Assembly automation: tolerance stacks, part supply and designing the product for it

Tolerances that people absorb, machines cannot

An assembly is a chain of features, each with a permitted variation, and the accumulated result at the final joint can be considerably wider than any individual tolerance suggests. Manual assembly hides this because the operator adjusts. Automated insertion into a hole that has drifted to one extreme while the pin has drifted to the other simply fails, and it fails on a fraction of parts rather than consistently, which makes it maddening to diagnose. Analyse the stack before designing the cell, decide which features must be tightened, and consider a self-locating design where one part guides the other rather than relying on absolute position.

Design changes that make automation feasible

Small product changes are frequently worth more than any amount of cell engineering. Adding a lead-in chamfer turns a marginal insertion into a reliable one. Making a part asymmetric enough to have only one stable orientation removes an entire feeding problem. Replacing a fiddly clip with a snap feature approachable from a single direction removes a wrist rotation. Designing so the assembly builds up along one axis from a base part avoids reorienting the stack midway. These decisions cost almost nothing during design and are difficult afterwards, which is why manufacturing engineering has to be in the room early.

Compliance and force feedback for the parts that resist

Rigid machines forcing rigid parts together break things. A compliance device at the wrist allows small lateral and angular corrections during insertion, which handles ordinary positional error mechanically and cheaply. Where the fit is tight or the mating features are hidden, force sensing lets the machine search, detect contact and confirm that a component has actually seated rather than jammed part way. Both are cheaper than tightening every upstream tolerance. What matters equally is what the cell does when insertion fails: a defined retry, then a controlled reject, rather than pressing until something deforms.

Verify each step rather than inspecting the finished assembly

Once a component is buried inside an assembly, confirming it is present and correct is expensive or impossible, and discovering the omission at final test means dismantling. Automated assembly should confirm as it goes: the part was picked, it was placed, it seated to the expected depth, the fastener reached its target, the connector latched. Each check is inexpensive at the moment it is made and gives an unambiguous location for the fault. It also produces a build record per unit, which is the only practical way to bound a containment action when a defective batch of components is later identified.

Flexibility costs rate, and the choice is a commercial one

A dedicated station built for one product achieves the highest rate and the lowest unit cost, and becomes scrap when the product changes. A flexible cell handling a family runs more slowly, costs more per station and survives the product transition. The decision belongs with whoever knows the product roadmap, not solely with engineering. A common compromise automates the operations that are stable across the family — the base part handling, the fastening, the verification — while leaving genuinely variant-specific steps manual or on quick-change tooling, so the investment is not tied to one configuration.

Frequently asked questions

How early should manufacturing engineering see a new product design?

Early enough to influence part geometry, orientation features and assembly sequence, which means during concept rather than at design release. The changes that make automated assembly feasible are trivial to make on a drawing and expensive to make once tooling exists. A short review focused on how each part will be fed, gripped, located and verified typically produces a handful of requests that cost the designer very little and remove entire subsystems from the eventual cell.

What should happen when an automated insertion fails?

A defined sequence rather than repeated force. Retry a limited number of times with a small positional adjustment, then stop and route the assembly to a reject or manual rework position with a record of which step failed. Pressing until the part goes in produces damage that may not be visible and may pass inspection. Recording the failing step is what converts a nuisance into diagnostic information, and clusters of failures on one step usually point to a supplier or tooling change upstream.

Should we automate the whole assembly or only part of it?

Partial automation is often the better economic answer, particularly where a few operations are difficult and the rest are straightforward. Automate the repetitive, ergonomically poor and quality-critical steps, and keep human judgement for variant-specific work, awkward flexible components and final verification. The main design requirement is a clean interface between automated and manual sections so that work in progress can pass between them without the whole line stopping when one side has a problem.