Machine Learning on IoT Devices Without Connectivity

Unavailable cellphone coverage pisses me off.

It’s a real issue when you want to collect data, send it to the cloud, apply machine learning and get the results back.

How to get around that? With a combined approach of analytics on devices and in the cloud.

Germany Land of No Cellphone Coverage

I’m always pushing that you need to collect data. You need to make sure you collect the right data to do analytics to drive business and create new services.

What really pisses me off is, when I don’t have cell phone coverage. You would think that in a country like Germany you always have cell phone coverage.

After all Germany is very high developed, high tech country of engineers. Right!?

No! Very often when you drive from A to B you don’t have cell phone coverage. When I drive from and to work I don’t have cell phone coverage at sometimes.

I’m not going crazy back roads. Even when you’re in the center of a town or mid sized city you don’t have connection. You go inside a building and the cell phone coverage disappears. Even if you are looking out a window. Internet surfing on a train? Forget it!

If I go from my local train station Würzburg to for instance Berlin, a lot of times during the journey there is no connection.

It drives me crazy.

The cell phone coverage in Germany is so thin that you basically cannot make sure that you have coverage even in populated areas.

But this issue of no connectivity is not only a problem for me.

Connectivity a Huge Issue For The Internet of Things

When you think about the Internet of Things connectivity is a huge issue. Building new services, creating new business models requires data and a connection.

Without cellphone coverage you cannot send send data to the cloud. You cannot make analytics.

What this means is that the analytics, the machine learning, the intelligence has to go away from a centralized cloud platforms approach.

Machine learning has to be implemented more into the actual devices. Otherwise you cannot use the function when the device is out of connection.

“Devices” like cars, trains, diggers your cellphone basically everything that is mobile needs to get smarter. The goal is to get intelligence into the digger, so you you are not dependent on the cloud.

What this means of course is then you have the problem of not getting the data. You want that to refine your business model, to create a better service, to to modify the machine learning.

This can be easily easily adjusted when you fit memory into the application and when it gets back to the home base it will transmit it to you.

Use-Case: Analytics in Container Ships

Container ships travel the world. Most of the time they are out of cellphone coverage.

What they have is satellite connection. The problem is, you cannot send a lot of data over satellite, because that is getting very very expensive.

Onboard analytics

What happens is that the that the ship does local analytics. Devices use sensors to collect data and apply machine learning to create on board analytics insight.

The deviecs also buffer the data while the ship is out of connection.

On board analytics allows companies to get real-time analytics insight. So, basically if something fails, or something is in danger of failing the onboard analytics already makes a recommendation.

That recommendation is a small amount of data and can be sent over the over the satellite communication back to the to the shipping company.

They can quickly order needed spare parts to the next harbor, while the ship is at the see. When the ship arrives the harbour there is already the part waiting to be installed and the problem to be solved.

Transmit data to the cloud to refine machine learning models

Back at a harbor the ship also sends out the collected data to the cloud over Wi-Fi or cell phone. The platform who is creating this analytic service gets the data that has been collected over the journey.

They can work on the machine learning models. Refine existing ones and create new ones to solve new problems.

Update features on the edge device with cloud models

Those new or refined models are then pushed to devices, back to the ship. Over the air updating function is the key.

This way the cloud gets more and mor data. The machine learning models get better and better. The local analytics on the ship also gets smarter.

I would really call that a win-win situation for everyone!

Shipping company and cloud analytics provider!