Historically there have been three drivers of productivity spikes in manufacturing:

The shift from every component being bespoke to interchangeability between parts, from manual labour to mechanised processes, and from skilled craft to automated decision-making.

Achieving any of these in isolation has driven large gains in productivity, but when all three are met together, productivity explodes.

Why does combining the three result in more than the sum of the parts?

Because while each advance individually improves productivity, combining interchangeability of parts with automatic, mechanised processes results in a scalable manufacturing process. You can just ‘copy and paste’ to expand. Ramping up production is almost as simple as buying another machine.

But right now, outside of mass production, these factors are not met together. There’s a missing piece holding manufacturers back.

This hints at a latent opportunity for a 10x productivity gain in most kinds of manufacturing.

 

Every item unique → Interchangeability of parts

It’s easy to take for granted that we can swap one bolt or part or lego block for any other, but the concept of interchangeable parts is barely 200 years old.

The same goes for the idea that one machine will produce the same components as its neighbours.

We no longer need armies of skilled ‘fitters’ to spend hours carefully filing, bending and pressing parts together till they fit.

 

Manual labour → mechanised processes

There was a time when the power of your machine was limited to the strength of your body, or your horse.

The advent of steam power, and then electricity, gave factories access to effectively unlimited power to run more and larger machinery.

 

Skilled craft → Automated decision-making

Manufacture of anything was once a skilled craft. There were no machines, so the manufacture of say your flint dagger relied solely on your personal skill.

The advent of machinery gave us a first level of automated decision-making – the ability to successfully produce the same part over and over, taking craftsmanship out of the process.

This, combined with steam power and the advent of interchangeability drove a trend of mass production that was born in the first industrial revolution and continues today.

A modern example would be interchangeable car bodies running down a mechanised and automated robotic assembly line, not a human in sight.

There’s no longer people standing behind the machines telling them what to do, but there are two problems with this image:

• This automation isn’t flexible. You can’t send a car down the line followed by a microwave.
• This automation is manually created. People spent thousands of hours programming decision-making logic into those robots for the specific purpose of building the specific car variants on this specific line.

This kind of automation is only available to mass producers for whom the high investment in first-time setup will be made back across tens of thousands of units.

Remember Tesla’s year of Production Hell building that Alien Dreadnought factory? Exactly the right idea, but as Tesla learned, automation is still super-hard.

 

So if manufacturing isn’t automated, how do things get made?

Across the 99% of manufacturing that is not mass production, skilled craft is alive and well. Expert humans are needed to run just about all types of machinery.

We can’t just give factories a 3D file and press go, because we have no automation of human decision making, and that is standing in the way of our productivity explosion.

This is why the next industrial revolution hasn’t hit manufacturing yet. It’s why we don’t have one-click manufacturing despite the hype (looking at you, 3D printing) – there’s a piece missing.

 

What’s the solution?

We need full automation of decision making in manufacturing.

That means turning 3D files into final parts and shipping them to customers without a person making a single decision in-between.

That’s not the same as having no humans present. It just means the humans, or robots, do what the system tells them to do and how it tells them to do it.

 

The catch is, this automation needs to be flexible – it must work on virtually all parts, even if its never seen them before.

Building flexible automation in e.g. fulfimlment isn’t that hard. Ultimately we’re locating items, putting them in boxes, and moving the boxes. There aren’t so many variables.

But if you want to be able to run a microwave down the production line after a car, that’s a whole different story.

 

What’s so hard about automating decision making in manufacturing?

Achieving this is so difficult that even accurately describing the scale of the challenge is problematic. My best attempt at boiling it down into a single sentence is:

Combinatorial explosions of possible solutions + non-idealised real-world physics & variability + finite computing power + slow feedback loops = very very hard.

Taking machining as an example, say we want to produce this part from a solid cube of metal.

One way of making this would be to sweep back and forth a long, low-diameter tool that fits everywhere, but this would be 1000x too slow and would miss some features like the threads.

Incidentally most 3D printing processes work in this way, and creating a path like this is computationally simple.

 

But we need to find a strategy that machines the part quickly, combining multiple tools, machining angles and changes to the cutting parameters (radial and axial depth of cut and rpm).

 

This problem is like the classic computer science problem “travelling salesman”, but multiplied.

Problems like this are called NP-Hard, and are classed as some of the hardest in the field of computational complexity. They have been proven to have no fast solution that produces the optimum result.

In the case of The Travelling Salesman, adding each new city makes the calculation take near-exponentially longer.

But for machining a part, our problem has many additional complexities compared to a conceptually simple problem like travelling salesman: It needs to be solved to an accuracy of <1/100th of a mm, in 3D, in varying materials, with varying machines, where each path between nodes could use a vast range of different tools and cutting parameters, and with a constantly changing field of obstacles to contend with as the stock material is gradually cut away.

A problem like this can’t be solved in an automated manner by any software built to date, at least outside of CloudNC.

The industry gets around this today by using the human mind to pick the tools, paths, cutting parameters etc. Computer aided manufacturing software is used to convert those decisions into machine-readable instructions known as G-code.

Humans are pretty good at coming up with a workable solution to machining a part, but these are sub-optimal, often contain mistakes and need iteration, and of course need a skilled human in the loop.

In machining this human is called a CAM Programmer. This is a super-skilled and quite rightly highly respected profession. Becoming a good one requires talent, creativity, and years of training. Time spent CAM programming a part varies from 30mins for super simple geometries, to 3+ months for the most complex.

 

So that’s machining. What about other manufacturing processes?

This combinatorial explosion problem is found in some form for most manufacturing processes, and answers the question of why we can’t run a microwave down our production line after a car.

Machines are automated. Programming them to do new things is not.

 

Why solve it? Why automate the programming of manufacturing equipment?

Achieving this automation of programming will unlock a scalable order of magnitude gain in productivity for most manufacturing process it is realised for, not to mention unlocking perfect quality, on-time performance, lead times of days, and all at a lower price point.

 

 

Taking precision machining as an example – the typical factory is small, its machines are cutting metal only 26% of the time, and this metal-cutting is slower than it could be. Why?

• 48% of machine shops have <5 employees and 98% have <100, because the complexity of running one scales up rapidly as machines are added, and the software needed to effectively automate management of high mix production has not been built. Their size is fundamentally limited by how much information can fit in people’s heads.

• Utilisation averages 26% as they A) rarely run 24/7 because it is challenging to find skilled labour to work weekends and through the night, and B) scheduling a high-mix factory is not only NP-Hard but ‘Garbage-In-Garbage-Out’ (guess your cycle time lengths or process steps wrong once and the whole schedule is invalidated).

• Production speeds (and so yields) are around 1/3rd of what is theoretically possible with the same equipment. Finding a fast solution to that 3D Travelling Salesman problem is well beyond the capability of the human mind.

A lack of machine programming automation is at the heart of all three of those problems, and solving it will enable factories to become rapidly replicable in a copy-paste manner, as the limiting factor of how much info can fit in people’s heads will have been removed.

Pricing/scheduling/design-for-manufacture/setup/quality inspection will be automated as well, enabling overall factory autonomy and near-perfect scheduling. This will drive a multiplication of machine output.

Lastly, near-optimal machining solutions will become possible, enabling a second multiplication of machine output.

 

It is these these advances multiplied together that will unlock that latent 10x productivity gain, and in a rapidly scalable manner. We will witness the beginnings of a productivity explosion unparalleled since the original industrial revolution, and the unlocking of a key factor essential to the achievement of a post-scarcity society.

 

The next industrial revolution isn’t here yet. Machines are automatic, but programming them is not.

Solving this is CloudNC’s mission. Learn more here…

• Share Post