Hydrocarbon Processing - November 2021 - 75

Digital Transformation
assay of the recently arrived crude oil to correlate to the new
heat exchanger fouling development.
A field operator will walk through the plant listening for possible
problems, such as a compressor with a knocking noise,
looking at steam traps releasing live steam rather than flash
steam, or checking lube oil leakage around motors and pumps.
The operator wears personal protective equipment (PPE) along
with a portable gas detector and performs an inspection of the
vibration in rotating machinery or a quick boiler flame inspection
using a pyrometer. Since most of these sensors are available
today with wireless connectivity, the need for walkarounds is
minimized. Similarly, a nurse will wear a mask to avoid contamination
while checking temperature, auscultating heart and
lungs with the appropriate devices.
The striking difference between the scientific progress in
testing the ApoE genotype and checking the health of a physical
asset is that some human patients do not wish to undergo the
genetic exam or know its result, while industrial " patients " want
to know about their condition as soon as possible. Plant operators
want to be proactive in avoiding the next unplanned downtime.
Manufacturers are eager to use all available data to predict
future failures and to prescribe action to avoid any production
loss, personal accident or environmental incident. There is no
" social barrier " in disclosing the health condition in an industrial
facility. It is exactly the opposite: insurance companies are
heavily investing in the IoT to reduce costs in the long term, and
share the savings with their customers.
However, huge potential exists for breaking through the
" social barrier " of harvesting the scientific evolution from
Isaac Newton and Antoine Lavoisier's times to the current era
of machine learning and AI. A polymerization reactor, a steel
mill blast furnace or a wood pulp digestor can be mathematically
modeled at the microscopic level using the Navier-Stokes
partial differential equations, which perform mass and energy
balances combined with kinetics mechanisms and integrated to
the boundary conditions of real-life operations. In the realm of
reliability engineering, there are more than 20 probability distributions
estimating failure rates along with time, including the
popular Weibull distribution. Statistics have been used heavily
in the analytics marketing field not only to predict customer
behavior (when the next purchase will take place), but also to
predict asset performance using machine-learning techniques
such as neural networks, multilayer perceptron (MLP), radial
basis functions, support vector machines (SVMs), Naïve Bayes,
k-nearest neighbors and geospatial predictive modeling.
With so much algorithmic and computational power available
today, some practitioners just want to throw more data to
machines to unravel correlations to predict future behavior.
Meanwhile, they are not taking advantage of the universal library
of scientific knowledge that has been built through human
evolution. They rely too much on data and intelligent machines
to almost completely replace the operational experience acquired
in starting new plants, revamping old units, undergoing a
multi-year turnaround, or facing the pressure of recovering from
an unplanned downtime in a sold-out economic scenario. Instances
have been seen of AI turning day-to-day operations into
a zero-sum journey to the detriment of natural intelligence (NI).
Often, the human element is being minimized throughout
the experience. A computer can be trained to play chess and it
FIG. 3. Passengers on US Airways Flight 1549 await rescue after
a successful emergency landing on the Hudson River.
will end up beating a human player. However, it does not know
how to play checkers, and it will be beaten by a human competitor.
It must be trained how to play checkers. What if another
computer, machine or robot teaches it how to play checkers?
This may lead to a machine training another machine how to
identify an issue with a fluid catalytic cracker in an oil refinery
and zooming into the root cause being its aging catalyst without
consulting a human subject matter expert (SME) and confirming
the diagnosis. This is a matter of leaving money on the table
by failing to integrate AI with NI.
Miracle on the Hudson. On January 15, 2009, US Airways
Flight 1549 (Airbus A320-214) struck a flock of Canadian
geese and lost all engine power two minutes after takeoff from
New York City's LaGuardia Airport. Unable to reach any airport,
pilots Chesley " Sully " Sullenberger and Jeffrey Skiles
glided the plane to a water landing in the Hudson River off
Midtown Manhattan (FIG. 3). The incident came to be known
as the " Miracle on the Hudson. "
Upon colliding with the geese, both of the aircraft's GE Aviation/Snecma-designed
CFM56 turbofans were damaged and
the associated electrical controls malfunctioned. Immediately
after the bird strike, according to the National Transportation
Safety Board (NTSB),3
Captain Sullenberger accomplished a
critical item that the flight crew did not reach in the checklist:
starting the Honeywell Auxiliary Power Unit (APU), a small
gas turbine engine that provides electrical power for starting an
aircraft's main engines and running air conditioning and other
systems while the plane is on the ground. According to the
NTSB report, " Starting the APU early in the accident sequence
proved to be critical because it improved the outcome of the
ditching by ensuring that electrical power was available to the
airplane " during the 58-sec elapsed time.
Hydrocarbon Processing | NOVEMBER 2021 75
FIG. 2. ApoE derives from the protein Apolipoprotein E.2
Hydrocarbon Processing - November 2021

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