The Pain Point: The First Article Was Perfect

A new radiator programme follows a familiar trajectory. The prototype core, hand-built and brazed in a laboratory furnace, meets its heat rejection target with margin. Tooling is released. Production begins, and the measured performance of the first production units is two to four percent below the prototype, with a wider spread across the batch.

Nobody has done anything wrong. The prototype was made under conditions that production cannot reproduce: one fixture, one operator, a small furnace with excellent temperature uniformity, and parts that were individually fitted. Production has several fixtures, three shifts, a large continuous furnace and parts made to drawing tolerance rather than to fit.

The gap between prototype and serial production in aluminium parts for radiators is almost entirely a gap in process control, and it shows up as scatter rather than as a single defect.

The Principle: Tolerance Stacks, Fixture Variation and Thermal Mass

Tolerance stack-up decides braze quality. The gap between fin and tube is not a specified dimension; it is the result of several tolerances combining. Fin height tolerance, tube flatness, tube position tolerance and fixture clamping all contribute. If the stack-up produces a gap outside the capillary range on some units, those units braze poorly while still passing inspection.

Thermal mass differs between prototype and production. A large furnace with a heavy fixture heats parts more slowly than a small laboratory furnace. The brazing profile that worked on the prototype may be too cold or too long on the production line.

Fixture variation creates unit-to-unit differences. Multiple fixtures, worn locating pins and inconsistent clamping pressure all shift the geometry. The result is a distribution of braze quality, not a single value.

Handling damage is cumulative and invisible. Thin fins deformed during assembly, transport or loading do not recover. The core may still pass a leak test and lose measurable performance.

Leak testing has a detection floor. A pressure decay test detects leaks above a threshold. Small leaks that are below the threshold at the time of test can grow in service. A helium test at a lower threshold catches more, but produces more false rejects unless the process is stable.

Volume changes the economics of every decision. At prototype volume, a part can be hand-fitted until it works. At ten thousand units a month, any operation that depends on fitting becomes a bottleneck and a source of defects. Processes that scale are those where the part locates itself - by a fixture feature, a formed tab or a brazed stop - rather than where an operator judges the result.

People and shifts are variables too. The prototype was built by the engineer who designed it. Production is built by three shifts with different habits - different clamping force, different stacking order, different judgement about whether a deformed fin is acceptable. Training reduces this variation but does not remove it, which is why the process must be designed to be insensitive to operator judgement wherever possible.

Equipment Classification: The Production Chain for Radiator Parts

Tube forming and extrusion. Roll-formed welded tube, folded tube or extruded multi-port tube. Dimensional consistency of the tube is the foundation of the whole core.

Fin forming. Roll forming of louvred or plain fin from clad coil. Fin height consistency determines the braze gap.

Core assembly and fixturing. Stacking, side plate placement and clamping. The most labour-sensitive step and the most common source of variation.

Brazing furnace. Continuous belt or batch vacuum. Profile control, atmosphere control and belt speed consistency determine joint quality.

End tank assembly. Crimping, welding or bolting, with gasket placement and torque control.

Testing. Pressure decay, helium leak detection, flow test, and destructive sectioning audit.

Machined and cast components. End covers, necks and brackets, where porosity and dimensional variation set the baseline.

Practical Guide: Closing the Gap

Build the prototype on production tooling wherever possible. A prototype made on a different process proves a concept, not a manufacturing route.

Define the critical characteristics, not the complete drawing. Braze gap, fin pitch, tube position and flange flatness matter far more than overall length. Put them on a control plan with real measurement.

Qualify the production furnace with a full thermal survey at three load conditions. Empty, partial and full load. The profile that matters is the one at full production load.

Use fixtures with positive, repeatable location. Spring-loaded locating features that compensate for tolerance are better than fixed pins that force parts into place.

Control handling from forming to furnace. Deformed fins are a permanent performance loss. Specify stacking limits, tray design and handling rules, and audit them.

Set a leak test threshold based on service risk, not on convenience. If a 5 × 10⁻⁵ mbar·L/s leak is unacceptable in service, do not accept it at the end of the line.

Run a first-article inspection that includes destructive sectioning. Dimensional first article alone does not prove braze quality. Section the first article at multiple locations and record fillet size.

Track capability, not conformance. A process running at Cpk 1.0 passes but will produce failures. Radiator joints need capability well above the minimum.

Audit the second and third shift as carefully as the first. Most process drift starts when the people who commissioned the line are not present. Unannounced audits at shift start are more informative than scheduled ones.

Freeze the process before you freeze the design. A drawing release with an unstable process is not a released product. Qualify the process, record its window, and treat any change to that window as a design change requiring revalidation.

Field Experience: The Two Percent That Cost a Programme

A supplier moved a passenger car radiator from a pilot line to a new plant. The pilot line had produced cores that consistently exceeded the heat rejection target by six percent. The new plant's cores met the target but with almost no margin, and a small number of vehicles in hot-climate markets showed elevated coolant temperatures under load.

The investigation compared cores from both plants sectioned at the same locations. The difference was fin pitch: the new plant's fin forming rolls had been set to the high side of tolerance, producing a slightly taller fin that compressed during stacking. Compression reduced the fin gap, which reduced airflow through the core, which reduced heat rejection.

Nothing was out of tolerance. The distribution had simply shifted to one edge of it. The fix was to centre the fin forming process on the nominal height and tighten the tolerance based on a capability study, plus a small adjustment to the stacking fixture to prevent compression.

The programme recovered its margin within one month, but only after the customer had demanded a plant-level performance audit. The supplier's own first-article inspection had passed, because the first article had been built by the most experienced operator on the day the line was commissioned.

Summary

Aluminium parts for radiators are produced by processes with many simultaneous variables, and the difference between a good programme and a difficult one is whether those variables are centred and controlled or merely within tolerance. The prototype proves the design; the control plan proves the process. Suppliers who treat the second as paperwork will spend the life of the programme explaining why the field performance does not match the sample.

Outlook

Production control in this sector is moving from inspection to instrumentation. In-line vision for fin pitch and core geometry, per-batch furnace profile logging, and automated helium leak testing with statistical process control are becoming normal rather than premium. The driver is not quality idealism but warranty cost: a field return on a brazed aluminium core costs an order of magnitude more than the part itself. Expect customers to require process capability data as a condition of nomination, and expect the suppliers who already have it to win the programmes.