EandM is an authorized SICK distributor serving manufacturers, system integrators, and industrial facilities across the West Coast — California, Oregon, and Washington. Among SICK's photoelectric sensors, safety sensors, safety scanners, vision sensors, encoders, and IO-Link devices, one of the most common questions we get is how to inspect for defects that don't follow a pattern.
Traditional machine vision works well when you can describe the defect in advance. A missing hole, an out-of-spec diameter, a code that won't scan — these are things you can count, measure, or match against a template. But a lot of real-world defects don't cooperate that way. They show up in random locations, in random shapes, on backgrounds that vary from part to part. Writing a rule for "anything that looks wrong" is nearly impossible, and that's exactly the gap AI anomaly detection is built to close.
Instead of programming a spec for what a defect looks like, the tool trains on images of good parts. It learns what normal looks like, then flags anything that deviates from it. You could also train bad parts, to further hone in on anomalies that should not pass through.
Where AI Anomaly Detection Outperforms Rule-Based Tools
Here's where that approach tends to outperform rule-based tools:
Surface defects on natural or textured materials
Wood, stone, and other organic surfaces have grain, color variation, and texture that changes from piece to piece. A pixel count or contrast check struggles here because the "normal" background is never quite the same twice. Anomaly detection handles this because it's not looking for a specific pixel pattern — it's looking for what doesn't match the learned baseline.
Cosmetic defects on produce
Bruising, blemishes, and discoloration on apples and other produce are a classic example. Every piece of fruit looks slightly different, so a fixed rule for "bruised" is hard to write and harder to maintain across varieties and seasons. Training on good fruit and flagging deviations is a more practical fit.
Weld inspection
Weld beads are rarely identical from one part to the next, even under good conditions. Porosity, spatter, and irregular bead geometry are difficult to catch with rule-based tools because there's no consistent shape or location to check against. Anomaly detection adapts to the normal variation in a good weld and still catches what falls outside it.
Packaging and print defects
Smudges, tears, and contamination on packaging are unpredictable in size and placement. A tool trained on clean packaging can flag irregularities without needing a rule for every possible flaw.
Textile and material surface inspection
Snags, stains, and weave irregularities behave the same way — inconsistent in shape and location, but easy for a trained model to flag as a departure from normal.
The pattern across all of these: if you can show a camera what "good" looks like more easily than you can describe what "bad" looks like, anomaly detection is probably the better tool for the job. It's also faster to get running than most teams expect — no vision programming background needed, just example images and a training step.
Frequently Asked Questions
No. The tool is built around teaching by example rather than programming rules. You capture images of good parts, train the model, and set a pass/fail threshold. No scripting or vision background required.
A working model can be trained on as few as 5 to 10 images of good parts. Bad images can be added but aren't required, since the model is learning what "normal" looks like rather than memorizing specific defects. The AI anomaly detection tool can read a max of 100 images on-device per tool instance.
Not necessarily, and it doesn't need to. Rule-based tools are still the better fit for known, measurable specs like dimensions or code reads. Anomaly detection is meant to cover the gap where defects are inconsistent in shape, size, or location.
The model can be retrained quickly by adding new example images, typically in well under a minute. It doesn't require starting over from scratch.
Yes — that's one of the strongest use cases. Because the model learns from real examples of the material itself, it adapts to natural texture and color variation instead of relying on a fixed rule that assumes every part looks the same.
EandM is an authorized SICK distributor serving manufacturers, system integrators, and industrial facilities across California, Oregon, and Washington. Reach out to EandM directly for current Inspector83x configurations, lead times, and pricing.
EandM is an authorized SICK distributor covering the West Coast, carrying SICK's photoelectric sensors, safety sensors, safety scanners, vision sensors, encoders, and IO-Link devices, including the Inspector83x line.
Yes. The Inspector83x ships with the Intelligent Inspection toolset available, which includes AI anomaly detection, AI classification, and AI pretrained OCR. EandM can help match the right sensor configuration — resolution, lens, and color or monochrome imaging — to your application.
EandM is the only SICK Elite Gold distributor on the West Coast — the highest recognition SICK awards to a distributor.
Dealing with inspection challenges that don't fit into a rule-based tool?
If you're dealing with inspection challenges that don't fit neatly into a rule-based tool, reach out to EandM. We can walk through your application and help you figure out whether AI anomaly detection is the right fit.