Weld camera arc glare is the reason most vision-based weld inspection systems fail in production, not cabling faults or vibration or sensor wear. A welding arc radiates broadband energy from ultraviolet through near-infrared at intensities that saturate every pixel on a standard sensor at once, leaving an overexposed white frame exactly where the weld pool detail needs to be. Solving this correctly — through spectral filtering, structured illumination, and camera placement that respects the physics of arc emission — is what separates a monitoring system that survives its first production shift from one that gets switched off within a week, which is the exact problem iFactory's weld inspection team works through with every deployment.
Four Reasons the Same Weld Camera Produces Different Images Every Shift
A weld inspection model trained on clean, well-lit sample images often performs poorly on the factory floor not because the model is wrong, but because the images it receives in production look nothing like the images it was trained on. Lighting inconsistency, more than model architecture, is the single largest driver of false positives and false negatives in deployed weld vision systems.
Arc Glare Saturation
The welding arc's broadband UV-to-infrared emission overwhelms a standard sensor, washing out the weld pool in an overexposed white frame during live arc conditions.
Reflective Surface Glare
Stainless steel and other high-reflectivity base materials bounce direct light straight back into the lens, creating hotspots that obscure the very defects the system is looking for.
Shift-to-Shift Drift
Ambient factory lighting changes with time of day, overhead fixture wear, and seasonal daylight, quietly shifting the baseline every trained model was calibrated against.
Fixturing Shadows
Clamps, jigs, and part geometry cast shadows across the weld bead that vary by part orientation, introducing noise that has nothing to do with weld quality at all.
The Illumination Stack, Built in Order
Solving weld lighting is rarely one technique working alone — production systems typically layer several complementary approaches, each addressing a different failure mode in the imaging chain.
Spectral Bandpass Filtering
A narrowband filter matched to a specific illumination wavelength isolates the weld pool signal from the arc's broadband background emission, with a filter full-width-half-max around 10 nanometers or tighter for reliable rejection.
Structured or Laser-Safe Illumination
An active light source paired to the bandpass filter provides enough signal to overcome the arc's brightness at the matched wavelength, restoring weld pool visibility that passive filtering alone cannot achieve.
Polarization Control
Polarizing filters reduce specular glare bouncing off reflective base metals, particularly stainless steel and aluminum, without dimming the diffuse detail the model actually needs to classify.
Multi-Angle Capture
Synchronized cameras at complementary angles fill in shadowed regions that any single viewpoint would miss, particularly around fixturing and clamps that vary by part geometry.
Matching the Technique to the Welding Process
Not every welding process fails in the same way, so the right lighting fix depends heavily on which process is being inspected and what its dominant glare source actually is.
| Welding Process | Dominant Imaging Challenge | Primary Fix |
|---|---|---|
| MIG / MAG Arc Welding | Broadband arc glare saturating the sensor | Optical bandpass filtering matched to illumination wavelength |
| Laser Welding | Coaxial beam path interference with pool imaging | Coaxial imaging synchronized to the beam path |
| Stainless Steel Joints | Specular reflection off high-reflectivity surfaces | Polarized illumination and lens filtering |
| Post-Weld Bead Inspection | Ambient lighting drift across shifts | Enclosed, controlled LED illumination housing |
A Practical Validation Checklist Before Going Live
Before trusting a weld vision system to run unattended in production, a short validation pass against real weld coupons catches most lighting problems before they show up as missed defects on the line.
Capture the same weld coupon under morning, midday, and end-of-shift ambient lighting to confirm the illumination stack holds image consistency across the full production day.
Run test coupons through every fixture orientation the process actually uses, not just the orientation used during initial camera setup.
Verify bandpass filter performance directly against live arc conditions rather than a bench test, since arc intensity in production often exceeds lab simulation.
Confirm image consistency holds after routine maintenance events like lens cleaning or filter replacement, which can shift calibration if not re-verified.
Choosing Sensors and Optics Around the Lighting Plan
Lighting design and camera selection are not two separate decisions — the wrong sensor can undo a well-engineered illumination stack, and the right one can make a modest lighting setup perform far better than expected. Global shutter sensors are generally preferred over rolling shutter for weld inspection because they capture the entire frame at a single instant, avoiding the smearing artifacts that a rolling shutter produces when imaging a moving weld pool or a part on a conveyor. Dynamic range matters just as much as resolution, since a sensor with poor dynamic range will either blow out the bright weld pool or lose detail in the surrounding shadow, no matter how many megapixels it has.
Global Shutter
Captures the full frame simultaneously, avoiding motion smear on moving weld pools or parts traveling through an inline inspection station.
High Dynamic Range
Preserves detail in both the bright weld pool and the darker surrounding base metal within the same frame, rather than forcing a tradeoff between the two.
Matched Lens Filtering
Lens-mounted bandpass and polarizing filters need to be selected to match the illumination wavelength and material reflectivity of the specific weld cell, not chosen generically.
What Inconsistent Images Actually Cost a Line
A weld inspection system that produces inconsistent images does not fail cleanly — it fails quietly, in the form of missed defects and false alarms that erode confidence in the system long before anyone traces the root cause back to lighting. The cost shows up in three distinct ways: escaped defects that reach the customer, false rejects that scrap good parts and slow the line, and the eventual decision to switch the system off entirely once operators stop trusting its calls. All three are avoidable with the same underlying fix — solving lighting before scaling the model, not after.
Frequently Asked Questions
Can existing plant lighting be used, or does every installation need dedicated illumination?
Ambient plant lighting is rarely sufficient on its own for consistent weld inspection because it drifts across shifts and seasons in ways a trained model was never calibrated to handle. Most reliable installations pair a dedicated, controlled illumination source with the camera housing so the inspection is insulated from whatever the rest of the plant's lighting happens to be doing that day.
How do you inspect a weld pool during active arc welding without the glare washing everything out?
The combination of a narrowband bandpass filter and a matched active illumination source lets the camera see past the arc's broadband emission at a specific wavelength, effectively filtering out the glare while preserving the weld pool signal. This laser-safe illumination approach is the standard solution for in-process MIG and MAG inspection specifically.
Does switching to a different base material require re-engineering the lighting setup?
Highly reflective materials like stainless steel typically need polarization added to the existing setup rather than a full redesign, since the core bandpass and illumination approach usually still holds. Contact ifactory Support to review your specific material mix before assuming a full lighting rebuild is required.
How long does it take to get a weld inspection system with proper lighting operational?
Basic implementations with a well-scoped illumination design are typically operational within four to eight weeks, though multi-angle or multi-process setups can take longer depending on fixture complexity. A scoping call early in the process is the fastest way to get an accurate timeline for your specific weld cell.
What is the single biggest mistake plants make when adding vision inspection to an existing weld cell?
Choosing the inspection model before the lighting is properly engineered is the most common and most costly sequencing mistake, since retraining a model repeatedly to compensate for inconsistent images wastes far more time than solving illumination correctly up front. Treating lighting as an equal engineering discipline to the model itself, rather than an afterthought, is what separates systems that survive production from ones that get quietly disabled.
