Page 24 - EngineerIT November 2021

AUTOMATION AND CONTROL

The interfacing requirements of these two types of sensors over the band of interest and designing the filter to have flat pass-
are also very different. Some MEMS accelerometers have digital band response.
output and can be hooked directly to a microprocessor. Most • Out-of-band signal rejection. Signals outside the band of
of the high-performance MEMS accelerometers have analogue interest are of no use to the CM system and can cost precious
output and require a data acquisition signal chain. MEMS processing power or even contaminate the signals of interest.
sensors can typically be powered by a single-ended 3.3 V to 5 V It is best for the DAQ signal chain to remove all signals outside
supply shared with the DAQ signal chain. In comparison, piezo the band of interest.
accelerometers with the IEPE interface are typically powered by • Noise. Just like signal flatness, it is desirable for the measurement
a ~4 mA constant current source generated across a 24 V supply system to have a uniformly flat noise spectral density (NSD) over
through a two-wire cable, with the sensor output being an AC the band of interest. The noise floor should be lower than the
signal on top of a DC bias voltage (typically 8 V to 10 V), which minimum signal amplitude of interest. The FFT process has the
typically needs to be buffered, attenuated and level shifted before added benefit of decreasing the overall noise floor in the frequency
it can be acquired by an ADC. domain output due to processing gain. A simple explanation is
that the more samples being processed, the narrower the bin
Channel count size and the lower noise, the power is inside each bin. This allows
Another sensor related consideration is the number of sensors to the measurement system to artificially increase its measurement
be used, which can directly impact the number of DAQ channels dynamic range (only in the frequency domain) to examine signals
required. A CM system may deploy the same sensor type in that will otherwise be under the noise floor. The limitation of
multiple locations to provide a more complete picture of the the processing gain is that it requires large memory and longer
asset condition. For example, a pair of vibration sensors can be processing time. The spurious-free dynamic range (SFDR) of the
placed orthogonally to provide more accurate information on the measurement signal chain can also set the smallest meaningful
magnitude of the asset vibration. A three-axis vibration sensor can signal amplitude to be measured.
be mounted with any angular position and still have full sensitivity • Dynamic linearity. Low harmonic distortion is important for
of the vibration in all directions. Certain fault diagnosis methods frequency domain harmonic analysis. Additional harmonics
also rely on the phase difference between multiple signals to caused by the non-linearity of the measurement signal chain can
triangulate the location of the fault. This requires that the CM mask the deviation of the real harmonic signals caused by the
system simultaneously acquires signals from multiple sensors of fault condition.
the same type, which translates into simultaneous sampling, phase
matching and channel sampling synchronisation requirements for Time domain analysis
the DAQ signal chain. Frequency domain analysis is limited to the monitoring of periodic
signals, such as those intrinsically produced by rotating machinery.
Analysis method For monitoring assets that operate in non-periodic fashion – for
The choice of analysis method also plays a key role in the DAQ signal example, with linear and reciprocating motion and for assets that
chain design decision making. operate based off specific timing, such as hydraulic/ pneumatic
cylinders – time domain analysis is needed. Even for monitoring
Frequency domain analysis rotation machinery, certain analysis methods, such as the shock pulse
Frequency domain analysis is a popular CM method for monitoring method, also rely on data analysis in the time domain.
moving machinery. Harmonics at multiples of the fundamental
frequency of a rotating machine can be detected through sensing The time domain information can be obtained by simply analysing the
modalities such as vibration, sound and power quality. Determining sampled data waveform. The key DAQ signal chain design parameters
the amplitude and frequencies of these harmonics is the first basic to consider for time analysis include:
step in analysing the operating condition of the machine. • Bandwidth of interest. The bandwidth of the measurement
signal chain shall be wide enough to not distort the signal
The frequency domain information can be obtained by converting waveform at the highest frequency of interest. It is usually
time domain samples using the fast fourier transform (FFT). The not the frequency of the transient event occurrence, but the
key DAQ signal chain design parameters to consider for frequency oscillation frequency of the signal resulting from the transient
analysis include: event that sets the measurement bandwidth requirement. In
• Bandwidth of interest. The measurement band of interest depends some cases, such as monitoring using the shock pulse method,
on the property of the asset being monitored and the type of the transient event induced signal oscillation is set by the
fault coverage. The vibration monitoring bandwidth required resonant frequency of the sensor.
for monitoring gearbox bearings can be significantly higher than • Sampling rate. As opposed to frequency analysis – where
for monitoring the structure swing of a wind tower. The overall the effective signal sampling rate in principle does not need
monitoring signal chain should have enough bandwidth to cover to be higher than twice the highest frequency component
the highest frequency component of interest. to be monitored – the sampling rate requirement for time
• Magnitude flatness. Flat magnitude response over the frequency of domain analysis may need to be much higher than the highest
interest is typically desired for frequency analysis – that is, the gain input signal frequency of interest. This is due to the transient
shall remain constant over frequency. The magnitude response nature of the signals being monitored. Oversampling of the
variation over frequency can come from both the sensor response transient signal makes it easy to analyse the profile of the signal
and the response of the filtering inside the DAQ signal chain. Good waveform, including its peak and valley magnitude and rate of
flatness can be achieved by choosing a sensor with flat response change. The maximum error to peak value ratio can be derived

EngineerIT | November 2021 | 22