© 2004-6, Ultramax Corporation,
Cincinnati, Ohio, USA. All
These are the major developments to
help problem detection (e.g., process incapabilities, process upsets), to
simplify installation and maintenance, and to increase speed of
optimization. The ULTRAMAX Method is
now positioned as Advanced Process Management™ (APM™)
to compare it with Advanced Process Control (APC) (both multivariable). There is a patent pending.
- The new
“Capability/Consistency Analysis” is used both for the up-front
Baseline Capability Analysis; and for detection of specific problems and
periodic auditing of process capabilities to satisfy business
requirements (much more than just quality issues). [detection]
Capability Analysis is a valuable tool even for customers not interested in
Note that the
Baseline Capability Analysis itself does not contribute to the effectiveness
of Sequential Empirical Optimization (SEO); the test is done for the customer
to understand the issues of process capabilities independently of the issues
of applying SEO. Also, passing this
test is a requirement for UMC to issue performance guarantees.
case of failing new sequential data validation, help plant personnel by
indicating which variables are most likely to be the source of the
problem, so as to be able to determine the cause and correct the
situation for the future (basic Quality Control procedures). [detection]
restarting a round of Sequential Empirical Optimization (SEO). This has been an important thrust of
product development for version 7.0 .
longer use Prior Region. This feature makes starting and
maintaining an Optimization Plan much simpler. [simplification]
- Automatic Scenario
feature enables representing various components which may or may not be
operating at any one time in one ULTRAMAX application file,
rather that having to design and maintain one application file for each
scenario. A scenario is a
different combination of operating process components.
- Elimination of application size limitation. This is particularly valuable to be
able to recall process behavior and optima for earlier combinations of
uncontrolled inputs, and thus avoiding having to re-learn and “climb the
mountain” under those conditions again.
Also there is no code limit to the number of inputs or
- Adjustable ALARM threshold values. For users to select the sensitivity of
the Alerts (ALARM, Warning, Note) for each application (product /
process), which affects messages and turning Closed-Loop off.
Management Screen (OMS)
at the optimization workstation: Simplify integration (automatic data
exchange with the DCS) by automatically providing an operator interface
before or instead the end-user installing the OMS directly in the
control room consoles. [simplification]
- Calculations: Do not require the customer to have
Fortran, or having to use it. [simplification]
Optimize without waiting for the transients due to readjustments to fade
away, useful for uncontrolled inputs that change a lot (e.g.,
Modulation: Speed up optimization when it is known
that variable values should follow certain patterns. [performance]
- 7.0 is
available for WinXP, Win2000 and NT4.0
services for 7.0 now include much more thorough management involvement
and follow-through support for the application. This, together with the simplification
steps, was done to avoid the following frequent syndrome. Once getting great paybacks if a few
months, ULTRAMAX fell into disuse after a few years – frequently
associated with major changes in the process and with changes in plant
Description of new features:
1. Up-front Baseline Capability Analysis: The analysis is the new Capability Analysis with regular
production data at baseline adjustments before running ULTRAMAX. This analysis is much more thorough and
discriminating than anything available before.
2. Capability /
Consistency Analysis: This is a new analysis that can be done
with the Baseline data, with regular SEO data, or with any operating
constraints were being satisfied by the data, and will be satisfied in the
future by the estimated distributions.
If constraints are
satisfied, then operating improvements are to improve the Scorecard; else
improvements are to learn first how to regularly satisfy constraints.
Any loss in
output “effective operating ranges” to provide safety against constraint violations;
which may reduce the ability to get more desirable Scorecard values. The loss should be approved by the
extreme of the above, whether the effective operating range is positive for
all outputs with upper and lower constraints; else the process is totally
incapable. Further, the effective
operating range needs to be larger than 3 sigmas for SEO to be
How close to
the optimum (in the effective operating range) SEO can quickly take the
process. The loss should be approved
by the end-user.
Until the last
three issues are resolved, the process is not sufficiently capable. To resolve the issues it will be necessary
to modify the Optimization Plan and/or to reduce noise – that is, be able to
explain variations in outputs better.
report with baseline data is mostly for the benefit of customer’s clarity and
perception; and in particular to understand process capabilities before doing
SEO. Neither collecting baseline
production data for the analysis, nor the report itself, is necessary to
obtain gains with SEO (if the process is capable). However, with the Baseline Capability
Analysis it will be simpler to evaluate gains – which is desirable for most
new customers. The baseline capability
analysis report values could be reproduced with other standard statistical
3. Data Validation. When
entering new data and doing a What-if, the Advice Report displays a new
column named “Deviation”. This
Deviation indicates, variable by variable, where each value stands in terms
of the distribution past data, in order to detect which variables may be most
different from past ranges – and thus possibly indicate a problem. This logic applies equally to inputs and
Note that it is
possible to have all individual deviations small and still have a large
multivariate deviation index, which is a case of having an unusual combination
restarting a round of Sequential Empirical Optimization (SEO)
This feature raises
the adaptability and flexibility of ULTRAMAX a significant notch, leaving
further behind model-based solutions (that would require off-line rebuilding
of prediction models – whether first-principle or Neural Network based). This feature will also reduce the instances
of people not re-starting ULTRAMAX -- after significant changes with the
process ‑‑ because of the relatively higher complexities involved
in the old method.
An important aspect of the ULTRAMAX® solution to optimize
operations is the quick, dynamic adaptation to changes in:
the process (e.g., new type of
burners, a new duct, new mixer, etc.)
discovering new variables to be
included in the Optimization Plan (e.g., include a new input to reduce noise;
a new quality characteristic)
These kinds of changes result
in the stored run data no longer being complete or no longer being valid
(i.e., no longer representing how the process will perform). In such cases one starts a new “round” of
SEO, basing optimization on a new set of now valid data. Continuous improvements with ULTRAMAX calls
for this situation to happen with some regularity.
Historically, several customers
have ceased to continue regular use of ULTRAMAX; partially because of:
the perceived difficulty or
uncertainty in re-starting a new round of SEO -- especially when the customer
continues on their own without requesting help from the Agent or Ultramax
Corporation (UMC) personnel; coupled with
the belief that they can
re-optimize performance based on personnel experience.
The table below describes
the new simplifications in starting a new round of SEO; possible because of
the new features “No Prior Region” and “Automatic Scenario Management”
Old procedures (up to 6.0)
New procedures (7.0)
Ignore old invalid or
(if any) add/delete
variables; update global factors (economic factors, requirements) and
If multiple scenarios, do both
tasks above in each of the scenarios’ application files.
Same, except that even in the case
of multiple scenarios, it is done in only one application file.
Set Prior Regions around current
best known adjustments.
If multiple scenarios, do it in
each of the scenarios’ application files.
No longer done
And then continue operating -- as
one would even without using SEO -- with the current best known
adjustments, ready to obtain available benefits through refining
5. No longer use the Prior Region
The equivalent function
is now defined by the first run data entered into (a new round of) SEO and by
the MIDs in the Optimization Plan. The
first run is the currently best known process adjustments. The definition of the input MID is
now changed (see below), roughly three times the value with the previous
the Prior Region has two mayor benefits:
· Simplifying the formulation of the
It also makes the
application impervious to errors in the definition of the Prior Region, as has
happened with some frequency in the past.
The input MIDs need to be determined a little more carefully, though.
faster improvements (and more boldness) in the earlier cycles of Sequential
Empirical Optimization (SEO)
This feature enables
representing various components which may or may not be operating at any one
time from one ULTRAMAX application file; that is, with one
Optimization Plan (conceptual model), one set of run data.
Alternate operating components may be: physical modules (burners,
filters, nozzles); raw material used (including alternate suppliers); supporting
material used (fuels, cooling fluids, tool bit alloys); procedures
(cooling or not); etc. Each possible
combination of operating components is called a “Scenario”. This
represents uncontrolled inputs without the traditional “gradual” requirement.
The benefits of managing all
scenarios from one application file is to provide much more flexibility and
simplicity (set up, automatic starting of a new scenario, restart SEO and
maintenance), plus faster overall optimization.
Previously, each scenario required
its own application file (as many as 60 in one case), with the Optimization
Plan with its particular operating variables and Prior Regions; all of which
represented significant overhead and care.
Also, general changes such as including new variables and changing the
values of the Global Factors to represent current management priorities and
economic factors, is now made only once, in only one application file.
A rewarding aspect of this
simplification includes not having to decide up front which scenarios will be
used in the future.
Three new elements enable this
Inputs: This is a new type of variable, with
the values of 1 if the
component is operating or 0
if not operating.
Component inputs are uncontrolled inputs (role 2) – i.e.,
with values determined externally to ULTRAMAX.
- Dependency field: This is a new
field in the Optimization Plan. The Dependency Field is either blank
or has the name of a component input.
Field can be used both on inputs and outputs.
If the dependency
field is undefined or blank then the variable is always operational
(the most common case, and as it happens when uploading older ULTRAMAX
application files). If the dependency
field is defined, then see next.
- Conditional Variable: If the dependency field is
defined then variable acts as a conditional variable, with values only
when its “component variable” is operating (has the value of 1); and
with undefined value otherwise (has the value -777).
applications with component inputs optimize faster because they are able to extrapolate
information from similar scenarios, particularly valuable before much
data is collected.
of application size limitation. There are no longer limits in the
application size (# inputs, # outputs, # run data).
This feature yields three main benefits:
Keep older historical data to know how the process behaved for older
experiences of combinations of uncontrolled inputs – resulting in faster
optimization when those combinations are (almost) repeated.
Keep old run data that is no longer valid or complete for eventual
plotting and analysis (rather than automatically loosing older data that
exceeded the available size)
Avoid the trouble of the customer having to get a larger version of
ULTRAMAX when wanting to overcome size limitations
As far as the code is concerned ULTRAMAX can work for any number of
inputs; however actual experience has been for at most 55 inputs.
There have never been any practical barriers to the number of outputs.
There is a limit to
size based on the computer itself. We
have not experienced these limits because most computers today have ample
capabilities. At present the following
requirements seems adequate for about 40 inputs, 60 outputs and 5000 stored
run-data sets: 500 Mb ram, 3GHz, 250Mb available in the hard disk. Each application of this size will takes
less than 20 Mb of hard disk.
8. Adjustable Thresholds of Alert
threshold values of alerts to trigger a Note, to Warning to ALARM is now
adjustable by the user, as part of the ULTRAMAX Parameter values (see Blue
Book). This enables the user to select
the sensitivity or frequency of the Alerts (and associated risks) given the
actual behavior of the process. As
problems are solved and the process becomes more reliable (Quality Control),
the user can tighten the thresholds.
Integration; Optimization Management
Integration is the
automatic data exchange with the digital control systems (one way or
There is a
new automatic OMS supported at the optimization workstation
The benefit is to simplify
integration by automatically providing an operator interface before or in
addition to the end-user installing an OMS – of their design -- directly in
the control room consoles.
Among various functions, the
optimization workstation OMS displays current data, run-average data, the
adjustment advice provided by the SEO technology, and the operator desired
next adjustments. There are buttons
adjustments (the whole set of control inputs in a predefined time profile)
three operating modes:
o Manual Supervisory Control: where the operator adjusts
setpoints, biases, etc. as he/she would in the console but in a central
location, without the aid of SEO. This
is very useful to develop practice and confidence that the integration
installation and data collection procedures are correct; and also to collect
Baseline operations data.
o Advisory Optimization: where the operator has to approve
each new adjustment advised by SEO. The
operator presses a button to implement the Advice or a revision typed in by
o Closed-Loop optimization: where the Advice is implemented
automatically. Closed-loop reverts to
Advisory automatically when the SEO detects an upset or when the operator
turns it off.
now supports OPC communications
protocol in addition to DDE. OPC is more robust and reliable that
DDE. Other special needs can be
satisfied as well by our integration supplier (at extra cost).
definitions in the configuration of the Ultramax Even Driver (UED) is now
updated automatically from the Optimization Plan. This reduces the work and the chances for
errors when changing the Optimization Plan.
The tags, of course, have to be defined and maintained by the
Note, as well, that we no longer
refer to the Character User Interface (CUI) version of ULTRAMAX.
Simple calculations are, as before,
provided by the GUI. More complex
calculations are now provided through DLLs (vs. direct linking). Ultramax Corporation (UMC) or the Agent can
provide the DLL.
11. Transient Optimization: With Transient Optimization data collection does
not have to wait to achieve some sort of operating steady state. Data collection starts immediately after an
One immediate result is faster optimization.
Transient Optimization requires new variables to represent past
behavior and which affect current transient performance. This reduces noise and thus enables higher
of the effects of having transient technology is to recognize that optimal
adjustments are different whether some uncontrolled input value is increasing
as compared to decreasing.
Cycle times (time between adjustment) is restricted to be larger than the
half-time of the transients.
12. Profile Modulation:
Profile Modulation is a mechanism to adjust similar inputs that are
next to each other (such as temperatures, dampers, etc.) and to apply plant
personnel knowledge that the adjustments should change gradually from one
location to the next. Based on this
knowledge, optimization can be approached much faster than optimizing the
inputs as independent physical adjustments.
can be defined in one dimension (like temperatures in an injection molding nozzle),
in two dimensions (like damper openings in an array of burners in a boiler),
The set of physical
adjustments across the different locations is called a “profile”, and the profile is determined by a few parameters, many fewer than the number
of physical adjustments. Optimization
decisions (virtual adjustments) are made on the parameter values, while
ULTRAMAX also displays the corresponding values of the physical adjustments.
parameters might be: average, slope, curvature, etc., or any other devised by
the user. This method could be used,
for instance, to assure some sort of symmetry.
parameters appear as control inputs (role 1) (i.e., the decision variables) in the Optimization Plan,
while the physical adjustments are ruled inputs (role 8). Thus, optimizing
the profile parameters provides improvements – while restricted to the
geometry of the profiles -- much faster than optimizing the physical
adjustments independently. The time
taken to approach optimal performance is about the ratio of number of
parameters to number of physical adjustments.
The number of
parameters can be increased with time to represent more complex profile forms
(e.g., like a Taylor expansion, more accurate with more terms) and eventually,
when there is plenty of data, easily transition to adjusting each
physical position separately for
the operations of a plastic injection molding machine includes optimizing the
adjustment of temperatures in the nozzle. This nozzle has one
temperature control every three inches, nine in total.
Take the adjustment of each of the nine temperature controls as a
separate control input in the Optimization Plan.
Profile Modulation approach:
Experienced people believe (or know) that the temperature should
change gradually in the nozzle from location to location.
One simple profile representation is to assume that the temperature
should follow a linear relationship along the location in the nozzle. Such a profile can be defined by two
parameters: average and slope (another possibility: first and last
With Profile Modulation the two parameters are control inputs
(decision variables). The nine
physical adjusted temperatures are “ruled inputs”, calculated by the Profile
The advantage of using Profile Modulation is to optimize faster (seven
fewer independent control inputs), although being restricted to the linear
relationship postulated by the user.
Note that after optimizing the average and slope, it is easily
possible to increase the flexibility of the profile. E.g., we could add as a new control input a
quadratic component -- which up to this time was zero -- to refine
There can be
several Profiles in one Optimization Plan.
such as Integration and Transient Optimization, is available at an extra
13. Minimum Important
With the elimination of the Prior Region, the meaning and demands
for the input MID has changed.
As in the past, the MID is often a somewhat fuzzy concept that does
not have to be defined with much accuracy in order to optimize effectively.
The new summary of definitions
The MID for the Objective
Function (Scorecard) is the minimum difference in value that managers
consider important from a business point of view (same as before).
The MID for
outputs with constraints are the maximum violation of constraints by
individual data that is
tolerable to management (same as before).
(Now it is also possible to enter MID=0 to represent a lack of any
gray area around the constraint, e.g., a gray area due to errors in
The two above
indicate that ideally, for prediction reliability from a business point of
view, the noise should be less than 1/3rd the output MID.
MID for the control inputs are:
defines the maximum change in adjustments considered acceptable from
one optimization cycle to the next – typically to assure stability of process
operations. (In previous versions it
should be sufficiently large such that they create an MID difference in
The MID for
uncontrolled inputs are the estimated total range of values.