Modular production: designing the interfaces before the modules

The interface document is the real deliverable

Committing to modules means committing to frozen boundaries: mechanical envelope, mounting points, connectors, protocols, and the tolerance budget shared between whatever meets there. Those definitions need controlling with more rigour than the modules themselves, because a change on one side propagates to everything that touches it. Tolerance allocation is the specific discipline most often neglected — deciding how much variation each module may contribute so the stack at assembly still fits. Plants that skip it find out when two individually conforming modules refuse to mate on the bench. Ownership should sit with one named engineer per boundary rather than with whichever module team happens to be busiest.

Testing moves upstream, and so does attribution

A module testable on its own bench converts a whole class of assembly-line problems into pre-assembly ones. Faults become cheaper to find and easier to attribute, warranty analysis improves because a returned module carries its own history, and the final line spends its time joining known-good units. The trade is real investment in test equipment and test definition per module, plus the discipline to reject at the module gate rather than pass a marginal unit forward. Where modules are only exercised inside the finished product, the model delivers complexity without benefit.

Parallel cells and the balance between them

Each module can occupy its own cell with its own equipment, staffing and pace, which lets capacity be added where it is short instead of across the whole product. That is the clearest operational advantage and also the planning headache, since module cells rarely balance: one runs at a different cycle, another needs longer curing or test time. Growth means adding a cell rather than duplicating a factory, and it stops where the final assembly point saturates, because every module still has to converge there. Sequencing matters too: a module produced far ahead of assembly parks value on the floor while the cell that made it looks admirably efficient.

Stock held as functions rather than as parts

Inventory shifts from thousands of components at the assembly point to a smaller count of completed modules. Counting gets easier, kitting simpler and the assembly area calmer, but value per stored item rises sharply — a module carries accumulated labour, test time and components, so the same money buys much less cover. The exposure worth watching is obsolescence at the boundary: an interface change strips value from every module already built, which is why interface revisions need a planned cut-in point rather than an immediate switch. Serial or batch identity applied at module level is what makes such a cut-in traceable, and retrofitting that identity later is far harder than designing it in.

Suppliers who own a module own part of the architecture

Handing a supplier a functional specification instead of a drawing changes the relationship. They carry the design, the component choices and often the testing, which lowers your engineering load and raises your dependence: a module supplier with design responsibility is expensive to replace, since a second source has to redevelop rather than simply quote. Purchasing correspondingly buys performance and interface compliance rather than parts, with acceptance criteria written as tests. Product structures must hold module bills of materials, module-level revisions and rules governing which versions may be combined. Escrow of the design record is worth negotiating while the relationship is still new.

Frequently asked questions

How do we decide where the module boundaries should fall?

Put boundaries where the interface can stay simple and where the parts on either side change at different rates. A boundary crossing many connections, tight tolerances or a shared structural load is expensive to control; one carrying a couple of fixings and a defined connector is cheap. Splitting along lines of technology life helps too — electronics that refresh often should not be structurally entangled with a chassis intended to outlast several product generations.

Does modular design always cost more per unit?

Usually a little, yes. Connectors, housings, fasteners and test points that an integrated design would not need all add cost, and modules carry their own packaging and handling. The offset comes from parallel build, less assembly rework, faster fault isolation in service, reuse across variants, and the ability to change one part of the range without redesigning the rest. Where a product has a single variant and a short life, integration is often the better answer.

What stops interfaces drifting once production has started?

A change process with real teeth and a named owner for each interface. Every proposed modification should be assessed against both sides of the boundary, with backward compatibility stated explicitly and a cut-in point agreed for stock and modules already built. Publishing the interface set as controlled documents, kept separate from the module drawings, prevents them being edited quietly during a routine revision — which is how most drift begins.