Agents in the Ether: They’re on a Machine Learning Mission

Agents in the Ether: They’re on a Machine Learning Mission

January 10, 2019

What do you think about when you hear the word “agent?” Maybe someone who’s on a mission, operating in the shadows to find a critical piece of information? There’s always a bit of mystery surrounding famous agents like James Bond or Mulder and Scully, but they still manage to get the job done and save the day.

When we talk about “agents” in the world of asset performance management, we’re actually talking about electronic agents that are acting on your behalf. Simply stated, they’re software components that take on the heavy lifting of machine learning — so you don’t have to!

Aspen Mtell® utilizes a pair of particularly effective Autonomous Agents that detect normal behavior and specific patterns leading up to failures in bearings, drive-couplings, seals and other components. Importantly, these agents also work to “gather intelligence” — that is, they learn and adapt over time.

These two fundamentally different types of Autonomous Agents — Anomaly Agents and Failure Agents — offer different ways to find degrading or failure conditions. Let’s learn a little more about them, including their missions and special capabilities.


Anomaly Agents

Anomaly Agents ask, “Is this normal behavior?” These agents have learned the precise patterns of normal, so they can detect failures early and with greater accuracy than modeling techniques. They also adapt to changes in process behavior to ensure the accuracy of all detected failures without false alarms.

Because of the waveform, pattern exclusion and machine learning used to define “normal,” Aspen Mtell’s detection of anomalies is far more precise and warns earlier than other systems. When it detects a deviation, it is real and accurate for any failure or “new normal” condition.

And when a deviation is declared to be a “new normal” behavior — that is, something never seen before — the Anomaly Agent retrains itself to recognize the pattern, so it will not send out false alarms. The Anomaly Agents actually capture and retain all the behavioral knowledge of changes in operations!


Failure Agents

Failure Agents are trained by measuring the actual behavioral patterns that begin early in root cause conditions and lead to very specific failures (e.g., a bearing failure). So, by constantly scanning incoming data to detect recurrences, Failure Agents provide far greater accuracy and much earlier warnings of machine and process issues.

Aspen Mtell analyzes data using machine learning to “fingerprint” the precise patterns for a specific root cause. Failure Agents are trained using Aspen Mtell’s proprietary data conditioning and sophisticated machine learning algorithms — all “under the covers” to keep it simple for end-users.

Where anomaly detection asks, “Is this normal?” a Failure Agent asks, “What is the exact pattern that led up to this failure?” So, the Failure Agent technique is based on actual failure patterns, proving to detect issues earlier and with greater accuracy than other contemporary systems.


What’s So Special About These Agents?

The differentiating feature of Aspen Mtell’s Autonomous Agents lies in their capability to learn precise patterns from the time-series data from sensors on and around machines.

These types of patterns do not emanate from models, determined by humans from thermodynamic, stoichiometric, physics, heat and material balance (or other mathematical/statistical equations and approximations). Rather, they are determined from the precise behavior of the machines, declared in the time-series data streams from the sensors that monitor them.

Both types of Autonomous Agents also come with the requisite advanced communication capabilities that any good agent needs, as they can send notifications via text messages, emails and dashboard alerts. Failure Agents can even send detailed descriptive information about the failure and advise prescriptive action!

The bottom line is, these are agents that you want on your side as you work to optimize asset maintenance and maximize system reliability. Learn how you can get them working for you in my recent white paper Ramp up Reliability With Low-Touch Machine Learning for Hyper Compressor Monitoring.

There was a problem storing your subscription

Leave A Comment


  • 5 years ago

    Very interesting topic, anxious to explore how these capabilities fit SKF interest to support bearing performance improvement with our Customers

  • 5 years ago

    Very interesting topic, anxious to explore how these capabilities fit SKF interest to support bearing performance improvement with our Customers