A Lesson From Dishy McFlatface: AI and the Benefits of Proactive Customer Support

Last week, Dishy McFlatface (aka our Starlink terminal hardware) died on me while I was on an important presentation. It seems like SpaceX is so busy hurling numerous satellites into space that they forgot to do some fundamental engineering work on… a cable. 

The cable between Mr. McFlatface and the router is notoriously poorly designed. The internet is flooded with complaints of people losing their connection to space because of an earthly cable failure. SpaceX has a simple but potentially costly hardware problem: Some cables deteriorate in the field, some faster than others, and they need to figure out which one is dying.

To minimize the impact of unhappy customers on its revenue and reputation, SpaceX developed a novel system to mine the boring BER (bit error rate) telemetry of every cable in use. If trends indicate a particular cable is deteriorating, SpaceX waives the $130 plus tax and shipping fee and sends a replacement.

The app shows that a poor cable connection is the culprit of my rudely interrupted presentation:

The moral of the story: SpaceX handles a hardware problem by pushing software to the user terminals to enable analysis and reporting of cable signal integrity. They can then proactively or reactively replace defective cables based on insights from analyzing telemetry.

Here at Tanktwo, we talk a lot and often about the value of telemetry. The reams of numbers may be uninteresting and unparseable to us humans, but a well-trained AI can plow through them effortlessly to identify patterns. 

Allow me to indulge the aerospace nerd in me and use this example: AI played a critical role in discovering extrasolar planets (exoplanets for short). These planets outside our solar system are notoriously difficult to locate because they don’t give off light like stars. The first confirmed exoplanet was discovered only in 1992 by radio astronomers Wolszczan and Frail — proving that planets existed outside our galactic cul-de-sac.

Once astronomers knew what to look for, things picked up. Over thirty exoplanets were bagged by 2000. But even the most introverted stargazers eventually got bored of wading through all those numbers, and someone came up with the idea of training a robot to do the heavy lifting. 

In this context, the robot is a machine learning algorithm (a branch of AI). This number-crunching machine went on to add almost four hundred exoplanets to the list over the next decade. Since most AI algorithms create a positive feedback loop, they get progressively better at their job. Today, we’ve found over 5,500 exoplanets.

It’s not too much of a mental stretch to go from astronomy to rocket science. It’s no surprise that Starlink uses machine learning algorithms to mine Mr. McFlatface’s terminal telemetry essential for mitigating the pedestrian but potentially costly cable problem.

Of course, all of this magic relies on software. As we explained in our recent post on  “software-defined” everything, using software to adjust hardware behaviors helps make complex technical systems better, cheaper, and more reliable. 

That’s what we do here at Tanktwo with software-defined batteries (SDBs).

We make it possible to apply what Starlink does with deteriorating cables to deteriorating cells in battery packs. Our technology can proactively identify upcoming failures and create reports for our customers to act upon.

For example, equipment vendors can see the health of every cell in a battery pack and identify those about to go kaput. Since our Dycromax Architecture allows mixing cells of different ages (which is not feasible with conventional battery technology), a vendor can simply drop a replacement cell in the mail and instruct the customer to make the switch before any unpleasant surprise happens.

Plus, our battery AI algorithm learns as we feed more data into the system to enable the deployment of ever-improving software — allowing our customers to see further into the future and address issues in a targeted manner.

The result: Lower cost, reduced support tickets, and happier customers.

No one likes downtime, and everyone likes visibility. Our self-learning machine systems will take us one (and hopefully more) step towards a future where electrification isn’t only good for the environment but also makes a lot of business sense.