---
lang: en
title: "AI defect detection and automatic rejection"
url: https://nabrio.com/use-cases/defect-detection-and-rejection
canonical: https://nabrio.com/use-cases/defect-detection-and-rejection
updated: 2026-10-27
publisher: Nabrio Co., Ltd. (nabrio.com)
---

# AI defect detection and automatic rejection

Nara finds defects such as cracks, chips, scratches, dents, discolouration, wet spots and missing seals, and signals your PLC or rejector to remove the item from the line. An image of every defect is saved for review and for retraining the model.

- Sectors: [Manufacturing](https://nabrio.com/industries/manufacturing.md)
- What it takes: Trained on your items. We train a model on your own products or defects in Reva, and Nara runs it on the line.
- Products: [Nara](https://nabrio.com/nara.md), [Reva](https://nabrio.com/reva.md)

## The problem

Defects must be removed from the line automatically, at line speed, without stopping production.

## How it works

- **Input:** Camera and part sensor on the line
- **Process:** Defect model trained in Reva on your own parts
- **Action:** Signal the PLC or rejector, save the defect image, update the defect-rate dashboard

A model trained in Reva on images of your good and defective parts runs in a Nara flow. When it finds a defect, Nara signals the PLC or rejector over Modbus or TCP/IP to remove the item, saves the image, and updates the defect rate on the dashboard. Saved images go back into Reva to improve the model over time.

## What you need

- A camera and part sensor at the inspection point
- A PLC or rejector reachable over Modbus or TCP/IP
- Images of good and defective parts to train a model in Reva

## Nara building blocks

- Camera nodes
- Custom defect model (Reva)
- Modbus and TCP/IP to PLC
- Image export
- Database table
- Dashboard with defect rate

## What it improves

- Defect escape rate
- Reject accuracy (fewer false rejects)
- Throughput kept at line speed

## Where it runs

- [Wet rubber clumps rejected before they move downstream](https://nabrio.com/customers/rubber-clump-rejection.md) (A rubber products manufacturer)
- [AI vision added to an existing conveyor](https://nabrio.com/customers/conveyor-sorting-retrofit.md) (A conveyor system operator)

## Frequently asked questions

### What kinds of defects can it find?

Cracks, chips, scratches, dents, discolouration, wet spots, missing seals and other visible defects, once a model has been trained in Reva on images of your parts.

### How does the model improve over time?

Nara saves an image of every defect. Those images go back into Reva, where the model is retrained and redeployed to Nara as a Nabrio Model (NAM) file.

## Related use cases

- [Visual quality inspection](https://nabrio.com/use-cases/visual-quality-inspection.md)
- [Foreign object detection](https://nabrio.com/use-cases/foreign-object-detection.md)
- [Assembly verification](https://nabrio.com/use-cases/assembly-verification.md)

Talk to us about defect detection and rejection: sales@nabrio.com. [Download Nara](https://nabrio.com/download.md).