AWS Machine Monitoring

AWS Machine Monitoring

Although I no longer work in machining, I still follow news from the industry. This evening I was surprised to see an article on Ars Technica entitled Amazon to roll out tools to monitor factory workers and machines.

Is Amazon getting into the machine monitoring business?

Not in the sense of providing a CNC machine monitoring solution.

Amazon is introducing two new services for AWS (Amazon Web Services). These services, named Panorama and Monitron, use web-connected sensors installed in a facility to feed data back to an appliance or to the AWS cloud, respectively, so that ML (machine learning) can do its magic.

Keeping in mind that I have not touched either of these services, and that my knowledge of them comes entirely from the news articles, here are my thoughts.

Panorama

AWS Panorama is described as an IP62-rated appliance that connects to existing video cameras on a network and enables development of CV (computer vision) applications.

Although IP62 is not suitable for environments with liquid exposure (e.g. coolant spray), presumably the appliance is intended to live in a server room, or at least in a far corner of the machine shop, away from any liquid danger.

According to the FAQ, Amazon envisions the appliance being used to automate quality inspection, locate process bottlenecks, and identify safety violations such as PPE not being worn. The Ars Technica article speculates Panorama could verify social distancing rules are followed. The industry will undoubtedly come up with many uses for a CV solution that utilizes existing video feeds.

Panorama is described as an edge CV service, which means video data is processed locally, not streamed to the cloud. This should appeal to shops with low-bandwidth internet connectivity. Shops concerned about security should also be pleased with this. However, Panorama is a managed solution that does require internet connectivity.

Monitron

AWS Monitron uses sensors to monitor vibration and temperature, and is intended for fault detection / prediction. The vibration sensor is a 3-axis MEMS device. A gateway appliance relays the sensor data to the AWS cloud.

The Monitron starter kit includes five sensors and a gateway, and is claimed to require no training. Amazon suggests attaching the sensors to machinery with adhesive.

The sensors take samples once per hour, and operate over BLE (Bluetooth low energy). The batteries are not rechargeable or replaceable, but Amazon claims the sensors will run for three years before needing replacement.

Placing the gateway for a solid BLE signal from all sensors may be tricky in a machine shop with a lot of metal components (moving and stationary). There does not appear to be a wired option.

Once placed, the sensors stream vibration and temperature information through the gateway to the AWS cloud, where ML magic happens, and you receive a notification via mobile app whenever a potential fault is detected. This requires an internet connection to the gateway with high availability, although bandwidth requirements should be low.

What’s the Potential?

The appeal of AWS Panorama seems to be the basic appeal of CV: the idea that an automated system could monitor a process just as well as a human.

It’s this idea that led Elon Musk to invest in vision-only autopilot for Tesla, without LIDAR or other active sensor systems.

By the same logic, why shouldn’t a computing appliance, monitoring security camera video, be able to look out over a factory floor and realize that there’s a machine stoppage down the line that’s slowing production? Or that the CNC in cell two is operating with its safety door open?

It makes sense that a computing appliance could do this. On the other hand, Tesla’s autopilot is still far from being completely automated. Panorama will likely deal with more limited use cases in controlled environments, but it may rise or fall on the ease of use of its development kit.

AWS Monitron is what caught my attention at first. “Is Amazon deploying a CNC monitoring service?” I wondered as I envisioned an AWS service that hoovered up XML data from OPC-UA servers and MTConnect agents and then, through the magic of Big Data, created forecasts for machine failures.

Neat as that idea might be, it isn’t Monitron, which simply monitors vibration and temperature. I am sure there are facilities and machinery where those two measurements could be meaningful.

But I’ve been in pipe mills where the whole facility shakes as steel pipe rolls down the line. I’ve seen struggles to get solid measurements due to the constant vibration of the concrete slab floor beneath the gauge.

I’ve been in machine shops in the north where the air temperature plummets 20°F in a few seconds when the huge material doors are opened in winter. And in shops with a foundry on-site, where you can close your eyes and feel the heat of molten metal being poured nearby.

Monitron is probably better suited for tamer environments than those.

The fact that Amazon suggests attaching these sensors with adhesive also does not fill me with confidence, nor does the disposable nature of the sensors. A three-year (non-replaceable!) battery life does not seem impressive on machines that will see decades of service.

What I do like about Monitron is that it strives to be as unobtrusive as possible. From installation (“attach sensor with adhesive”) to usage (“the app will tell you when there’s a problem”), it seems the idea is to make the system easy for a non-technical user.’

So does it have a future? It really seems intended for facilities other than machine shops. (Amazon claims it has installed 1,000 Monitron sensors at a fulfillment facility to monitor conveyor belts. Useful, but not the same as monitoring a line or a cell in a machining facility.)

But if its sensor line could be expanded, or if it could integrate with existing sensor systems, the idea of an easy-to-install system that “just works” (tells you when something unusual happens, and learns what “unusual” means as time passes) could be very appealing.

Conclusion

Again, I should point out that I have not worked with either Panorama or Monitron. All I’ve done is read the news articles and the AWS overview and FAQ pages.

From what I’ve seen, it seems that Amazon does not (specifically) intend these services for CNC machine shops. Machine shops might find ways in which Panorama and its CV development kit is useful. Some shops might find the limited sensor feedback of Monitron useful, but its usefulness in a machining environment probably depends on expanding its capabilities.

Although as I write this, I’m reminded of Amara’s Law:

We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.

Roy Amara

If I sound dismissive of these services, it’s because in the short run, based on my past experience, these devices do not seem useful for the machining facilities I’ve seen.

And if I sound enthusiastic about their potential, it’s because in the long run, it’s because Matt Garman, AWS’s head of sales and marketing, sounds like he “gets it”. He’s quoted as saying:

There’s a ton of data in a factory, or manufacturing facility, or a supply chain. It’s just locked up in sensors, locked up in machines that a lot of companies could get a lot of value from.

Matt Garman

This is the same line of thinking that led to the development of MTConnect, and the idea that manufacturing data needs to be unlocked from its machines to be fully utilized is a philosophy I share and promoted for years.

Leave a Reply

Your email address will not be published.

%d bloggers like this: