Improving Processes by using Sequential Empirical Optimization (SEO)

  • September 13, 2013
  • Ultra Electric & Automations
  • Feature

By Bill Lydon, Editor

The ability to improve manufacturing operations is growing with the application of software that can leverage lower cost computing. Back in 1982, the ULTRAMAX Corporation commercialized an interesting software product called ULTRAMAX to optimize the operating performance of industrial processes. I recently caught up with Carlos Moreno, the Founder and CEO of the company, to learn more about the product and their philosophy and perspective of industrial process optimization. This article outlines my interview questions and Moreno’s responses.

Based on your experience, how many industries do you believe are running near optimal performance?

I have not found a relationship between industries and the level of suboptimization of production/manufacturing operations with existing assets. So, what follows applies broadly to all industries.

Of the processes optimized with ULTRAMAX, about 1% ended up proving – probably for the first time – that they were already running near-optimally. On the other extreme, about 5% ended up proving that the processes were incapable of meeting minimum requirements – something else was needed to be done with the process. About 94% did get improvements in operating performance - some more, some less.

I tend to believe that this is representative of industry in general. As a matter of principle, exceptions ought to apply to processes with very thorough and well maintained regulatory and supervisory control systems.

With so many process optimization software approaches on the market today, what is different about ULTRAMAX?

I cannot respond authoritatively for the technologies to optimize Regulatory Control. I’ll restrict my answer to technologies for optimizing Supervisory Control: finding the best possible balance of the operating performance metrics by determining the best combination of setpoint values. Setpoint values are the directions given to the existing Regulatory Control on how to run operations.

Just to add a detail, Supervisory Control can also provide feed-forward, longer-response regulatory control for those desired targets beyond the reach of the regulatory control in the process control system, such as for certain product characteristics.

The difference in the ULTRAMAX solution for supervisory control is that it is based on the technology of “Sequential Empirical Optimization (SEO)”. This offers two main benefits:

First, neither the user nor the supplier needs to start by creating a process prediction model (outputs as a function of input values). The SEO cycles (see diagram below) creates the models automatically, in the background, in the course of implementing and optimizing a sequence of setpoint adjustment decisions. This provides speed and simplicity.

Second, the paradigm in ULTRAMAX is to implement repetitively the same intensive and sophisticated analytics every time new run data is obtained (the bottom function in the graphic below). It refines models with each new run data and advises the next set of adjustments suggested for the system. This allows the operator to manually accept or modify adjustments and when they have confidence they will press a button to engage closed loop optimization if installed. Since the operator makes the final decision this is a very safe environment.

Other solutions have particularly intensive technologies when creating the models up front, and they required that associated expertise.

Optimizing Supervisory Control

Editor’s Note: This diagram illustrates SEO

Just to be clear, ULTRAMAX does not require the development of a model of the process?

Correct, ULTRAMAX does not require development of models. In that sense is it is unique and this means it is easy for anyone without knowledge of the analytics of model building.

How would you describe the ULTRAMAX technology?

Production efficiency is managed by decisions on the adjustment values of the controlling setpoints. Production efficiency is evaluated by how well it operates within required input/output constraints that represent basic requirements, plus improvement goals (e.g., cost reduction, increase througthput, sharpen quality) - all based on user-collected metric values.

As operations proceed, the ULTRAMAX’s intelligent analytics learns from the accumulation of the run data generated and determines better set-point adjustment values (tuning) for the following runs.

Improvements continue until you get to the optimum performance capable by the process equipment and process control system or until the user is satisfied with the results.

Thus, the user maximizes contribution to enterprise goals (e.g., productivity, economics) while maintaining reliability and complying with basic requirements including safety, quality, regulations and production commitments.

Editor’s Note: This diagram illustrates SEO Optimizing information and decision flows

How does ULTRAMAX perform when there are varying raw material and environmental conditions?

Varying raw material characteristics, ambient conditions, and demand on the process are represented by the “uncontrolled inputs” that also affect outcomes. Uncontrolled inputs which vary but are not measured appear as “noise” or unexplained variations in output values, together with sensor errors and imperfect regulatory control. Those which are measured are dealt with very effectively by the ULTRAMAX version of SEO in that it modifies adjustments to maximize performance for the uncontrolled input values. This dynamic feature is a critical advantage for some processes, where otherwise fixed adjustments give significant inferior overall performance.

How much can traditional regulatory control systems be used to optimize processes?

Regulatory Control, traditional or advanced, is a system that can be optimized in its own right, namely, how best to make process variables have the values the user wants.

It leaves unanswered the question and opportunities of optimal Supervisory Control, given the existing process and process control, what setpoint values maximize overall value contribution to the business. This is a higher order level of control.

If the process control system has control logic to define values of setpoints depending on conditions, supervisory control optimizes performance by optimizing the “bias” that is added to the simpler control logic setpoint values.

In turn, there are even higher level systems in production/manufacturing, such as supply chain, production planning and scheduling (what product to make, how much, when), transportation, storage centers, etc.; each of which can be optimized, and depend to some extent on the implementation of the other systems.

Different systems have different frequencies of interventions/decisions. The more frequent the interventions, the higher the demand for automatic implementation.

How do you determine the best possible process operation in a plant?

Optimizing the Regulatory Control function has many solutions offered by several control technology companies.

Optimizing the Supervisory Control function has had several solutions in the last century, such as First Principle modeling, Design of Experiments, EVOP (Evolutionary Operations), Sequential Empirical Optimization (SEO) and Neural Networks – in order of invention. All but SEO start by creating a process model. If the models are sufficiently accurate predicting the process response and sufficiently complete representing business needs, applying the optimization of the models will result in optimal process operations.

How does ULTRAMAX differ from model based methods?

As mentioned above, the main difference is that with SEO, the user does not need to create a process model up front.

To clarify, ULTRAMAX is also “model-based”, but the models are built automatically and kept in the background as it implements the SEO cycles. A benefit of this feature is that continued use of ULTRAMAX will keep its process models up to date with changes in the process. The internal models are available for observation and plotting.

What changes are required to existing controls and automation when applying ULTRAMAX?

If ULTRAMAX is applied manually, that is, manual data entry into the software and manual readjustments of setpoints, then obviously no changes are required.

If ULTRAMAX is applied closed-loop (with automatic data exchange with the digital process control through OPC) certain things need to be added to the process control system such as a “heart-beat” detectable by ULTRAMAX, and to let ULTRAMAX have supervisory control of the selected setpoints. An implementation principle is that if the user were to stop using ULTRAMAX the process continues working as before, what we call “non-intrusive”.

Also, if the process control system includes control logic to adjust setpoint adjustments and the setpoint does not include a “bias”, then a bias needs to be added, as the bias is what ULTRAMAX optimizes.

Also, since ULTRAMAX makes it so simple to be more business-wise comprehensive by adding more metrics of performance to be balanced, some metrics may not have sensors or tags, and these will have to be added. This first application of ULTRAMAX is usually done based on the existing tags.

What special skills are required to apply ULTRAMAX?

There are two main tasks: planning and execution.

Planning is done by engineers who understand the process and managers who represent business requirements and economic factors. They need to know their processes and their business.

The execution is very simple. Operators have applied manual ULTRAMAX after participating in a few SEO cycles. However, there are reports to be interpreted to evaluate performance and novel situations. This requires an understanding of basic statistical measures such as average and standard deviation, and of the reports available in ULTRAMAX.

How long does it take to be trained to use ULTRAMAX?

To be introduced to the rather novel concept of sequential optimization, the planning for it, the execution and interpretation of reports takes about eight hours or training. It takes deep participation in a complete operations optimization project, including documentation, to become confident about the different procedures in applying ULTRAMAX. After some half-dozen applications in different processes, one is nearly an “expert” in the application of ULTRAMAX. Each optimization study takes about ten team personnel-days above and beyond regular operations.

What are the risks in applying ULTRAMAX to process applications?

A risk shared with other supervisory control optimization is for the user not to know the process sufficiently well to know which metrics properly represent the operations impact on the business, and which adjustments and uncontrolled inputs affect them significantly.

At the early stages of SEO cycles, SEO is still learning the basics of the process response patterns. The generation of adjustment of advice is based on max allowed change from run to run and the level of knowledge SEO can harness from the limited operating data.

Note that the penalty/risk in operating performance in the earlier SEO cycles is FAR LOWER than the penalty from experimental runs for design of experiments and neural networks, and validation runs for models created by any means.

Since the run data has noise - unexplained value variations - the prediction models created by SEO are not perfect. Noise is created by variations in unknown uncontrolled inputs, imperfect regulatory control, sensor errors, and occasionally missing data. Thus, run data may sporadically run the risk of violating output constraints. Constraints reflect basic operations requirements, such as for safety, quality specs, regulations, commitments, etc.

SEO Operations Optimization generates quickly a higher return from the existing assets. Not a risk, but a missed opportunity, would be to be satisfied with those gains and not to apply engineering analysis to the optimal operations to learn what makes the gains happen, to lead to better understanding in future operations and discover opportunities for process redesign.

Herein the larger macro-cycle with SEO: optimize operations with the current assets with a round of SEO cycles; analyze and redesign the process; and re-optimize operations, and so on; perhaps once a year. This is another example of continuous improvements.

What type of ongoing maintenance is required to keep ULTRAMAX operating properly?

Like all supervisory control optimization, the configuration of the optimization -- the I/O variables, constraints, cost factors, and performance calculations -- need to be updated to reflect important changes in specifications, material costs, limited availabilities, etc. This is a generic problem with analytical solutions; they need to be kept up-to-date to represent reality.

In addition, when the operating procedures are to have constant adjustments – i.e., no dynamic readjustment procedures to compensate for known values of uncontrolled inputs – the optimization needs to be repeated as the process change with time. This is done automatically for on-going on-line applications of ULTRAMAX Advisory or Closed-Loop.

Can you provide some examples of applications of ULTRAMAX and the benefits?

A food processing cooking application achieved gains of $1,080 per hour ($2 million per year) due mostly to an increase in production rate of 1.1 Klbs/hr and corresponding increased sales in a capacity-constrained environment, and an increase in yield. These gains were achieved Stand-Alone in a study of about three weeks of production.

In a 550 MW utility boiler in Zou-Xian, Shandong, China with 5 coal pulverizer mills, efficiency was improved while maintaining safety and reliability considerations. ULTRAMAX was integrated with the control system for automatic data exchange and efficiency improvements were obtained in the first 500 hours of operations.

When should somebody consider using ULTRAMAX?

When at least one important process output cannot be easily predicted with sufficient accuracy… In this situation operations need to be tested to see what really happens, namely, it requires an empirical solution (DOE, SEO or NN).

When the returns on engineer’s time to achieve improvements is competitive to other opportunities at the plant… It takes about 10 team personnel-days per optimization study (including about 3 days for formal documentation, presentation and approval). In high volume production, annual gains are usually from a few $100K to a few million dollars per process. Recall that about 1% of operations are already near-optimal to start with; where supervisory control optimization cannot yield any major benefits.

When there is a culture of doing nearly-as-well as possible with existing assets, to avoid waste, somewhat similar to Lean; but in this case not through redesign but through refined operations management. In addition, supervisory control optimization detects some barriers to further gains, which generates ideas for re-engineering changes, perhaps aided by Lean and/or Six-Sigma principles.

Learn More

Did you enjoy this great article?

Check out our free e-newsletters to read more great articles..