# AI visual quality inspection on the production line

Castings and machined parts needed consistent, traceable visual quality inspection on the line. Deep-learning inspection where every decision is linked to its images, model version and confidence score.

- **Sector**: Manufacturing
- **Region**: India
- **Disciplines**: Computer vision · MLOps
- **Practices**: [AI & machine learning](https://www.winapps.co.in/services/ai-ml/)
- **Client**: Under NDA. References available on a call.

## The challenge

A manufacturer in India needed visual quality inspection for castings and machined parts on its production line, with a clear record of why each part passed or failed.

## Our approach

We built deep-learning inspection in PyTorch, running at the edge on the line. Every decision is traceable: it is linked to its images, the model version that made it and its confidence score. An MLOps pipeline manages training and versioning of the models.

## Architecture

- **Production line**: Images of castings and machined parts
- **Edge**: Deep-learning model (PyTorch), Edge inference
- **Traceability**: Decision record, Images, Model version, Confidence score
- **MLOps**: Model training and versioning

Illustrative architecture, simplified to protect client confidentiality.

## Results

Outcome details for this engagement are shared under NDA. We're happy to walk through results and put you in touch with a reference on a call.

## Stack

- PyTorch
- Edge inference
- MLOps

Next case study: [Fleet management and route optimisation](https://www.winapps.co.in/work/fleet-route-optimisation/) ([Markdown](https://www.winapps.co.in/work/fleet-route-optimisation.md))

## Facing something similar?

A senior engineer, not a salesperson, replies within one business day. NDA available before any detailed discussion.

- [Book a 30-min intro call](https://www.winapps.co.in/contact/)
- Email us: [sales@winapps.co.in](mailto:sales@winapps.co.in)
- Or message us on [WhatsApp](https://wa.me/917015796830), or call [+91 70157 96830](tel:+917015796830).