How to Survive the Oncoming Train
“Our Inescapable Data vision suggests that it is not just advances in each of these technologies, it is the combination
of these fundamental elements that will break barriers and magnify gains to levels not yet anticipated. We think these new combinations will lead to an explosion of benefits driving both higher personal and economic satisfaction.” This is from a new book, Inescapable Data: Harnessing the Power of Convergence
out of IBM Press, by Chris Stakutis and John Webster. Stakutis is the IBMer, CTO for emerging storage software. Webster is founder and Senior Analyst for Data Mobility Group, a TLA consultant based in Boston. Their thesis is that the convergence of technologies and ubiquity of data will drive a revolution. Applying this to process automation, we get the vision of tomorrow’s plant we’ve given you. If you want to live long and prosper in this new world of inescapable data, you will need, as Rich Merritt encourages in his column, to “get your mind right about technology,” and start thinking in fourth order terms. Remember, nobody cares how the watch is built, or how time is kept.
What If? Virtual Plant Reality
By Gregory K. McMillan
, CONTROL columnistTHE CHEMICAL
industry has had the software since the turn of the century to create a virtual plant where the actual configuration of DCS resides in a personal computer with a dynamic simulation of the process that has evolved to be a graphically configured process and instrument diagram (P&ID). At the same time, the retirement of experienced engineers and technicians at the plant, the disengagement between the operator and process from the use of advanced control, and world wide competition has created a greater than ever need for training and optimization. Potentially, the virtual plant can go where the real plant hasn’t gone for good or bad reasons and replace myth with solid evidence and engineering principles.
Why are process simulators so far behind flight simulators?
What if a virtual plant offered the ability to explore new operating regions and abnormal conditions and help develop and prototype new control systems while meeting the basic but growing need of keeping operator and process engineer skills fresh? Simulations Behaving Badly
The results in the chemical industry have been spotty at best with some notable successes but with too many disappointments. Dynamic simulations of sufficient fidelity to design control systems and train personnel in process operation often require an investment in software and engineering of more than $100K, so when the behavior of the model is erroneous, you have unhappy customers. A misleading simulation is worse than no simulation.
Unfortunately, so called “high fidelity” process simulators show bizarrely unrealistic dynamic responses and may even develop fatal numeric errors for startup, shutdown, extremely abnormal situations, and instabilities. These process models behave analogous to a flight simulator that is great if the plane is cruising but would be erratic or crash during a takeoff, landing, wind-shear, or rudder problem.
Forget about trying to accurately simulate batch operations and exothermic reactors with simulators designed for continuous steady state operation. Empty vessels, zero flows, non-equilibrium conditions, and process non-self-regulation are not properly addressed.
Even at normal operating conditions, the total loop dead time is often off by 1000% or more. Since the integrated absolute error in most key control loops is proportional to the dead time squared, the simulation gives no inkling of the control problem. The mismatch between the process time constant of the plant and model fares somewhat better but errors of 50% or more are quite common. The process gain is the generally the most accurate, but even then it is not suitable for direct use for tuning or model predictive controller. Why are process simulators so far behind flight simulators?
Flight simulators can focus on the servo response of hydraulic controls for well-defined components in air or the ions in space, whereas chemical processes have significant dead times, pneumatic actuators, and thousands of poorly defined compounds. The physical properties (e.g., density, mass heat capacity, and boiling point) as a function of composition, pressure, and temperature of mixtures are missing. Often hypothetical compounds must be created and the simulation obligated to estimate the relationships. Just try to find hydrochloric acid and sodium hydroxide in a physical property package. If you combine this data problem with the uncertainties, disturbances, nonlinearities and slowness of the process, control valves and sensors, you are set up for a failure. (Click the Download Now button at the bottom of this article for a .pdf version of this chart.)
Chemical processes have significant dead times and thousands of poorly defined compounds.
IF YOU LOOK
at the block diagram in Figure 1 of a control loop (above), good physical property data and a “high fidelity” simulator can yield an accurate process output for a process input at steady state. When process simulators are developed by process engineers for process design, this is the beginning and end of quest. You have a design point and the information you need for a process flow diagram (PFD).