Machine Modeling Explained, Oil and Gas Examples

On the third day of implementation, one anomaly agent alerted and exposed the cause of a compressor failure that had plagued the refinery for over a decade.

Pilot projects implementing prescriptive, predictive or proactive analytics applications in manufacturing are being documented quite regularly now, be it aluminum production or wind power generation. One of the exciting trends within these disciplines is machine modeling and the importance of cleaning data, time-stamp or event-based data, found in decades-old plant historian databases,

At the recent 2018 WindPower Conference in Chicago, Dr. Sandeep, ceo of Ensemble Energy, Inc. , talked about this very topic of data cleansing. “With machine modeling, cleaning up data is approximately 80% of our solution. The question is how well the data is structured.” I’ll have more from on Ensemble Energy’s data modeling and physics-based approach — first principles thinking.

A recent white paper from Aspen Technology, Inc., titled, “Seeing Into the Future With Prescriptive Analytics: A New Vision for Asset Performance Management,” discusses data cleansing and other reliability issues related with machine modeling for three different pilot projects. See below:

For example, a major oil and gas (O&G) company was experiencing recurring, unexplained breakdowns of compressors at one of its refineries. The staff was a mature implementer of reliability-centered maintenance methodologies and used state-of-theart vibration systems, but still the breakdowns occurred.

The author, Robert Golightly, senior product mkt. mgr. at Aspen Technology, Inc., goes on discuss how the machine modeling solution allows the O&G company to get identify the root cause or provided a failure pattern recognition with subsequent notifications to follow.

The heart of the data cleansing is the platform’s Aspen Mtell® functionality with its low-touch machine learning approach that eliminates much of the manual effort involved in “data wrangling,” according to Aspen Technology.

Another interesting aspect of these pilot projects is the time period for completion. Aspen Technology says “the pilot projects were all completed in less than a month and, on average, in about 2 ½ weeks.” The paper isn’t clear on the total timeline and what’s involved, but  Sandeep at Ensemble Energy cited “3 to 6 months” for their machine modeling approach. 

If you need an accessible explanation of machine modeling in action and case applications examples, download this white paper here.

>> White Paper | Seeing Into the Future With Prescriptive Analytics: A New Vision for Asset Performance Management