Optimization means to maximize (or minimize) a Performance Index that reflects
an economic (or equivalent), balanced composite of multiple objectives while
providing safety from violating Constraints. The customer places
constraints on any variable to represents absolute requirements, capacities,
Ø ULTRAMAX delivers the highest
cumulative process performance of any operations optimization technology
today for processes that run sequentially, such as production
equipment. This means highest cumulative profit impact, if those were
the objectives. This means fast optimization and being close to the
optimum, even as optimum conditions change due to changes in uncontrolled
inputs (for licenses dedicated to a process). This applies under most
practical conditions – unless the process was adjusted optimally at the
Ø ULTRAMAX gets quickly as close to
the optimum as about the noise of the Performance Index and
Constraints. Noise is the variations in outputs for constant inputs,
due to unexplained causes.
Ø Implementation: It takes about one
week to train and plan the optimization strategy. If the license is
dedicated to process and is integrated, it takes about 2 weeks to integrate
with DCS. It takes 2 weeks of regular (old) operations to collect data
for the Process Capability Analysis. Then Production with Sequential
Optimization is started. Noticeable improvements are frequently
obtained in about 2 weeks (50 to 200 re-adjustments, or more for very large
applications, unless there are problems with the process). The time to
converge to near-optimal operations depends on how far from optimum the
optimization was started. If production performance is still not
satisfactory, the method induces plant personnel to discover bottlenecks to
be resolved, sometimes noise reduction which calls for the use of engineering
solutions such as Six-Sigma.
ULTRAMAX uses Sequential Empirical Optimization (SEO)
technology. Empirical means that is based on data, not on
first-principle engineering models – thus it always represents reality
limited in accuracy by the noise. SEO assimilates operating data as
it is generated by the production process. The information in the
data is used immediately to refine knowledge of process behavior
characteristics, and from it to determine how to re-adjust the process
better. Unlike classic approaches such as Neural Networks and Design of
Experiments, the information is not lying unused until the analysis at the
end of data collection. Therein lies the great efficiency and
effectiveness of sequential analysis, especially when imbedded in software
with hundreds of applications and many years of experience and development
making it superbly reliable.
SEO includes Feed-forward optimization, to provide optimal adjustments
that change with the values of (known) uncontrolled inputs such as raw
material characteristics, process states (e.g., temperature of cooling
water), and environmental conditions.
The Dynamic Optimization features of ULTRAMAX, in addition to
Sequential Optimization and Feed-forward optimization, enables frequent
changes to the Optimization Plan, including changes in constraints,
calculations, etc. to reflect changes in business conditions and changing
levels or awareness as to what is truly important.
If the process itself is changed so much that the older data no longer
represents how the process behaves today, then old data is simply ignored and
a new round of SEO is started.
What-if Analysis gives predicted values
and consequent Alerts for exploratory input
The current version optimizes steady-states. This means that for
continuous processes, after making adjustments to the control inputs, one has
to bypass the transient time and wait for near-steady-state before collecting
output operating data. In steady-state, the outputs are mostly a consequence
of the current inputs, and not of older ones. (There is a Transient
option as well.)
There may be variables which are calculated as a function of other
variables. The calculation can be anything that can be programmed.
Graphics: plain data and 3D rotating model depiction: run data,
prediction errors, predicted values, empirical model quality, how close
to (local) optimum, etc.
Analysis capability: prediction model equations and dimensionless
interpretation, detected effects of the variables on outputs, actual results
versus predictions, historical data reports, etc.
10. Alerts: 15 different process and
optimization performance status alerts (notes, warnings and alarms),
including run-by-run statistical quality control.
11. Integration capability: Integrated
for automatic data exchange (both ways) with DCS, SCADA in Advisory mode and Closed-Loop
Stand-alone mode from an optimization workstation, entering data and
making adjustments by hand, useful for engineering and development studies.
12. Ability to analyze the sensitivity
of constraints to evaluate the benefits of moving or removing them, and
evaluate associated re-engineering projects.