Replacement parts for obsolete equipment: reverse engineering

A replacement part for obsolete equipment is a measurement problem before it is a manufacturing problem. The original maker has stopped shipping the machine, the drawing went out with the last service manual, and what you have is a used bracket with fifteen years of wear in two bores, a twist you can see against a straight edge, and a line that stops in three weeks.

The part in your hand is not the part that was drawn. Every dimension on it carries four contributions added together: the nominal dimension, the tolerance the original supplier held, the wear since commissioning, and the distortion the part picked up in service. Then the instrument you measure it with adds a fifth. A drawing that copies the measured number pastes all five into the new part, which is why reverse-engineered spares so often fit worse than the worn original they replace.

What follows is the method that separates those contributions: an error budget for the measurement, a bend deduction you can reproduce from a formed part, a material identity check that does not need a mill certificate, and a wear allowance rule with a number attached. The wider set of application-led fixes sits in the applications and cases hub.

What can a worn part actually tell you?

A worn part tells you its function, its interfaces and its clearances reliably. It tells you its original dimensions only at surfaces that never carried load. The first job is to sort the part's features into those two groups before touching a caliper.

Take a clevis bracket with a 12 mm pin bore. The measured bore is 12.06 mm. The nominal could have been 12.000 mm with a plus tolerance, or 12.050 mm with a minus tolerance; the bore cannot say which, and the difference is 0.05 mm of the clearance the joint needs to work. What the bore does tell you is that 0.06 mm of material has gone somewhere, and that the pin has worn to match it, because a pin and bore that run together wear into each other.

Sort the features three ways.

Preserved features. The first 1.5–2 mm at the mouth of a bore is usually unworn because the pin never reaches it. A machined pad that sat against a washer keeps its original thickness. A spot-faced land keeps its original diameter. These carry the design intent and are measured first.

Functional features. Centre distances, hole pitch, the height of the pin axis above the mounting face, the angle between two flanges. These are what the machine cares about, and on a distorted part they have to be re-established from three mutually perpendicular faces rather than from whichever face looks flat.

Free features. Flange heights, bend radii, corner radii, the outline. These were set by the tool, not by the function, and they are the safest to reproduce approximately.

The measurement budget, instrument by instrument

The values below are the ones that decide whether a written dimension is honest. They assume a part at shop temperature between 20 °C and 30 °C, measured by hand, with no coordinate measuring machine on site.

Two corrections turn a hand measurement into a dimension. The first is temperature. Steel expands at 11.7 µm per metre per kelvin and aluminium at 23.6. A 412 mm aluminium centre distance measured at 35 °C reads 0.146 mm longer than it would at 20 °C, which is half a typical tolerance band; the same steel feature moves 0.072 mm. Measure aluminium parts in a cool room or write the temperature on the sketch.

The second is instrument bias. A caliper with worn jaws reads small on outside measurements by 0.02–0.05 mm and this error is consistent, so it disappears from a ratio but not from an absolute dimension. Check the caliper against a gauge block or a known shaft before the session, and note the check.

Recovering the flat pattern from a formed part

The flat pattern is not on the part, and it cannot be measured directly. It has to be reconstructed from two outside lengths, the thickness and the inside radius, and the reconstruction has to state the K-factor it assumed so the shop can check the number before cutting metal.

Take a Z-section stiffener in 2 mm CRCA. Measured outside lengths are 80.0 mm and 60.0 mm, measured thickness is 1.98 mm, and the inside radius reads 2.2 mm on a radius gauge. The bend allowance for a 90° bend is:

BA = (π/2) × (R + K × t) = 1.5708 × (2.2 + 0.42 × 1.98) = 4.76 mm

The bend deduction is twice the outside setback minus the allowance:

BD = 2 × (R + t) − BA = 2 × 4.18 − 4.76 = 3.60 mm

So the developed length is 80.0 + 60.0 − 3.60 = 136.4 mm. Had the shop assumed a sharp punch radius of 2.0 mm and a K-factor of 0.33, the deduction would come out at 3.83 mm and the flat at 136.17 mm: 0.23 mm short on one bend, and 0.7 mm short on a part with three. On a 1,200 mm mating panel that 0.7 mm becomes the gap the fitter closes with a hammer.

Which of the two is right depends on the tool that made the original, and that tool is gone. The honest answer is to specify the radius and the K-factor you will use, publish the flat, and check the first article against the mating part rather than against the worn original. Ganesh returns the flat pattern with the first article for exactly this reason.

Reading a radius that has been deformed

An inside radius on a used part is rarely the radius the punch had. A 2 mm punch puts roughly 2.2–2.4 mm of inside radius into 2 mm CRCA in air bending, because the material wraps the punch with a little springback. If the part has been straightened in service, the radius at the bend is larger and the angle is not 90°, and both errors feed the flat pattern.

Check the angle with a protractor or a digital angle finder before trusting the radius. A bend at 87° instead of 90° changes the developed length by roughly 1.5 mm on a 60 mm flange, which is more than the radius uncertainty contributes.

Identifying the material without a mill certificate

You can establish thickness, hardness and steel family in ten minutes. You cannot establish the grade, and any drawing that claims a grade from a magnet test is guesswork dressed as data.

Steel family. A magnet sticks hard to CRCA, hot-rolled mild steel and 430 stainless, and weakly or not at all to annealed 304. Cold-worked 304 from a bend or a draw becomes slightly magnetic, so a weak pull proves nothing on its own; a strong pull plus visible rust in a scratch is enough to call it a ferritic or low-alloy steel.

Coating. A galvanised part carries 20 µm of zinc per side at the common coating class, which is 0.04 mm of total thickness that an ultrasonic gauge reads as steel unless you zero on the bare substrate. A painted part carries 60–80 µm of powder. Both matter: a replacement in CRCA where the original was galvanised will rust at the cut edges within a monsoon, and a replacement deliberately galvanised where the original was painted changes the bend behaviour.

Grade. Optical emission spectrometry on a 30 mm sample settles carbon, manganese and chromium in a few minutes at a laboratory. That test is worth paying for when the part is structural or welded to a pressure boundary, and it is wasted effort on a cover panel.

For most legacy parts the practical route is different and cheaper: specify the grade you intend to buy, quote it against the coil certificate, and confirm the first article by thickness and hardness. A DC01 or DC03 equivalent to IS 513 D grade covers the great majority of painted brackets, and a 304 covers the great majority of stainless covers. Saying so on the drawing is more useful than a guess at what the original mill rolled.

How much of the wear goes back on the drawing?

None of it, unless the wear is the function. The rule that keeps a replacement part interchangeable is a comparison between the measured wear and the tolerance band the feature needs.

Measure the feature, subtract the unworn reference where one exists, and compare the result with the band. A 12.06 mm bore against a 12.000–12.027 mm band has worn 0.033 mm beyond the top limit, so the replacement bore is specified at 12.020 mm, in the middle of the band. The worn 12.06 mm is not carried across. If the mating pin has also worn, from 12.000 mm to 11.940 mm, the assembly has 0.12 mm of diametral play, which on a 60 mm lever arm gives 0.11° of angular slack before the joint takes load. That slack is what the operator feels as a rattle, and it will be there again in the new part unless the pin is replaced with the bracket.

Three decisions follow.

• Wear under 25% of the band: reproduce the nominal dimension to the same tolerance class. The wear is normal service and the new part will outlast the old one by the same margin.
• Wear over 25% of the band: measure the mating part and size the replacement to restore the original clearance, not the original dimension. The pair, not the part, is the unit of replacement.
• Wear that is the failure: stop. A bore that has worn oval, a hole that has elongated along the load line, a crack from a hole to an edge — each is a design signal. Reproducing the geometry reproduces the failure, and the redesign belongs in the drawing you are about to issue, not in the next spare.

What Ganesh runs for legacy and replacement parts

Legacy parts go through the same route as new parts, which is the point of using a fabricator rather than a repair shop. Profiles are cut on the 12 kW fibre laser with its 40 ft × 16 ft bed, so a 3 m bracket that was originally a weldment can come out as one piece. Forming is on the 200 T, 5-axis CNC press brake with a 4,450 mm bed up to 10 mm, or on the mechanical presses from 20 T to 300 T where the original was a stamping and the volume justifies a tool.

The in-house tool room matters more here than on a new part. Press tools, welding fixtures and checking fixtures are built and maintained in-house, so a replacement for a part whose tool was scrapped can be produced without a second supplier in the chain.

Material range is 0.8 mm to 40 mm, with ±0.5 mm typical on formed sheet metal and tighter limits on critical features by agreement. First article inspection and PPAP documentation are available, and the drawing formats accepted are PDF, DXF, DWG and STEP. Customers including Indian Railways, Mahindra and TKE run legacy parts through this route; the documentation that travels with them is the same as for a current part.

What to send with a legacy part

• Photographs of all six sides with a scale rule in frame, plus one close-up of each worn feature and one of each marking, stamped number or label on the part.
• A measured sketch, not a modelled guess, listing every dimension with the instrument used and the room temperature.
• The wear direction marked on the sketch with an arrow, so the supplier knows which side of a bore was loaded.
• The mating part, or its measured dimensions, with the same treatment.
• The function in one sentence: what the part locates, what load it carries, and roughly how many cycles it sees in a shift.
• The failure history, if there is one: how many replacements per year, at what hours, and where the crack or wear started.
• The quantity now, and the annual quantity afterwards, because a part needed twice a year and a part needed twice a week are different manufacturing problems.

The sketch is the deliverable that decides the job. A supplier who receives a worn part and no sketch will measure the outside dimensions, draw what they see, and reproduce the distortion, because nothing in the package says which surface was datum. The ten drawing notes that prevent this are set out in drawing notes for sheet metal, and the route from a sketch to a first article follows build-to-print manufacturing.

When reproduction is the wrong answer

Three cases end the reverse-engineering argument.

The part failed in service. A crack from a hole to an edge, a weld toe that has been the start of every failure, a bore worn oval — the drawing you would produce reproduces the stress concentration. Redesign the detail, then reproduce.

The original process has gone. A drawn cup with a compound-curve form, a coining operation that needs a tool nobody has, a heat-treated alloy with no equivalent coil: each needs a new route, and the new route holds different tolerances. Say so before quoting, not after the first article.

The quantity does not carry the tooling. A laser-cut and press-braked replacement is the right answer at 50 pieces a year; at 5,000 a year the same part should be a stamping, and the tool cost has to be weighed against the piece price over the order horizon. That crossover is a quoting question, and the stamping capability page sets out what a progressive tool changes about tolerance and unit cost.

What we cannot do is invent the missing half of a drawing. Where the original tolerance class, datum scheme or material grade is unknown, the replacement carries our proposal as a proposal, marked on the drawing. The first article proves it against the machine and the drawing is revised once, with the revision recorded.

Send the part with the sketch and the mating dimensions to start a quotation, and we will return the features we can reproduce as measured, the ones that need a new datum, and the ones that should be redesigned before anyone cuts metal.