How condition monitoring led the way to Industry 4.0
发布时间：2024年6月4日 10:46
200
(Credit: Shutterstock)
An often-cited advantage of Industry 4.0 is the use of embedded sensors to provide real-time data.
An often-cited advantage of Industry 4.0 is the use of embedded sensors to provide real-time data.
This will reduce the downtime required for maintenance of machinery, prevent catastrophic failure and allow systems to be optimised.
These ideas are not, however, all that new – vibration analysis has been carried out using permanently installed sensors for decades. To get an insight into the tangible benefits that Industry 4.0 might provide, a good place to start is the condition monitoring of high-value machinery, using vibration analysis.
By analysing the frequency, amplitude, phase, position and direction of vibrations in machinery, it is possible to identify many common faults. An analyst can, for example, differentiate between wear on a specific gear or bearing, a lack of lubrication, an imbalance, a misalignment, a loose mounting or an electrical fault. Fault finding and acquisition of replacement parts can be carried out before a machine is stopped, reducing downtime to its bare minimum. Early detection and preventative maintenance can also prevent more serious faults from developing.
Vibration is most commonly measured using an accelerometer, although displacement or velocity measurements are sometimes used. Magnetic accelerometers are often used for manual inspection, while embedded sensors are built into high-value machinery such as aero-engines, steam turbines and wind turbines.
Identifying faults
Regardless of whether acceleration, velocity or displacement is measured, the other parameters can, of course, be obtained by integration or differentiation of the measurement signal. Some faults, such as a shaft alignment or imbalance in rotating machinery, will produce a sinusoidal signal, with a frequency that is some multiple of the shaft running speed. Electrical motor faults are often a multiple of the AC supply frequency.
Other faults, such as damaged gear teeth or bearings, will produce brief, periodic impacts. The frequency of these impacts will depend on the number of teeth on a particular gear, or the rate at which balls precess around a bearing. Where multiple bearings of the same size are located along a shaft, the faulty bearing can be identified by the location on the machine where the vibration has the greatest amplitude.
Surface-mounted accelerometers normally measure the acceleration normal to the surface. It may, therefore, also be useful to place sensors in different positions to identify the direction in which the vibration has the greatest amplitude. An example would be to determine whether a shaft misalignment is producing axial or radial vibrations.
In real machinery there are normally many sources of vibration. This means that analysis is not as simple as looking at the measured signal and identifying the waveform caused by a particular fault.
Spectral analysis is, therefore, often used to identify the underlying sources of vibration within a complex signal. This uses fast Fourier transforms to transform the time domain signal into the frequency domain, allowing the frequency and amplitude of the vibration sources to be identified.
In rotating machinery, the frequency corresponding to the shaft running speed is normally identified as X. The frequencies are then converted into a multiple of X.
The type of fault finding carried out by a skilled vibration analyst requires knowledge of the machine’s mechanics. The required information may include running speeds, gear and pulley sizes, details of bearings and electrical supply frequency. Measurements may be taken over each bearing location.
Machine learning
As we transition to Industry 4.0, there is an expectation that machine-learning algorithms will be able to identify faults without requiring human analysis of the data. It is often suggested that these algorithms will not require knowledge of the mechanics in the way that a human would. This assumes they will be able to empirically perform diagnostics, based only on historical data correlating measurements with fault conditions. However, such an approach requires very large data sets before accurate results can be obtained.
Machine learning may not replace humans until it is able to make model-based inferences, just as humans do.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.
These ideas are not, however, all that new – vibration analysis has been carried out using permanently installed sensors for decades. To get an insight into the tangible benefits that Industry 4.0 might provide, a good place to start is the condition monitoring of high-value machinery, using vibration analysis.
By analysing the frequency, amplitude, phase, position and direction of vibrations in machinery, it is possible to identify many common faults. An analyst can, for example, differentiate between wear on a specific gear or bearing, a lack of lubrication, an imbalance, a misalignment, a loose mounting or an electrical fault. Fault finding and acquisition of replacement parts can be carried out before a machine is stopped, reducing downtime to its bare minimum. Early detection and preventative maintenance can also prevent more serious faults from developing.
Vibration is most commonly measured using an accelerometer, although displacement or velocity measurements are sometimes used. Magnetic accelerometers are often used for manual inspection, while embedded sensors are built into high-value machinery such as aero-engines, steam turbines and wind turbines.
Identifying faults
Regardless of whether acceleration, velocity or displacement is measured, the other parameters can, of course, be obtained by integration or differentiation of the measurement signal. Some faults, such as a shaft alignment or imbalance in rotating machinery, will produce a sinusoidal signal, with a frequency that is some multiple of the shaft running speed. Electrical motor faults are often a multiple of the AC supply frequency.
Other faults, such as damaged gear teeth or bearings, will produce brief, periodic impacts. The frequency of these impacts will depend on the number of teeth on a particular gear, or the rate at which balls precess around a bearing. Where multiple bearings of the same size are located along a shaft, the faulty bearing can be identified by the location on the machine where the vibration has the greatest amplitude.
Surface-mounted accelerometers normally measure the acceleration normal to the surface. It may, therefore, also be useful to place sensors in different positions to identify the direction in which the vibration has the greatest amplitude. An example would be to determine whether a shaft misalignment is producing axial or radial vibrations.
In real machinery there are normally many sources of vibration. This means that analysis is not as simple as looking at the measured signal and identifying the waveform caused by a particular fault.
Spectral analysis is, therefore, often used to identify the underlying sources of vibration within a complex signal. This uses fast Fourier transforms to transform the time domain signal into the frequency domain, allowing the frequency and amplitude of the vibration sources to be identified.
In rotating machinery, the frequency corresponding to the shaft running speed is normally identified as X. The frequencies are then converted into a multiple of X.
The type of fault finding carried out by a skilled vibration analyst requires knowledge of the machine’s mechanics. The required information may include running speeds, gear and pulley sizes, details of bearings and electrical supply frequency. Measurements may be taken over each bearing location.
Machine learning
As we transition to Industry 4.0, there is an expectation that machine-learning algorithms will be able to identify faults without requiring human analysis of the data. It is often suggested that these algorithms will not require knowledge of the mechanics in the way that a human would. This assumes they will be able to empirically perform diagnostics, based only on historical data correlating measurements with fault conditions. However, such an approach requires very large data sets before accurate results can be obtained.
Machine learning may not replace humans until it is able to make model-based inferences, just as humans do.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.
Share:
Read more related articles
Manufacturing
Read more related articles
Manufacturing
