Internet of Things (IoT) Applications

Grid Connect Makes a Smart Cord for Industrial Applications with Machine Learning and AWS IoT

Grid Connect wanted to add AWS IoT and Machine Learning to their smart power cord to go beyond just monitoring.

When most people hear “smart device” their first thought is a home appliance they control with Alexa—but Grid Connect is thinking much bigger. They specialize in making smarter edge devices for industrial automation, and have designed everything from sensors and intelligent gateways to their latest product, a smart power cord for industrial machinery. The cord uses AWS IoT to monitor fluctuations in things such as amps, voltage, and watt-hours to provide proactive alerting to factory floor workers when things go wrong.

To understand why this is so impactful, say there’s an industrial cooling system that is critical to keeping a factory running. Suddenly, a belt snaps, stopping a fan. The cooling system is no longer operating as expected, which could impact the health of other nearby machines or cause downstream production issues that halt operations for days. Grid Connect’s smart power cord is able to quickly send alerts about the failure when it occurs, helping workers resolve the issue as quickly as possible and get production back on track.

While their first smart cord prototype was able to see outright failures, Grid Connect saw a chance to make it even better. What if the cord used machine learning to detect anomalies that may have flown under the radar before? With IoT and machine learning, Grid Connect could create a power cord that alerted workers about potential failures before they even had the chance to occur, allowing them to perform predictive maintenance.

Building such a device would be a big undertaking. Grid Connect would need to not only monitor data but collect it and analyze it with machine learning. They would need robust alerting for abnormalities, automated responses, and easy data visualizations with AWS IoT SiteWise to help workers understand and control what was happening with their machines. Most industrial machines wouldn’t be natively able to connect to the cloud, either, which meant building out a lot of custom infrastructure to give workers this level of data access and control. And because of its strong tie-in with AWS services, Grid Connect would need to go through all the processes to help them get the cord approved and endorsed by AWS itself.

Curious about IoT? We can help!

IoT Expertise

“We wanted to work with an AWS partner who could help us push this project along and get it approved by AWS internally, as well as make sure we were doing everything with AWS IoT in the most futureproof way from day one,” said Steve Sanders, Lead Application Developer at Grid Connect.

“From our first conversations, we had no doubts the people at Trek10 were very knowledgeable about machine learning and IoT, as well as the AWS approval process.”

Building the Machine Learning Component

The idea was to make the machine learning model a doctor of sorts. It would take data such as current fluctuations over time, and use that data to diagnose existing or potential failures in machinery.

While the cord was already built to gather lots of data on machine inputs and outputs, that raw data had to be sent in packets to the cloud on a regular basis and then converted into a format machine learning could interpret. Part of the challenge involved would be managing cost; the more frequently the data was analyzed, the more expensive it would be for the cord to run.

As the cord collects readings for volts, amps, temperature, and other critical data​​, it gets sent to SiteWise. Then, a Kinesis firehose sends it to a data lake, after which it is converted into a row and column format for the machine learning algorithm to ingest.

This process triggers the machine learning model that does anomaly detection as it reads through the data. If any anomaly is detected, users are notified through email or SMS via an SNS topic. By default, the algorithm collects, aggregates, and draws instances on the data once per day, though the frequency is configurable for end-users to help them control costs.

Another aspect that makes running machine learning over power cord data particularly tricky is that a power cord can be unplugged at any time. “It makes machine learning hard because it breaks the series data and confuses the algorithm,” said Fenil Patel, cloud engineer at Trek10.

Instead of using a service such as AWS Forecast, which is time-dependent, Trek10 opted to use the random cut forest method. This approach would allow them to configure the model so that an unplugged power cord didn’t read as a failure.

Getting the Cord to the Quick Start Phase

Before Grid Connect could submit the cord to AWS for approval, they needed to have a fully-documented quick start that worked in a user’s hands. This would require thorough testing of the cord in various industrial conditions, as well as making sure it was easy for users to connect it to their own AWS accounts.

Since in-house user testing wasn’t possible at this stage, Trek10’s engineers built a simulator that could account for both optimal working conditions as well as various failure modes—for instance, rising resistance caused by increased friction on a fan belt. Over the course of several weeks, the cord was put through a variety of scenarios and refined.

“It can be a lot of work to retrofit legacy manufacturing equipment,” said Gary Marrs, Solutions Architect at Grid Connect. “One of the things our smart cord does is let you plug in ‘dumb’ equipment and almost immediately start commanding it and monitoring it.”

As such, the quick start documentation also focused on helping users connect their own AWS account. This is advantageous because it enables users to keep all their data in-house, as well as completely control the frequency of data collection and see all the metrics in the SiteWise and QuickSight dashboards they are already using for other machines.

Once the end-user has connected their machines to AWS, they can have actionable data and a trained machine learning model within an hour. This not only makes it easier to address urgent issues with their new smart machines but makes everything on the factory floor easier to manage and more efficient to run. The cord can be used for daily operations such as monitoring power consumption, turning off equipment when it isn’t being used to save energy, and predicting maintenance needs.

Submitting the Quick Start to AWS

Once data is in the cloud, it becomes much more useful and actionable to end-users. Ultimately, the engineers at Grid Connect who designed the smart power cord wanted to give end-users the power to use that data to make their work much easier.

“This project had three parties involved: us, Trek10, and AWS,” said Grid Connect’s Sanders. “The people at Trek10 weren’t just excellent engineers, they helped us navigate a lot of red tape and push everything forward with AWS.”

Grid Connect is moving forward with a few clients who want to use their new power cord for predictive failure in order to continue testing and get the product ready for market.

“This smart cord has everything it needs to make a tremendous impact for our customers,” said Marrs. “With Trek10, we were able to do everything the way it needed to be done, as quickly as it needed to be done, to get it in users’ hands.”