Every inspection technology has a physical limit built into it, and understanding where that limit sits is the difference between a quality program that catches defects and one that ships them. AI vision systems read light bouncing off a surface, which means they see everything a camera can resolve on the outside of a part and nothing that sits below the top layer of material. X-ray inspection reads radiation passing through a part, which means it sees density variations inside the material and only indirectly tells you anything about what the surface looks like. Neither is a replacement for the other, and choosing between them without understanding which class of defect each modality can physically detect is one of the most expensive misalignments a quality team can make — talk to iFactory's deployment engineering team about mapping your defect types to the right inspection modality.
AI Vision vs X-Ray Inspection: When Visual Detection Reaches Its Limits
Surface defects belong to AI vision. Internal defects need X-ray. The question isn't which one is better — it's which one is physically capable of seeing the defect class you actually need to catch, and where the two modalities meet on a production line to give you complete quality coverage.
Two Fundamentally Different Ways of Seeing a Part
The comparison between AI vision and X-ray inspection is often framed as a cost or speed decision, but the more useful framing starts with what each technology is physically capable of detecting. AI vision systems capture visible or near-infrared light reflected off a part's surface and use trained models to identify defects in that reflected signal. Anything the light doesn't reach — a void two millimeters beneath a casting's skin, a crack running through the center of a weld, a solder joint hidden under a ball grid array — is invisible to the camera regardless of how good the model is, because the information never enters the sensor in the first place.
X-ray inspection works on the opposite principle. A radiation source projects X-rays through the part, and a detector on the other side measures how much of that radiation is absorbed along each path. Dense material absorbs more; voids, cracks, and low-density inclusions absorb less. The resulting image is a density map of everything the radiation passed through, which means an internal void shows up as a bright spot on the detector image and a crack running perpendicular to the beam direction appears as a dark line. What X-ray cannot easily tell you is what the surface finish looks like, whether the paint has scratches, or whether a printed label has been applied straight, because surface-level cosmetic variations don't create meaningful density differences.
This physical divide is not a limitation vendors are working to eliminate — it's the fundamental property that makes each modality useful for the class of defect it's designed to catch. A quality program that treats the two as substitutes will always leave one class of defect uncovered, which is why serious inspection strategies pair them at different points on the line rather than choosing one.
Reflected Light vs Transmitted Radiation
The diagram makes the trade-off concrete: reflected light gives you a rich, high-resolution picture of every surface feature but tells you nothing about what's below the skin, while transmitted radiation gives you a density map of the entire volume but treats surface finish as noise. This is why a shop trying to catch both cosmetic scratches on an anodized housing and porosity in the aluminum underneath will always need both technologies — one modality cannot physically deliver what the other one specializes in.
Which Defects Each Technology Actually Catches
The clearest way to decide which modality belongs at which station on your line is to start from the defect list itself. Every part has a specific set of failure modes that matter for its function, and each of those failure modes has a natural home in one modality or the other. Trying to force a surface-level inspection technology to catch an internal defect, or vice versa, almost always ends in a compromise that misses both classes.
- Surface scratches, dents, and cosmetic blemishes on painted or coated finishes
- Missing components on printed circuit board assemblies visible from above
- Label placement, orientation, and print quality on packaging
- Color variation across dye lots and coating batches
- Bead geometry on welds — width, height, toe angle, leg length
- Dimensional gauging of external features against CAD reference
- Assembly presence and orientation verification at high line speeds
- Surface porosity and pit detection on machined or cast faces
- Internal porosity and voids in castings, forgings, and welds
- Solder joint quality under ball grid arrays and other hidden pads
- Foreign object detection in packaged food, pharmaceuticals, and sealed goods
- Fill level verification through opaque containers
- Internal cracks in welded pressure joints and structural members
- Wire bond and lead frame integrity in encapsulated electronics
- Delamination in multi-layer composites and laminated assemblies
- Density variation within sealed assemblies where surface access is impossible
A useful test when you're not sure which side of the table a specific defect falls on: ask whether the defect changes what light bouncing off the part's surface looks like, or whether it changes how much material a beam would have to pass through to reach the other side. Surface scratches change the reflection; internal voids change the density path. Defects that answer neither, or both, are the edge cases worth a conversation with an inspection engineer before you commit to a modality.
Not Every Defect Belongs to Vision, and Not Every Line Needs X-Ray
iFactory's deployment engineering team helps quality leaders map defect classes to the right inspection modality before capital gets committed — so you don't buy X-ray capacity for defects vision could catch faster, or oversize vision for problems that live below the surface.
The Comparison Every Buyer Needs Before a Purchase Decision
Purchase decisions get easier once the two technologies are compared on the same axes rather than in isolation. The table below covers the dimensions that actually drive cost and quality outcomes on a production line, not the specification-sheet numbers vendors optimize for on their own terms.
| Dimension | AI Vision Inspection | X-Ray Inspection |
|---|---|---|
| Defect visibility | Surface, geometry, color, presence | Internal voids, cracks, density variation |
| Inspection speed | Milliseconds per part, matches line takt | Seconds per part, often sampled not continuous |
| Coverage model | 100% inline on every part | Sampled or batched on critical parts |
| Capital cost range | Lower — camera, lighting, compute | Higher — source, shielded enclosure, detector |
| Operating cost | Minimal, no consumables | Tube replacement, radiation safety compliance |
| Facility requirements | Standard production floor | Radiation shielding, regulatory licensing |
| Operator training | Standard vision system operation | Radiation safety certification required |
| Software maturity | Trained deep learning models, active retraining | Established image analysis, slower model evolution |
| Best-fit stage | Inline, every-part, high-volume | Critical part verification, safety-related joints |
| Data output | Continuous images and measurements, part-linked | Radiographic images, often archived per part |
The row worth lingering on is coverage model. AI vision's ability to run continuously on every part at line speed changes what quality data looks like — instead of a sampled record with statistical inference filling the gaps, you get a defect record on every unit that shipped. X-ray's slower cycle typically means it runs on a critical subset, which is exactly right for defect classes where a single missed void could cause a field failure, but wrong as a general coverage strategy for high-volume production.
Why the "Which Is Cheaper" Question Is the Wrong Question
The instinct on any capital decision is to compare purchase prices, but inspection technologies don't produce their value at the purchase stage. Their value shows up in defects caught over the life of the equipment, and that value has to be measured against the class of defect each modality can physically detect — a cheaper technology that can't see the defect class you need to catch has an infinite cost per defect prevented, because it prevents zero of them.
How to Decide Which Modality Belongs at Each Station
A working decision framework doesn't start with the technology — it starts with the defect list from your last twelve months of customer returns, field failures, and internal scrap. Each entry on that list has a natural inspection modality, and mapping them one by one produces a coverage plan that treats vision and X-ray as complementary tools rather than competing budget line items.
Build an Inspection Strategy That Sees the Whole Part
iFactory's vision platform handles the surface, geometry, and assembly defects that belong to visual detection — and integrates cleanly with X-ray data so internal defect records live alongside surface records against the same serial number. That's what complete coverage actually looks like on a shop floor.
The Line Layouts Serious Quality Teams Actually Deploy
Once you accept that vision and X-ray solve different physical problems, the interesting question stops being which one to buy and becomes where each belongs on the line. The layouts that work in practice share a common logic: vision runs early and often, X-ray runs at the gates that matter most, and both feed a single quality record so the part's inspection history is coherent rather than fragmented across two disconnected systems.
The pattern is consistent across industries: vision handles the volume and the speed, X-ray handles the depth and the critical checkpoints, and the two together deliver a defect coverage that neither could deliver alone. Trying to stretch either modality to cover the other one's domain almost always costs more in missed defects than the second modality would have cost to deploy in the first place.
The mistake I see most often on capital reviews is teams treating this as an either-or between vision and X-ray, when the honest answer is that they solve different physical problems and a serious inspection strategy needs both. I've watched a plant spend eighteen months trying to tune a vision system to catch internal porosity on aluminum castings — it couldn't, because the porosity was two millimeters below the machined surface and the light never reached it. They eventually added X-ray at the final gate and the defect escape rate dropped inside a quarter. The other direction happens too, plants that put X-ray on every part when a 100% inline vision system would have caught the same defect classes at a tenth of the cost per unit. Match the modality to the physics of the defect, not to the sales narrative you heard last.
Frequently Asked Questions
See How iFactory's AI Vision Platform Fits Alongside Your X-Ray Strategy
Every serious quality program eventually pairs vision with X-ray, because no single modality can physically see every class of defect. iFactory's platform handles the surface, geometry, and assembly inspection at line speed on every part — and integrates cleanly with the X-ray data you already collect on critical joints.
