Koch Ag & Energy High Value Digitalization Deployments Leverages AWS

Summary

Dave Kroening, IT leader, and Martin Miller, data analytics leader with Koch Ag & Energy, discussed two high value digitalization applications using AWS (Amazon Web Services) including Monitron and SeeQ software in a re:Invent session. Both applications followed a basic digitalization strategy:

• Think Big
• Start Small
• Scale Fast

Basic building block: Monitron Sensors

The basic building block in both applications described are highly cost-effective AWS Monitron IoT sensors (pictured above) that non-intrusively capture equipment vibration and temperature profiles communicating data to AWS cloud applications.  The Monitron devices are self-configuring requiring no applications programming and in most applications are simply glued onto equipment.

Application: Less critical equipment predictive maintenance

Monitron was applied to less critical equipment many times referred to as balance of plant to achieve predictive maintenance improving uptime and availability. The low cost and ease of application of Monitron sensors and AWS predictive cloud software has now made this practical. Continuous monitoring of vibration and temperature as key indicators coupled with cloud analytics provides operating and predictive information. The Monitron sensors are simply epoxy glued onto equipment.

Plant people set up Amazon Monitron Sensors using their smart phone following a few simple steps in the free Monitron App and picking the ISO 20816 vibration standard for vibration for the equipment being monitored and cloud system uses this information to automatically select the proper Machine Learning (ML) model.  The AWS application automatically detects abnormal machine operating states by analyzing vibration and temperature signals and provides notifications to plant people via Amazon Monitron App. Plant people can also review and track these abnormal machine states within the app in any time. Closing the loop plant people can use the app to enter feedback on the alerts received, such as failure mode, failure cause, and action taken. Amazon Monitron learns from that feedback and continually improves over time.

These are the steps following the company’s digitalization strategy:

• Think Big: monitor vibration on all assets to provide insights to health, increasing asset availability
• Start Small: quickly deploy sensors on a small subset of assets gain insight that enable action prior to failure
• Scale Fast: The first step was deploying 125 sensors across five sites proving the concept and is now being scaling to over 2,000 in the next 18 to 24 months.

This represents a shift in the maintenance philosophy for this equipment that had been running these devices to failure including pumps and motors. 

Shift application: Modeling equipment behavior & diagnose issues

This application uses existing plant sensors, Monitron sensors, Amazon Lookout and SeeQ software to implement predictive maintenance on more complex equipment. The goal accomplished was successfully implementing predictive maintenance requires data from thousands of sensors to gain a clear understanding of unique operating conditions and applying machine learning models to achieve highly accurate predictions. In the past modeling equipment behavior and diagnosis issues requiring significant investment in time money inhabiting scaling this capability across all assets. 

All user interaction an information displays are through the SeeQ interface which is designed for plant people. Using AWS Lookout Machine Learning models are developed from machine sensors with minimal or no IT or Operations expertise in hours. The user selects an asset and sensors inside of SeeQ and selects training model with expertise from Black & Veatch. The data is then moved to AWS S3 cloud database and Lookout for Equipment to train a model using the asset data and is completed in less than one hour. At that point Lookout is ready to accept real-time data for analysis. 

These are the steps following the company’s digitalization strategy:

• Think Big: cost effective holistic approach to leverage all data from machine sensors
• Start Small: deployed across five sites and 41 critical assets
• Scale Fast: scaling to over 100 critical assets in the next 12 months

Accurate & transparent models

Martin Miller described how this approach met their goal to have accurate and transparent ML models, “We didn’t want a black box, we wanted the ability for the operators and engineers as they built these models to get the feedback to understand the conditions, operating states, what is a model actually telling me and what’s driving our potential failure.”

Thoughts & observations

The applications at Koch Ag & Energy reflect the commoditization and simplification of applying analytics and machine learning using IoT edge devices. Commercial analytics and machine learning technologies have advanced far beyond proprietary industrial and process automation offerings with a wide range of tools easily accessible with cloud applications. The combination of powerful analytics and machine learning combined with software such as SeeQ that have been created for manufacturing and process industry people who are subject matter experts certainly expands opportunities for manufacturers to be more productive and profitable.