Tolerancing a part with no original drawing
Every reverse engineering supplier quotes their scanning accuracy. Almost none of them explains the part of the job that actually decides whether the replacement works.
Measurement accuracy is not the hard part
Go and read the reverse engineering pages of a dozen UK suppliers and you will find the same number in different clothes: twenty microns, three microns, thirty-five microns. It is presented as the answer to the question “will the part be right?”
It is not, and it is worth being precise about why. Scanning accuracy tells you how faithfully the measurement reproduces the object in front of you. The object in front of you is a component that has been running for eleven years. It is worn where it rubs, corroded where it was damp, possibly bent, and quite possibly not to the original drawing even when it was new. Measuring it perfectly gives you a perfect record of all of that.
The engineering question is different: what was this part supposed to be? Answering it is judgement, not metrology, and it is where the value of the job actually sits.
Sort the dimensions into three groups
Before tolerancing anything, every dimension on the part gets assigned to one of three categories. This is the single most useful discipline in the whole exercise and it takes an hour.
Functional
Dimensions that determine whether the part works: bearing and bore fits, spigot registers, shaft diameters, sealing faces, thread sizes and classes, keyway widths, hole positions that have to line up with something else, and any dimension that sets a clearance or an interference. These get a tolerance derived from the duty — a fit class, a positional tolerance, a geometric control — and they are the dimensions worth arguing about.
Interface
Dimensions that must match the mating part but where the mating part, not the drawing, is the authority: bolt hole patterns, flange diameters, the envelope the component has to sit inside. Where you have the mating part, measure that and derive these from it. It is more reliable than measuring the worn item.
Incidental
Everything else — the majority of dimensions on most parts. Web thicknesses, cast radii, chamfers, overall lengths that nothing depends on. These get a general tolerance stated once in the title block and no more thought than that. Applying a tight tolerance to an incidental dimension is not caution; it is a cost you are adding for nothing, and it is the most common way a reverse-engineered part ends up more expensive than the original.
Recovering a nominal from a worn feature
When a functional surface has worn, the measured value is wrong by definition. There are several ways back to the number, and the good jobs use more than one and check them against each other.
- Measure the unworn part of the feature. Wear is rarely uniform. A journal worn in the loaded arc is often close to original ninety degrees round.
- Work back from the mating component. If the housing bore is 62 H7 and the design clearly intended a running fit, the shaft nominal follows.
- Look for standard sizes. Designers use bearing bores, stock bar diameters, standard threads and preferred numbers. A measurement of 24.87 on a bearing seat is telling you 25.
- Use the fit the assembly needs. Sometimes the only reliable route is to decide what fit the joint must achieve and dimension for it.
- Find the witness marks. Machining marks, a step at the edge of a wear scar, or an unworn shoulder often preserve the original surface exactly.
Whichever route is used, it goes on the record. A drawing that quietly states 25 h6 where the measured value was 24.87 is making an engineering judgement, and the person who has to live with the part is entitled to know that a judgement was made.
Copying a worn part gives you a replacement that is already part-way through its service life on the day it is fitted. It will fit — that is what makes it convincing — and it will fail early, and the failure will look like a material problem rather than a dimensional one.
Two more things that change the answer
You may not want the original tolerance
If the part failed early in service, reproducing the original specification reproduces the original failure. Sometimes the right answer is a harder material, a larger radius at the fracture site, or a tighter fit than the drawing would have carried. That is a design change and should be recorded as one, with a reason, rather than smuggled in.
Quantity changes the sensible approach
For one part, tolerance so that it can be made and fitted, and expect to fit it. For a batch that has to be interchangeable across three machines without selection, everything tightens, because you have removed the ability to fettle. Say which of these you are buying before the drawing is produced, because the drawing is different.
What good looks like on the finished drawing
- A general tolerance stated once, and the majority of dimensions relying on it.
- A small number of dimensions with specific tolerances, each traceable to a function.
- Datums that reflect how the part is actually located in service, not how it was convenient to measure.
- Material specification stating whether the grade was tested or inferred.
- A note recording which dimensions were recovered rather than measured, and on what basis.
That last line is the one almost nobody includes, and it is the one that will matter in five years when somebody orders another.