Aconundrum: what do you do with all that intelligence which resides in state-of-the-art sensors? Ignore it or use it?
The good news is: smart sensors could be smarter than you think – and able to decipher a lot of those riddles arising from the shortcomings which are a natural yet anomalous by-product of problems associated with measurement, process or plant operations.
Mostly, of course, information that could be extracted from smart sensors is just ignored, simply left to evaporate since, to recover it and analyse it has until now been technically forbidding and prohibitively expensive.
And we’re talking here specifically pressure transmitters; smart sensors capable of telling us much more than previously we’ve been able, or willing, to hear, yet which beckon unrelentingly to deaf ears.
Last year we reported on the secret codes buried in ‘noise’, that commodity widely regarded as a nuisance and of no use whatsoever. We reported on the work of Professor Ted Higham at Sussex University and of Dr Joe Amadi-Echendu at Greenwich university, who told us valuable messages that could tell us much about the condition of process plant and measurement system status were simply being ignored and/or lost.
Amadi-Echendu’s work has moved on, as has operational validation elsewhere, including Oxford University, where Dr Manus Henry has conducted extensive research on this very subject.
So how does it work? A spectral analysis of a high resolution (up to 16 bit) provides the raw material, while the focus of the exercise is to generate what Amadi-Echendu refers to as ‘change detection metrics’. These in turn can tell us a lot about what is happening with the measuring device, a combination of the interfacing process and the sensing component.
These metrics enable us to determine whether any abnormality in the signal is due to the device or the process itself – a kind of ‘validation hierarchy’.
Truth is that with increased automation, validation is vital for effective control, maintenance and management of industrial plants and processes.
Amadi-Echendu points out that validation should be traceable, starting from the interface to the process up to the highest level of operations. At present, validation is focused at the systems level and current methods are based on condition monitoring, and fault detection and isolation techniques using functional, analytical and hardware redundancy.
But validation of the measurement output from a pressure sensor is, says Amadi-Echendu, ‘imperative in situations where the cost of redundancy may be prohibitive. It should provide both measurement and condition information.’
And the contention is that the condition information should be used to assess the validity of the measurement by identifying metrics which describe the sensor and process conditions, respectively.
Work in this area continues at the University of Greenwich. Dr Yong Yan, of the School of Engineering, has concentrated research on the wideband signal taken directly from sensors that can be processed to provide two metrics for validating measurement devices – but which can also provide valuable information in the context of plant condition monitoring.
The idea is to identify three main levels of validation: measurement validation; process validation; and then plant validation.
The approach, says Yan, strongly depends on successful extraction of the wideband signal – that is the electrical signal that is obtained from the sensor with minimal electronic filtering.
Then the approach uses an artificial neural network as a computational tool to identify the device and process states using partial information arising from time series analysis and wavelet transforms.
It sounds good; and it is good. But, as Amadi-Echendu is the first to admit, the technology comes with a significant price tag attached: if you want to use it to do real condition monitoring, then you’ll need someone with the right credentials and experience to interpret the results.
It could, in short, lead to the creation of a whole new real time analytical discipline.