Before they launch a high-profile gadget, consumer electronics companies usually fly armies of engineers to China, where they spend weeks in remote factories hammering out manufacturing problems that can turn hot new electronics into a hot mess.
div > div.group > p:first-child”>
Inside the industry, it’s called the design process, and it’s a slog. “Nobody builds it perfectly the first time,” Instrumental co-founder and former Apple engineer Anna Shedletsky said.
So Shedletsky founded Instrumental with an ex-Apple colleague, Sam Weiss, who had also spent months in China troubleshooting Apple products before they were officially launched. The 4-year-old start-up builds a system of cameras for assembly lines, then couples it with software that uses machine learning to detect defects before they go into production.
If Instrumental catches on, it could enable more companies and start-ups to produce high-quality gadgets in Asian factories without the resources of a consumer electronics giant.
“What the process means for consumers is that they are less likely to get a product that has an issue or has a failure of the unit,” Shedletsky said. “Look at the Samsung issues they’ve had over and over — issues that got shipped that should never got shipped.”
One of Instrumental’s top clients is Motorola, the Lenovo subsidiary, which has been using Instrumental’s cameras since early 2018 to make its smartphones.
“Instead of sending an army of engineers at an enormous cost, every engineer can view assembly data on every unit from their desks thousands of miles away,” Motorola senior manufacturing manager Kevin Zurawski said in an email.
Tiny defects that start on the manufacturing line — for example, a seam that is a few millimeters larger than its tolerance or a component that has decreased in quality — can end up causing major embarrassment for electronics companies when the devices are sold to the public.
High-profile manufacturing issues in recent years include Samsung’s Note 7, which had a battery issue that caused the phone to catch on fire, or Samsung’s Galaxy Fold, which has seen its release date pushed back indefinitely after early units broke. Even Apple recalled one of its older chargers earlier this year.
“Look at Amazon reviews — there are a lot of dead on arrival products, or 1-star review experiences. It impacts the company’s ability to sell products, and companies are more and more aware of the impact of a bad experience on social media,” Shedlesky said.
Fixes happen in what’s called the design verification testing phase. Basically, before an entire factory gears up to assemble phones, companies put a single manufacturing line together to make sure the process works and all the pieces fit together correctly.
That’s how Motorola used Instrumental’s camera technology on the manufacturing line. That camera takes a picture of each device that goes through the assembly process, and then compares the pictures to figure out if there are places where parts are missing, or assembled wrong, or have some other problem that stands out.
This is what one of Instrumental’s pictures looks like:
“Instrumental’s machine learning allows for unit-to-unit comparisons, build-to-build comparisons, immediate operator feedback when defects are detected, therefore reducing high dollar rework and component scrap costs,” Zurawski said.
“It saves money in several ways,” Shedlesky said. “The Motorola team believes that it accelerates what’s called ‘product maturity,’ or finding the issues earlier, and fixing them with real engineering fixes — not bubble gum and tape.”
Instrumental has raised $10 million from venture capitalists and has 26 employees in the U.S. and China. More funds may be needed as more companies start to embrace machine learning as a core tool to reduce the amount of human labor required in manufacturing.
One of Instrumental’s competitors, for example, is Landing AI, which was founded by deep learning pioneer Andrew Ng. It has a deal with Foxconn, Apple’s main contract manufacturer, according to a 2017 post.
“In developed economies, deeply integrating AI into manufacturing will also pave the way to power a new generation of products, devices and experiences,” Ng wrote in the post.
One possible stumbling block for these companies is the intense secrecy around manufacturing, especially in the electronics world.
It’s not just companies like Apple or Motorola trying to prevent leaks around their upcoming products, but also the fact that the actual manufacturing fixes discovered in the design phase are treated as valuable core intellectual property.
“Secrecy has made manufacturing a lagging industry in terms of applying new technologies, but there are ways to respect the secrecy and get the benefit of new technologies,” Shedlesky said.