The fascinating (nuclear) history behind Oracle’s new anomaly detection service

July 12, 2021 | 2 minute read
Text Size 100%:


This month, Oracle unveils a new AI Service called Oracle Cloud Infrastructure Anomaly Detection Service. We are thrilled to bring you this powerful diagnostic service for highlighting multivariate anomalies in complex systems. As new as it is, however, I’d be remiss if I didn’t also tell you how old it is — 21 years old, to be exact!

Spotting problems before they become problems

At the core of OCI Anomaly Detection Service is an innovative algorithm called multivariate sampling estimation technique (MSET). Originally developed in 1996 by the US Department of Energy’s (DoE) Argonne National Labs, MSET was designed to monitor nuclear power plants and ensure they are safe and secure. Today, all 93 US nuclear plants have incorporated MSET for monitoring services, and nearly all the 450 nuclear plants worldwide use it as well.

In the early 2000s, a startup called SmartSignal commercialized MSET, successfully expanding its use cases to include monitoring:

  • Crewed and crewless NASA space shuttle flights,
  • Structural and ride safety systems at Disney's 10 global theme park rides, and
  • Jet engines with Delta, Lufthansa, and Southwest airlines.   

In 2011, GE Digital acquired SmartSignal and made it the foundation for its popular Predix asset monitoring business.

153 patents now at your fingertips

So how did Oracle get involved with MSET? In 2000, Dr. Kenny Gross, the lead developer of MSET, left Argonne Labs and joined Sun Microsystems. Sun began investing mightily in the promising technology, earning multiple patents and upgrading the algorithm to a version called MSET2. In 2009, Oracle acquired Sun and continued to discover new ways to apply MSET2's wide-ranging applicability for watching time series data from complex business assets. Over the last 21 years, Oracle Labs has built up a dense portfolio of 153 patents integrating MSET2 with other prognostic algorithms, increasing sensitivity while reducing false positives (often beating neural nets, support vector machines, and kernel regression in this respect).

Thanks to the success we’ve had helping customers with MSET2 in varied verticals — including utility, oil and gas, transportation and manufacturing — Oracle decided to release a generic AI service so that developers could easily incorporate MSET2 into their own asset monitoring and diagnostics systems. The OCI Anomaly Detection Service includes all the patented capabilities for data quality, preprocessing and early detection techniques and is generally available on July 15, 2021.

In summary? A product built over decades of research and tested by extreme conditions (nuclear plants, rockets, roller coasters…) is now at your fingertips. Try it in our live lab today!


Jim Rohrkemper

Jim Rohrkemper is a senior technology executive with deep industry knowledge in automotive, manufacturing, and technology. His leadership experience includes client executive at IBM, VP/CIO at Flint Group, president at Precisia LLC, and director of IT PMO at Ford Motor Company. Jim has led and sponsored many complex projects in ERP, supply chain, procurement, and customer experience.

Previous Post

Graph Machine Learning with PyPGX and OML4Py

Rhicheek Patra | 14 min read

Next Post

Deploy remote diagnostics without in-house data science and ML experts

Viji Krishnamurthy | 3 min read