Industry 4.0: Generalized Technologies Replace Purpose-Built Ones On The Factory Floor

A version of this originally appeared on Forbes.com.

Industry 4.0 is an incredibly broad concept than encompasses a lot of the technological advancements going on in the manufacturing space today – so it’s easy to get lost. One of the core opportunities of Industry 4.0 is the development and increasing availability of generalized solutions that in the past had to be purpose-built. These generalized solutions are inherently more scalable and are connected by default.

Generalization: Three examples

First, let’s start with the basics: generalization of tools. It used to be the only way to make 100 plastic parts was to kick off an aluminum tool and to use an injection molding machine. In the past decade, it’s been possible to get these parts from the numerous companies selling 3D printers or 3D printing services. What about 100 metal parts? Now instead of stamping or forging tools you can use CNC mills or metal 3D printers. In tooling, there’s been a definite shift from specialized tools to generalized ones that have led to reduced costs of prototyping and faster speed to market.

The second manufacturing generalization is of machines. The mechanization of manufacturing mostly started with tools – but then went on to include machines with the purpose either to do things humans could not or to obviate them. In the electronics space, as automation has dripped downstream from chip fabrication and tooling shops and has found fertile ground in circuit board surface mount assembly (SMA). Those lines are filled with single-function machines (paste deposition, pick and place, ovens) that are incredibly interoperable – and easily reprogrammable. This automation drove a significant increase in the yields of SMA to over 99% as an industry standard.

But automation met a hard stop at the doors to the final assembly workshop floor. Unlike SMA where most circuit boards follow the same fundamental rules, the problems on the workshop floor are as varied as the products they produce. That didn’t stop manufacturers from trying to automate the workshop. In the relentless pursuit of increased operating efficiency and headcount reduction, manufacturers who weren’t building automation in house partnered with systems integrators who capitalized on the opportunity to build customized equipment. Instead of investing in the development of general purpose machines that do basic operations (screw in screws, attach labels, combine part A and part B, etc), manufacturers pushed systems integrators for purpose-built equipment to attach a specific label or to combine two specific parts. While it enabled systems integrators to sell new equipment for every program, it has set the industry back as a whole: for decades automation was associated with high-cost, high-complexity solutions.

Things are starting to change. While purpose-built machines are still in vogue, multipurpose cells and omni-purpose robotic arms have picked up speed, growing at double-digit annual growth rates. Several companies are building robotic cells that are designed to be networked, multi-function, and massively reprogrammable. Only time will tell if the generalized hardware technology these companies produce will be able to displace the purpose-built systems the industry relies on today for less cost.

Our third example is the generalization of inspection – formerly done by humans, now often being replaced (when possible) with cameras and algorithms. While the camera itself is a highly generalized tool, its efficacy is entirely determined by the algorithms it runs to provide judgements on the line. For decades, it’s been possible to get a computer vision system that used classical computer vision methods to take measurements or to identify previously programmed defects. These are rules-based systems that are incredibly specific and require custom programming, often by a specialist. This has made them expensive. Sometimes the rules can be combined with knowledge to identify whether a specific cluster of voids in a turbine blade x-ray indicate a faulty part – but they are still based on rules and the ingenuity and limitations of the person writing the algorithm. This makes the algorithms short-sighted: there’s a lot of things they can miss.

Generalized and self-programming vision algorithms are coming online – that’s what Instrumental does. Our algorithms are powered by machine learning methods – which not only enables these systems to program themselves, but to identify defects that weren’t even anticipated. Our algorithms are able both to identify what to look for and how to measure it, with very little human supervision. As a result, Instrumental has faster bringup times than conventional vision and provides significantly more value because it can be used in many different ways in places where vision wasn’t well-suited before.