater insight into the mechanical operation of a wide range of industrial equipment and applications. These solutions will change the way maintenance is performed and the way machines operate, ultimately reducing downtime, improving efficiency, and enabling new capabilities for the next generation of equipment.