Reverse Engineering

How it works

1. Send photographs, a sketch, or scan data, or ask us about shipping the part itself.
2. We review what you have and send a free quote within 12 hours.
3. Our engineering team models the part, and stops to ask whenever the sample cannot answer a question.
4. A second engineer verifies the model, then the files reach you by email or secure download link.

Steps 1 and 3 are where reverse engineering goes wrong, and they fail for different reasons.

Step 1 fails when the input cannot support the output you need. A reverse engineering job is limited by what the source material physically contains, and photographs in particular contain far less than they appear to. Three photos of a cast housing will not tell anyone the wall thickness or the internal rib layout. Neither will thirty. If you need an NDA in place first, ask and we will sign one before any files move. The quote form takes files up to 5 MB, and anything larger is handled after first contact.

Step 3 fails when a worn sample gets treated as a dimensional authority. A part that has been in service is not the part that was designed. A shaft journal measuring 24.87 mm was probably a 25 mm nominal feature before it spent years in a bearing, and the difference between recording what you measured and recovering what was intended is the entire job. Ambiguity stops the job. Where the surrounding dimension chain leaves exactly one arithmetically possible value, that value gets closed without asking, because that is arithmetic, not assumption. A value that merely looks plausible is still a guess, and it comes back to you as a question. Open points arrive batched as one written list rather than a trickle of messages through the week.

Three ways in

Reverse engineering has three genuinely different starting points. They are not interchangeable, and picking the wrong one is the most common way these projects go sideways.

From photographs and sketches

Suited to: parts whose geometry is mostly prismatic and externally visible. Brackets, plates, levers, simple housings, weldment components. Also useful alongside any other route, because a photo of the part in its assembled context answers questions a dimension sheet cannot.

Cannot resolve: anything you cannot see. Internal cavities, wall thickness, blind features, and the far side of the part are invisible to a camera. Thread specifications cannot be read off an image. Neither can surface finish or material. A photograph also carries perspective distortion, so scaling a dimension off the image produces a number that looks real and is not.

You supply: photographs from several angles including straight-on views of each face, a dimensional reference in frame such as a scale rule, and critically, a set of measurements you have taken yourself with calipers. The photographs establish shape. Your measurements establish size. A photo set with no measurements is a shape study, not a reverse engineering input.

Cost and time: from $125. The cheapest route to start and frequently the slowest to finish. Every feature you did not measure becomes a question, and queries pause the clock.

From a physical sample sent to us

Suited to: prismatic and machined geometry, worn components where the geometry needs interpretation rather than transcription, parts with internal features accessible to measurement, and any case where you cannot reliably measure the part yourself. The value of this route is that our engineering team takes the measurements rather than relying on yours.

Cannot resolve: organic or freeform surfaces. The part is measured by our engineering team, and measurement describes a surface at discrete points rather than tracing it continuously, so a compound curve cannot be recovered this way. If that is your geometry, the scan-data route below is the only one that fits. A sample also cannot tell you what the part was before it wore, or which of its features were designed and which are artefacts of how it was made, which is covered further down this page.

You supply: the part, and any context you have about its function, material, and how it failed. If the part mates with something, send the mating component or a drawing of it. Shipping instructions and the destination are confirmed at quote stage. Samples are returned on request, dispatched within a few days of delivery, with return shipping at your cost.

Cost and time: from $625, the highest of the three, and the route with the most non-engineering overhead. Shipping takes as long as shipping takes, and if this is the only surviving example, you are without it for the duration.

From 3D scan data you already have

Suited to: customers who have already had a part scanned and hold the mesh, and to organic or freeform geometry of any kind. Scan data is the only input that captures a compound curve, so this is the route for cast, moulded, and sculpted shapes rather than the physical-sample route above. We accept STL, OBJ, and PLY, and rebuild them as solid CAD. For organic shapes that will not decompose into prismatic features, this runs as a scan-to-CAD surface modeling workflow.

Cannot resolve: the gap between a mesh and a solid, which is larger than the file extension suggests. A scan is a dense triangle shell describing a surface. A CAD model is a B-rep solid built from faces, edges, and features carrying dimensional intent. Converting between them automatically produces a solid with no features and every scanning artefact faithfully preserved as geometry. It is the same category of problem as automated PDF to STEP conversion: the data in the file is not the data you need.

You supply: the mesh, the scale units it was captured in, and any notes on scan coverage. Scans routinely have holes where the scanner could not see, and undercuts and deep bores are the usual casualties.

Cost and time: from $500. Faster than photographs when the scan is clean, because the shape questions are already answered. A scan with significant dropout can be slower than either other route, because it looks complete and is not.

Prices are floors. The final figure depends on complexity, source quality, and output requirements, and reverse engineering is quoted per project rather than hourly because the analysis phase varies so much with input quality. See pricing.

If what you actually have is a drawing rather than a part, none of these apply. See legacy drawing conversion.

What we deliver

Native CAD files with an editable feature tree — on SolidWorks, Inventor, and PTC Creo. SolidWorks as SLDPRT and SLDASM, Inventor as IPT and IAM, and PTC Creo. A hole is a hole feature with a diameter you can change, and that is standard on every project delivered in those three. AutoCAD is delivered as a 3D solid instead, because AutoCAD has no parametric feature history for a tree to live in. Imported STEP geometry arrives the same way, a featureless solid where the same change means surface editing. Both get called 3D models, which is why the difference is worth confirming with any vendor.

Neutral formats. STEP AP242 — geometry as standard. Full PMI, GD&T, and model-based definition available on request. Also STEP AP203 and AP214, IGES, Parasolid X_T, and STL where a mesh workflow is the destination.

2D drawings on request. Manufacturing-ready drawings can be generated alongside the 3D model, delivered as DWG, DXF, or PDF. Dimensioning follows ANSI Y14.5 or ISO 1101, with GD&T per ASME Y14.5.

Assembly structure decided at quote stage. Either a mated assembly or individual parts. Assemblies run up to 1000+ parts.

Two-tier QC. Modeling by a specialist, verification by a second engineer, before anything reaches you.

One revision round is included per project, with additional revisions billed at the rate specified in your quote. Revisions are tracked in writing and you confirm acceptance. Errors caused by our team are corrected at no charge within 30 days of delivery, separately from the included revision.

When this gets difficult

The modeling is rarely the hard part. Deciding what the part is supposed to be is the hard part, and a physical sample is a worse witness than most people expect.

Wear has removed the original dimension. Bearing journals, bushing bores, gear teeth, sealing faces, and any sliding surface are smaller or rougher than they were designed to be, and the sample carries no record of the difference. Rounding up to the nearest standard size is the obvious move and it is a guess wearing a convincing costume. The part may have been undersized by design. Where the surrounding chain forces exactly one possible value, we close it. Where it does not, you get the measured value and the candidate interpretations, and you decide.

Design intent versus manufacturing artefact. A parting line, an ejector pin mark, a gate scar, a machining witness step where two setups met: none of these are features of the part, and all of them are present on the sample as real geometry. Model them and you have baked one factory's tooling into the definition of the component. Ignore the wrong one and you have deleted a functional feature. The distinction usually depends on what the part does, which is why we ask about function rather than guessing from shape.

Draft angles and cast fillets. Cast and moulded parts carry draft on nearly every vertical face, and the draft belongs to the tool rather than the design. Whether it should be reproduced depends on how the replacement will be made. A part being re-cast wants the draft. A part being machined from billet usually does not, and carrying it through means every face sits a degree or two off perpendicular for no reason. Cast fillet radii also vary along a single edge on real parts.

Features that are nearly symmetric. A sample will measure 40.1 mm on one side and 39.9 mm on the other, and the question is whether that is a tolerance band around a symmetric design or an intentional asymmetry that matters. Modeling it as measured produces a part that is subtly wrong everywhere. Modeling it as symmetric produces a part that is wrong in one place that might matter a great deal. This is a question every time.

The sample was made from a worn tool. The part in your hand may be a faithful copy of a die in its last year of service. Reverse engineering it reproduces the die's condition rather than the component's design. Where multiple samples of the same part exist, sending more than one materially improves the result, because features that agree across samples are design and features that vary are not.

What this costs you

Queries pause the clock. A worn casting with an unclear parting line and two ambiguous bores will take longer than a clean prismatic bracket, and if a question sits in your inbox for three days, delivery moves by three days. Batching the questions into one written list limits the number of interruptions but does not remove the wait.

A vendor who never asks anything will always look faster. The model arrives on schedule with every ambiguity resolved by someone who has not seen the assembly and will not tell you which choices were made. That cost is invisible until a replacement part reaches a fitter.

Who this is for

The part is obsolete and the supplier is gone. It still runs in your equipment, nobody sells it, and the only definition that exists is the one sitting in the machine.

You have a part and no drawings at all. Nothing in the archive, nothing from the original manufacturer, no digital record anywhere.

The only surviving example is worn or damaged. It came out of service because it failed, and it is simultaneously your only reference and a poor one.

You need to modify an existing component. Changing a mounting pattern or a port position requires editable features, and a mesh will not give you one.

You are re-sourcing and the shops want 3D. Machine shops and casting houses increasingly decline to quote from a physical sample or a sketch.

Work spans industrial machinery, automotive, medical devices, aerospace, architectural hardware, and furniture.

If what you have is drawings rather than parts, that is legacy drawing conversion instead. Different inputs, different process, and usually cheaper.

FAQ

What accuracy can you hold?

We do not publish an accuracy figure, because a single number would be misleading. Achievable accuracy is set by the input, not by the modeling. A clean scan of an unworn machined part supports a much tighter result than a photo set of a worn casting, and no amount of care at the CAD stage recovers information the source never contained. What we will tell you at quote stage is which features we expect to resolve confidently and which ones will need your input.

Do we have to send you our only sample?

No, but understand the tradeoff. Photographs plus your own caliper measurements work for prismatic parts, start at $125, and cost you nothing in downtime. Working from the sample itself is materially better for worn parts, for internal features, and wherever measuring it yourself is impractical, and starts at $625. Our engineering team takes the measurements instead of you. Organic and freeform shapes are the exception: those need scan data, not a sample. Shipping instructions and the destination are confirmed at quote stage, before you send anything, and samples are returned on request within a few days of delivery, with return shipping at your cost. If the part is irreplaceable and currently in service, say so and we will tell you honestly whether the photo route is viable for that specific part.

What if the part is worn?

Wear is expected and it is the reason this work needs a human. A worn sample records its current state, not its original dimensions, so recovering the design means reasoning from the surrounding dimension chain, from mating components, and from what the part does. Where the chain leaves exactly one possible value, we close it. Where it leaves several, you get the measurement and the options rather than a silent decision. Sending a second sample of the same part helps considerably.

Can you work from photographs alone?

For simple external geometry, sometimes. For anything with internal features, wall thickness, or threads, no, and a vendor who says otherwise is planning to guess. Photographs establish shape and proportion, not size, because perspective distortion makes scaling off an image unreliable. Photographs plus caliper measurements you take yourself is a legitimate input and covers a large share of bracket, plate, and lever work.

We already have an STL from a scanning service. Is that enough?

It is a strong start and it is not a CAD model. An STL is a triangle mesh describing a surface with no features and no editable dimensions. Rebuilding it as a solid is modeling work rather than a format conversion, and mesh to solid CAD conversion is exactly what this route does. Send the mesh along with the units it was captured in and any notes on where the scan has dropout, since undercuts and deep bores are the usual gaps.

Who owns the resulting model?

You do. Upon full payment you receive full ownership of the deliverables, and we retain no rights in them. Source files are deleted from our systems within 90 days of project completion, and our production partner retains files under its own policy, typically no longer than twelve months. Uploaded files and delivered models are not used to train AI models, and partners are not authorized to do so on our behalf. An NDA is available on request and is signed before any files are shared.

How long does it take?

Simple single-part work returns in 24 to 48 hours. Medium complexity, meaning a multi-feature part or a small assembly, runs 2 to 3 business days. Complex multi-part assemblies run 4 to 6 business days. Rush turnaround of 12 to 24 hours is available for files received before 11am, subject to availability. These ranges assume source material we can work from without coming back to you, since every query pauses the clock.