Q&A ’15: Predictive analytics allow operators to maximize refinery data

By Ben DuBose
Digital Editor

NEW ORLEANS, Louisiana -- Predictive analytics have the potential to significantly enhance the performance of refineries and petrochemical plants in the years ahead, the principal consultant at Emerson Process Management said this week.

Speaking at the Plant Automation & Decision Support session, Emerson's Doug White explained how extensive developments in predictive analytics have greatly improved the potential quality and accuracy of future plant behavior, including potential production and supply chain alternatives, early detection of potential equipment problems, and product quality issues.

"We're now able to move the data to the expert, not the expert to the data," Mr. White said.

The Emerson consultant used the example of "Moneyball" in baseball to explain how analytics had transformed an industry, and he believes the same can happen for refiners and petrochemical operators.

Mr. White sees predictive analytics helping in four primary areas. One is safety, led by avoiding incidents through the early detection of potential hazardous situations. Another is availability/reliability, which predictive analytics can aid through anomaly detection by identifying precursor events to unscheduled equipment outages or problems. Likewise, performance monitoring can allow operators to detect a loss in process or equipment performance before it impacts production.

Meanwhile, the third target area is sustainability, where analytics can allow the industry to better compare the current usage of energy resources to expected usage and determine possible causes of variation.

The fourth and final emphasis comes within financial optimization. There, analytics can help operators to detect and dissect complex interacting constraints on production and also determine reasons for issues in product quality and yield.

"It helps us understand patterns and relationships by developing statistical models that explain them," Mr. White said.

Gathering the necessary data is not easy, though. For starters, the industry needs to be able to preprocess and shape data before it can be used to generate models, he said. Moreover, the same techniques must be able to be used off-line to generate models and on-line to preprocess data before it is used to generate predictions, provide fault detections, and generate recommendations.

Beyond that, operators must also remember that correlation is not causation, Mr. White said.

The Emerson consultant then shared a case study on how the use of predictive analytics worked at a wastewater treatment plant. At the site, multiple effluent NH3 spikes in outfall were seen violating the consent decree, thus incurring fines for the company.

Making matters worse, the residence time of the treatment plant was greater than two weeks, which meant that making an issue ID would be difficult. In fact, more than 200 measured variables were considered as possible predictors.

But the use of predictive analytics allowed the company to analyze five years of data for correlations with a simultaneous identification of time lag. From that, the company identified a strong correlation with the pH of effluent from one of the plants, where a large drop in pH preceded excursion by one day.

The vessel cleaning was then done in correspondence with the pH drop.

Likewise, Saudi International Petrochemical Co. (Sipchem) recently implemented a complete integrated maintenance program using analytics at its world-scale Jubail complex. The project scope included the implementation of asset management, including reliability-centered maintenance, on 6,000 assets while using predictive analytics and diagnostics from field instrumentation.

The final result was a 12% decrease in maintenance costs and a 2% increase in plant availability.

"Predictive analytics is an evolving technology with many potential applications in refining and petrochemicals," Mr. White said. "The current applications in process fault detection, availability, safety and optimization already have proven value, and there should be even more in the future."

Related News

From the Archive

Comments

Comments

{{ error }}
{{ comment.comment.Name }} • {{ comment.timeAgo }}
{{ comment.comment.Text }}