AWS digs into its new machine learning industrial products
Amazon Monitron and Lookout for Equipment use sensor data to help industrial customers predict when their machines will break. Amazon's GM of machine learning and AI explains.
In this Q&A, Vasi Philomin, general manager of machine learning and AI at Amazon, discusses why AWS is releasing these new machine learning industrial products now and explains how they work.
Can you go into detail on Amazon Monitron?
Vasi Philomin: Amazon Monitron is essentially an end-to-end condition monitoring system that uses machine learning to detect abnormal behavior in industrial machinery. Essentially, it helps our industrial customers to implement predictive maintenance and reduce unplanned downtime.
If you look into the industrial space, around 82% of companies over the last three years experienced some form of unplanned downtime, and 70% of them are not fully aware when their equipment is due for maintenance or upgrade or replacement.
The average cost of unplanned downtime is about $250,000 an hour, depending on the size of your company. This is an average across small, medium and large businesses. Obviously, the smaller you are, the costs will be lower, and the larger, your costs will be higher.
Amazon Monitron are sensors that industrial clients can use to help predict when their machines will break.
The Amazon Monitron sensor is the size of my little finger. It comes with epoxy and you can just glue it to any part of a machine where you want to monitor vibration and temperature, which are key to telling you if something bad is going to happen over time.
The most places where we are seeing them being used are in places where there are rotating parts, like pumps and bearings.
The sensors communicate with a slightly bigger gateway. A lot of sensors can talk to the gateway via Bluetooth Low Energy. The gateway device connects to your Wi-Fi network and brings all of the sensor data to the cloud. There's an app that goes with it, which you can download to your smartphone. Within 10 minutes, you can start seeing the vibration and temperature readings.
In the cloud, we apply machine learning so that we can tell you if the machines are behaving abnormally. It takes a few days of measurements to establish what normal behavior is, and then the user can use the app to provide feedback, whether maybe load change on a day is expected or unexpected, and the machine learning model learns from that.
AWS also launched Amazon Lookout for Equipment. Can you explain how that is different from Amazon Monitron?
[Amazon Monitron] helps our industrial customers to implement predictive maintenance and reduce unplanned downtime.
Vasi PhilominGeneral manager of machine learning and AI, Amazon
It's a typical AWS service, in the sense that it's a service in the cloud that you can integrate with your own applications. It's meant for people that already have a lot of the sensors set up. Think of it like the brain of Amazon Monitron, but now users have that brain separately and they can upload their own sensor data. It can understand correlations between all kinds of sensor measurements. It no longer has to be just vibration or temperature -- it could be RPM, it could be the flow pressure, it can be any kind of sensor data users may actually have.
Without a data scientist, but with some development effort, users can visualize their data and understand how their equipment is operating.
By development effort, I mean that you need people to know how to integrate the web service APIs. But you don't have to be a data scientist to do it.
Right now, Amazon Monitron sensors only read vibrations and temperatures. Are there plans for future iterations to read additional metrics?
At this point, I don't have any exact plans to share. But right now, we've got version one of a sensor, so you can definitely anticipate multiple versions of it with more kinds of modalities. It's all going to be dependent on how customers want to use it, and what they are talking to us about. As usual, we're going to be informed based on what they tell us.
