Manufacturing from measured reality

A drawing or a model is a statement of intent. It describes the article that should exist, together with the tolerances inside which a real article is still acceptable. The tolerances are the honest part: they are an admission, written into the definition itself, that the manufactured article will not match the nominal geometry.

Every real article differs from nominal. Stock varies. Material springs back and relaxes. Fixtures deflect under clamping load. Machines have their own geometric and thermal behaviour. Prior operations leave the part in a state the next operation inherits. None of this is failure. It is the normal condition of physical production.

What a fixed process assumes

A conventional program encodes a set of assumptions about all of that and then replays them. It assumes the stock is where the fixture says it is, that the datums are where the model says they are, and that the machine will behave as it did when the process was proven. When those assumptions hold, the result is excellent and extremely repeatable. That is precisely why the approach dominates in high-volume work.

When they do not hold, the program has no way to know. It executes the same motion into a different reality. The usual mitigations are to constrain the variation until the assumptions become true: tighter incoming material control, more rigid part-specific fixturing, more inspection after the fact. Each of those costs money for every new configuration.

Working from state instead of assumption

The alternative is to establish the relevant state before acting on it, and to make the decision to act conditional on what was found. In principle that means a system needs four things it can actually defend, not just assert.

• Measured state. Where the article actually is and what it actually looks like, to a known uncertainty. A measurement without an uncertainty is an opinion.
• Characterized capability. What this machine and this process can actually hold, established from observed behaviour rather than from a specification sheet.
• Approved operating limits. The allowables the finished article has to sit inside, which come from the design authority, not from the manufacturing system.
• A decision. Whether, given the first three, this operation can proceed on this article and still land inside the limits.

Metrology-assisted and adaptive manufacturing already do versions of this. Probing to locate a part before cutting is routine. Adaptive machining against a measured surface is established in repair and in castings and forgings work, where the incoming article is known to vary. The concept is not exotic.

Refusal is a feature

The part that receives less attention is what happens when the answer is no. A system that can only proceed will proceed, and a system that always proceeds is not really deciding anything.

Bounded action means the system has an explicit envelope and does not act outside it. If the measured state, the characterized capability, and the engineering limits do not agree that an operation can land inside the allowables, the correct behaviour is to stop and escalate to a human rather than to attempt it and inspect afterwards. A refusal is cheap. A scrapped forging, or worse a non-conforming article that passes inspection, is not.

This is also what makes verification meaningful rather than ceremonial. Confirming the result on the part closes the loop that measurement opened. Without it, adaptive behaviour is just a more sophisticated way of hoping.

What this does not solve

Working from measured reality does not remove quality obligations, and it should not be sold as though it does. First-article requirements, process qualification and customer acceptance exist because someone has to be accountable for the article, and that is discussed in a separate note on first article and manufacturing reuse.

It also introduces its own difficulties. Measurement takes time and adds its own uncertainty. Capability models can be wrong. Deciding what to measure, and how much uncertainty is tolerable before the decision is unsafe, is genuine engineering work rather than a configuration setting. These are the questions physical validation exists to answer, and they are why Syntrex treats the systems it is building as something to be tested rather than something already established.

Sources

• NIST: Model-Based Enterprise / model-based definition program
• NIST: Engineering metrology and dimensional measurement