Machine Learning is Everywhere, What’s the Difference?

There is a lot of talk these days about analytics and machine learning. It seems that machine learning is everywhere.  But not all analytics and machine learning are industrial-strength and few can help the world’s largest, most important industrial companies drive best-in-class performance. What does it really take to create a world that doesn’t break down?

You can start on-line with self-help machine learning, with projects kits from Microsoft in Azure, MatLabs, and others. It’s all the same; right? Not quite; we’re talking machine learning for asset performance management; with potentially messy and incomplete data and requirements to collect from real-time data sources – not data in a file. 

Industrial Strength Machine Learning

First, machine learning is not a silver bullet. It needs guidance by people who understand the processes and machines to be monitored. Those people give context to machine learning to make sure the discovered correlations make sense. It is not just a case of crunching data to see if anomalous patterns appear. Additionally, if machine learning is to be valuable, it cannot be an offline tool – that takes too long and takes great skills and experience. It must work fast, inline, in real-time, all the time giving accurate advice when machine and process behaviors indicate precisely know failure patterns and deviations from normal behavior – both. 

In-line and real-time coupled with an automated framework can deliver the earliest, most accurate warning of equipment failures and unplanned downtime. Further effort can incorporate the capability to prescribe detailed actions to mitigate or solve detected issues. Additional to machine learning, such prescriptive guidance is based on established root cause analysis and presents information on the approach that will proactively avoid process conditions that cause damage and/or advise on the precise maintenance required to service the asset. 

Make it work for Joe Normal

Build into that framework an abstraction mechanism so that engineers who understand the problems can exercise the machine learning without intense data science skills and now you have an application that fits precisely with the work processes and skill sets available at the processing plant. “Joe Normal” can do it easily, rapidly, accurately, and scale his work output across a plant unit, the plant site, and even across the multiple locations in a corporation.  

His results deliver extremely early trustworthy alerts and drastically reduced false alarms; providing very high accuracy degradation pattern recognition that eliminates the need for additional resources to interpret the data and consult on what action should be taken to avoid failure. And it’s not just late-stage detection of damage, but detecting the upstream root causes of damage to advise adjustments to avoid process conditions that cause the most failures. Operators and maintenance staff are now confident in the failure prediction and act together to avoid unplanned production downtime. 

Pick the industrial-strength machine learning application.

About your Guest Blogger:  

Michael Brooks is Senior Director, Business Consulting, Asset Performance Management (APM) Business Unit at AspenTech. Previously, Brooks was the President & COO of Mtell. Founded in 2006, Mtell is a privately-held company, providing software solutions for managing the health of industrial equipment. Brooks also was a Venture Executive with Chevron Technology Ventures and held senior roles at Infobionics, Vision Solutions, INDX Software Corporation and Wonderware. He started his career as an engineer at Chevron and Esso.  Brooks has a degree in chemical engineering from the University of Bradford.