For a large midwestern energy production company, safety, environmental stewardship, and operational efficiencies were priorities. Additionally, minimizing processing and maximizing output was key to lowering costs and increasing profit. Particularly compelling was the chance to address process variability. The goal was to identify periods of peak performance at several plants, understand which control settings were used, then replicate peak performance.

After reviewing the functionality, ease of use, price, and support of advanced analytics offerings, TIBCO® Data Science software was determined to be best suited for the full list of requirements. Its intuitive user interface streamlines data mining, preparation, and modeling and further simplifies use by modern citizen data scientists by easing preparation of structured and unstructured data. Reusable process templates make it even easier to share and distribute analytic workflows to non-technical users.

The company also appreciated seamless connectivity to its OSIsoft PI System, which captures, processes, analyzes, and stores any form of real-time data. This direct link would enable team members to browse and import data from the PI System into TIBCO Data Science software to streamline both simple and multivariate process monitoring. TIBCO also offered an advantage in training and troubleshooting. TIBCO's advanced analytics experts were always available, while a comprehensive set of online tutorials was extremely useful in helping team members take full advantage of all functionality.

"The direct link to our PI System was desirable. It's a major advantage to browse and import data from the PI System into TIBCO Data Science software to streamline both simple and multivariate process monitoring," says the asset performance and benchmarking engineer.

Immediate Feedback for Process Improvements

The key would be determining how a host of controllable and uncontrollable variables impact performance so modifications could be made in real time. With predictive data mining and analytics, the company could get immediate feedback, resulting in significant process improvements.

"The more efficiently we process gas, the better service we provide our customers while supporting our bottom-line business and profit objectives. The power of TIBCO Data Science software is that it enables us to use past performance indicators to maximize future plant performance."

By replacing manual data mining and analytical methods, TIBCO Data Science software enabled the company to quickly and easily harvest data on controlled and uncontrolled process variabilities. TIBCO Data Science software then directly links to the OSIsoft PI System for further analysis, providing operators with real-time data on how to maximize processing performance.

Reduced Costs from Analyzing Performance Metrics

In processing natural gas, a team of engineers works to remove CO2 contaminants. The goal is to determine how much gas requires treatment and how much can be bypassed. The right treatment/bypass ratio saves money.

In examining the process, inconsistencies were discovered. Performance often fluctuated from shift to shift, based on minor modifications of controls. Operators from different backgrounds with different experience and comfort levels made subjective decisions on modifications.

The company created a roadmap to remove subjectivity. The team gathered and analyzed data to determine which factors most affected peak performance—the amount of gas bypassed, ambient and lower amine outlet temperatures, inlet CO2 levels, flow rates, and steam flow rates were examined. Initially a heavily manual effort including gathering data and entering it into a spreadsheet, the more sophisticated TIBCO Data Science solution automated data mining and predictive modeling and brought meaningful insight needed to quickly improve processes and operational efficiencies.

Ability to Determine the Right Control Settings

With interactive general classification and regression trees (GC&RT), for example, data is separated into packets of roughly homogeneous inlet CO2 levels to help predict continuous or categorical variables from a set of predictors and/or factors. What makes this particular data mining method so effective is that a tree can be followed from the beginning to identify the exact set points of the controllable parameters. The team also leverages multivariate adaptive regression splines (MARSplines) to iterate further and perform what-if analysis to fill any remaining gaps in set points. This completes the puzzle and assigns the right values to the appropriate set points.

Armed with this level of detail, the company then leverages OSIsoft to provide operators with real-time data for current operating conditions. A color-coded display makes it easy to see everything at a glance without having to translate or calculate numbers, which streamlines any required processing modifications.

With the TIBCO Data Science solution, this energy leader is prepared to handle ever-increasing amounts of big data without concern for limits or analytical constraints. "I have yet to come close to pushing any boundaries on TIBCO Data Science software," says the asset performance and benchmarking engineer. "I look forward to delving into all that it has to offer as it's already provided us with real, measurable value."