Experience that connects software to production reality

Where we are now

The first steps are already rather more practical than the usual discussion about AI and manufacturing might suggest.

★ It is already possible to examine an Edgecam post processor, compare its configuration against the requirements documented for a particular CNC controller or machine, and produce a detailed change specification identifying what needs to change and where.

That is quite different from asking a general-purpose AI system to guess how a CNC machine works. The important part is giving the system access to the real manufacturing knowledge: CAM documentation, controller manuals, machine information, existing post processors and proven examples.

★ Large bodies of Edgecam, Code Wizard and related technical knowledge can already be organised so that the system can retrieve and cross-reference the information required for a particular problem.

This starts turning specialist manufacturing knowledge into something a computer can use rather than simply something a person can read.

From advising the programmer to doing some of the work

At present, most AI systems sit outside the CAM environment. You ask a question. They give you an answer. That boundary is beginning to disappear.

★★ Work is already moving toward systems that can observe what happens inside the CAM environment, understand the before-and-after state and build a record of successful programming decisions.

Once that is reliable, the role of the system changes. Instead of saying, “Here is how I think you should program this,” it can eventually say, “Here is the process I have created, here is why I selected it, and here is the evidence behind the decisions.”

This does not require AI to invent machining technology. CAM systems already have sophisticated machining strategies. Machine simulation already exists. Digital machine models already exist. Post processors already translate toolpaths into machine-specific NC code. Tooling and workholding can already be represented digitally.

The opportunity is to connect those pieces and increasingly automate the decisions between them.

Now take that idea one stage further

Imagine a manufacturing company having an online account that describes its real manufacturing capability. The machines are known. Their travels, spindle capabilities and axis configurations are known. The controllers are known. Approved post processors are associated with each machine. Available tooling and workholding are known. Machine rates and production preferences can also be known.

Then somebody uploads a solid model.

★★★ The system examines the component, chooses an appropriate machine and workholding arrangement, selects machining strategies and tooling, estimates cycle time, generates the CAM process, produces machine-specific NC code and verifies it through simulation.

The user does not have to understand every detail of CNC programming to begin the process. The manufacturing knowledge is built into the system.

And it need not stop at programming

Suppose the model is being uploaded during quotation rather than after the order has been received. The system may see something that a conventional estimating process misses.

★★★ It could recognise that a machine table has room for another two vices and calculate the effect of machining three components per cycle instead of one.

Rather than simply reporting a cycle time, it could explain that there is sufficient table capacity for two additional vices and calculate the predicted change in unit manufacturing cost.

Or it may determine that a component can be manufactured on one machine, but another machine eliminates two setups and materially reduces total production time.

Now the same intelligence that ultimately produces the CNC program is contributing to the quotation. A quotation can increasingly be based on an actual proposed manufacturing method rather than an estimate made before anybody has seriously considered how the component will be produced.

The factory itself becomes digitally understandable

Today, knowing what a particular machine shop can really manufacture usually requires somebody who understands its machines, people, tooling and experience. But all of that can increasingly be represented as data.

★★★ A manufacturing company could eventually have a digital description not simply of the machines it owns, but of the components it is genuinely capable of producing economically.

What if the component could find the factory?

Now imagine the person uploading the model is not the machine-shop owner. It is the designer or company that needs the component. They upload the solid model and effectively ask: Who can make this?

★★★ The system could compare the manufacturing requirements of the component with the known capabilities of participating manufacturers and identify suitable companies, machines and production methods.

One company may have the necessary machine but unsuitable workholding. Another may be ideal for ten components but uneconomic for ten thousand. Another may already have exactly the right machine configuration, tooling and available capacity.

Instead of sending the same drawing to ten companies so that ten people independently analyse it, the component's manufacturing requirements could be matched directly with suitable manufacturing capability.

In effect, the model begins looking for its factory.

From quotation to production

Follow that path far enough and the process starts looking remarkably simple from the customer's side: upload the model, specify material, quantity and delivery requirements, assess manufacturability, select the manufacturing method, calculate cost, select the manufacturer, generate the CAM process, produce the machine-specific NC program, verify it, and generate tooling, setup and operator information.

There should still be approval points and independent verification. There will always be unusual jobs requiring experienced people. But experienced programmers no longer need to spend all of their time repeatedly solving problems that have already been solved before.

Instead, their expertise can be captured. Solve a difficult problem once, validate the solution, and the system can potentially apply that knowledge again.

That may be the real answer to the skills shortage

Perhaps the future is not about replacing CNC programmers. Perhaps it is about giving every good programmer much greater leverage: capturing what they know, making proven knowledge reusable, automating repetitive decisions, and leaving skilled people to deal with the genuinely difficult and new problems.

Some of what I have described here is still well ahead of us. Some is under active development. And, perhaps surprisingly, some of it can already be done.

That is why I think the next few years in CAD/CAM and CNC manufacturing are going to be particularly interesting.