Hydrocarbon Processing - June 2021 - 68

Digitalization
hit-or-miss. This can be tolerable if the results are generated
quickly, but there is always a time lag.
Improvements can be provided by creating a software sensor.
With sophisticated analysis of the process from historical
data (Phase 2), it is possible to build a picture of what the process
looks like when it is operating. This provides ranges for
the variables that are possible to measure through conventional
instrumentation. By applying the soft sensor, all those critical
variables can be monitored as a continuous value like temperature,
flowrate and pressure, helping to determine if the process
is running as it should. This provides a better handle on what
the process is doing, but frequently leaves it running with too
wide a deviation from the target.
The next level of sophistication is applying multivariable
model predictive control (MMPC). With this approach (Phase
3), the controller has a model capable of expressing dynamic
characteristics of the process. Control is now based on prediction
by the model, which can make operation far more stable
than is possible with regulatory control built on multiple proportional,
integral and derivative (PID) loops. Variability common
to the previous control approach can be tightened up, keeping
the critical attribute within a much narrower range. The result
is operation closer to the desired goal, regardless of its nature.
It is now possible to determine exactly where to place the operating
level, whether it is closer to an upper limit (Phase 4) or
somewhere else dictated by big-picture corporate objectives.
Capabilities of a digital twin. The kind of analysis used to
build a software sensor can be extended to support a digital twin,
which provides a simulation designed to respond to changes in
the same way as the actual plant. If a setpoint is changed, the digital
twin responds in a similar manner to the actual process plant.
This allows operators to test setpoint changes and other process
adjustments to determine if they will have the desired effect.
This capability is especially valuable since parameters change,
which can make achieving optimal operation a moving target.
For example, the crude oil coming into a crude distillation unit
could easily be from a different source than the crude running
yesterday. Will the new crude allow the same draw rates of desired
products today as yesterday?
Rather than experimenting with the actual distillation tower,
operators can program the digital twin with the new crude attributes
and let the simulation do its analysis. Likely, the analysis will
call for setpoint and constraint adjustments to achieve the desired
product distribution. These adjustments can be implemented automatically,
or they can be provided as advice to operators.
Assembling the pieces. The parts just described can be
combined so they build on each other, so it soon becomes possible
to implement a largely self-functioning process control
mechanism able to move into a new level of total plant productivity
with real-time optimization (RTO). This type of mechanism
uses an advanced process control (APC) technique, such
as MMPC, to help analyze all the trade-offs that occur when
making individual unit optimizations. Simulation can be used
to find optimum operating points for multiple process units by
using current process data.
Greater detail is needed on how these elements interact
(FIG. 3). At the top are the operators, but this level goes beyond
the control room as it involves corporate management and
company-wide production scheduling as much as the individuals
in vests and hardhats.
People in the control room now understand how their efforts
fit into the bigger picture and exactly what is expected of
the specific unit and overall facility. Their tasks are simplified
because the control systems can automatically implement all the
constant tweaking and loop adjustments connected with traditional
regulatory control.
At the bottom of the diagram is the process itself, still controlled
by the distributed control system (DCS). This remains
as it always has been because it is still the mechanism that collects
data from instrumentation and sends instructions to actuators.
With traditional systems, the DCS would be controlled
by the operators directly, but now the APC system monitors
the DCS and supplements the efforts of operators. Setpoints
in the DCS can now be tweaked, as needed, automatically to
maintain target values.
The APC system is under the control of the RTO system,
which is programmed with economic targets. In a perfect
world where nothing changes, this would be enough, but nothing
is perfect. As mentioned earlier, feedstocks change, and
with something as complex as crude oil, the variability can be
subtle or drastic.
This is where the digital twin comes into play. It functions as a
FIG. 3. New platforms placed on top of the DCS use improved control
methodologies as compared to conventional regulatory practices.
68 JUNE 2021 | HydrocarbonProcessing.com
steady-state simulator representing the actual process. The simulator
has a library of information on specialized feedstock characteristics,
so it can determine exactly how the optimizer must
adjust the process to accommodate changes in feedstock characteristics.
This avoids any trial-and-error hunting for setpoints.
The process remains stable and within desired operational parameters,
even with these types of changes, as well as others.
http://www.HydrocarbonProcessing.com
Hydrocarbon Processing - June 2021

Table of Contents for the Digital Edition of Hydrocarbon Processing - June 2021
