Expecting routine operations at the start of their weekend shift, the crew was soon confronted with a concerning revelation: differential pressure (DP) across a fixed-bed reactor was increasing. It was a particular concern for a reactor whose catalyst substrate was relatively “soft” by design, and contained enough precious metals to purchase the yacht of an oil oligarch. So the crew started making adjustments to reduce the DP, several of which reduced production at a time when the business was desperate for product. Hence the ops manager was dismayed, particularly when it was revealed—they should not have believed their eyes.
In the 20th century, professionals deploying controls and measurement appliances, “instruments” if you will, might recall a day when the end user—operations—was quick to dismiss an instrument whose output didn't agree with their notion of what was really happening in the process. Specifications for accuracy, repeatability, temperature and pressure effects, and “conversion”—i.e., producing a 4-20 mA or 3-15 psi signal from some electronic or mechanical transducer—were literally orders of magnitude looser than what's typical today. “Unrevealed” failures caused by oil, water, desiccant or fragments of Teflon tape in the air supply were routine. The traces on a circular strip chart or analog meter might not even reveal the subtle changes troubling the crew that weekend. They would least of all be capable of using two pressure transmitters to compute a differential pressure that was one-fortieth (2.5%) of their individual full-scale pressure, with an accuracy of 2% of reading.
Digitally integrated instruments connected by HART, fieldbus or Ethernet sacrifice almost nothing to conversion, as the measurement seen at the host is precisely what the device computes and to more decimal places than is typically useful. They innately compensate for the effects of temperature, static pressure and similar effects that would otherwise skew accuracy and repeatability. One can purchase a transmitter with a five- or 15-year stability guarantee. The capacity for this extraordinary quality and reliability of measurements given to the end user has never been better.
This unprecedented reliability and stability have contributed to attrition in the ranks of instrument maintenance. Instruments simply don’t fail like they used to. Transmitters capable of extraordinary turndown can be adapted to a new, full-scale span with a few keystrokes on a handheld communicator—often to the dismay of those maintaining the database. Consequently, we're finding that our new generation of end users—operators—are less likely to question what they see on their screen. Should we train them to doubt measurements like their forebears? When the one-in-a-thousand (or more) malfunction threatens to lead them to ill-advised remedies, what guidance would you give?
'Digital twins' offer needed assurance
Could our newest imaginary friend, the digital twin, be of any value? A twin can be just a simple model correlating, for example, measured flow to valve position. Does the flow rate match the delta-level in the downstream tank? A more complex possibility would be mass and energy balances around a distillation column: steam to the reboiler should correlate consistently (if not precisely) with feed rate, condenser duty, reflux and product draw. Advanced controls engineers have been using similar methods for decades to validate lab measurements or on-line analyzers. Those fortunate enough to have the funding and resources to design and maintain a simulator based on first principles (physics) are able to do these comparisons on an entire unit, testing potential moves and investigating possible equipment and measurement shortcomings.
When the inferred measurement or model differs from the “real” one, do our end users become the person with two watches? Uncertainty fosters robotic reactions, which diminishes the value of having a human operator. Next month, we’ll consider additional strategies for furthering thoughtful analysis of reliable measurements.
