Case study · Manufacturing
Every part inspected, not just a sample
Manual inspection catches defects on a sample, and tired eyes miss some of those. This is how we added cameras and computer vision to a production line to inspect every part, flag defects in real time and learn from operators' decisions.
Project overview
Automated visual inspection with computer vision.
- Industry
- Manufacturing
- Client
- Confidential
- Solution
- Visual quality inspection
- Platform
- Edge AI + Azure ML
- Timeline
- 12–20 weeks
- Team
- 3 people
- Inspection
- 100% of parts
- Latency
- Real time at line speed
- Integration
- PLC reject signal
Expected business outcomes
What the client set out to achieve
Detection rates are validated on labelled images from the client's own line.
-
95%+
Defect detection on validated defect types
Measured on a held-out set of labelled images.
Target for defect types with enough examples
-
100%
Of parts inspected
Every part imaged at line speed instead of sampling.
Where the camera station covers the line
-
Lower
Customer returns for visual defects
Fewer defective parts leave the factory.
Measured over several months after go-live
Business challenges
Where the client started
Inspection relies on people and sampling.
Inspection
Operators inspect a sample of parts by eye.
Consistency
Results vary by shift and person.
Records
Defects noted on paper with little detail.
Customers
Some defects reach customers and come back as complaints.
Speed
Line speed limited by how fast people can check.
Proposed solution
What the build covers
Imaging quality matters as much as the model.
-
Imaging station
Cameras and lighting designed for the defects that matter.
-
Defect models
Models trained on labelled images of good and defective parts.
-
Edge inference
Inspection at line speed on an edge device next to the line.
-
Line integration
Reject signals to the PLC and results to the quality system.
-
Operator review
Operators confirm or correct uncertain cases, labelling new data.
-
Quality dashboard
Defect rates by type, shift and batch.
Project requirements
What the client asked us for
The request was automated inspection. Success starts with imaging: consistent lighting and angles make defects visible to the model.
Operators stay involved, reviewing uncertain cases, which keeps quality high and improves the model over time.
What the project had to deliver
- 01Every part inspected
- 02Real-time rejection
- 03Operators review uncertain cases
- 04Defect data for improvement
Key challenges & considerations
What made it harder, and what we did
-
Rare defect types with few examples
What we did
Examples of rare defects were collected deliberately and added to the training set over time.
-
Variable lighting and surface finishes
What we did
Controlled lighting and an enclosure were designed before any model was trained.
-
Line speed and space constraints
What we did
The cameras and the edge computer were chosen to fit the line speed and the space available.
-
Changes in product variants
What we did
New variants were added through a retraining process run with the quality team.
What we watched for
- Prove visibility before building models
- Invest in lighting
- Collect examples of rare defects
- Keep operators in the loop
Development process
Imaging, data, model, line
A feasibility test on real parts comes before any line work.
-
Feasibility
Sample parts imaged to confirm defects are visible.
-
Data collection
Images labelled with quality engineers.
-
Models
Models trained and validated on held-out images.
-
Line integration
Edge device and reject signal installed.
-
Operation
Operator review loop and monthly model updates.
Technology stack
Edge computer vision
Inference on the line, training in the cloud.
Vision
- Industrial cameras
- Controlled lighting
Machine learning
- PyTorch / ONNX models
- Azure Machine Learning
Edge
- Edge GPU device
- Low-latency inference
Integration
- PLC signals
- Quality system
Quality assurance & deployment
Measured against a test set before launch
AI quality was measured, not judged from a demo. A test set of real examples with agreed answers decided whether the system was ready, and the same set was re-run after every change.
- A test set of real examples with agreed answers
- Accuracy thresholds agreed with the business before launch
- Safety checks: refusals, permissions and prompt-injection tests
- A person reviewing low-confidence output
- Cost and response time measured at real volumes
Deployment
- 01Line integrationEdge device and reject signal installed.
- 02OperationOperator review loop and monthly model updates.
- 03Pilot, then staged rolloutA small group of real users first, with results reviewed weekly before widening access.
- 04Monitoring and warrantyAccuracy, drift, usage and cost tracked after launch, and six months of free bug fixing.
Final deliverables
Quality at line speed
Every part is inspected at line speed, defects are rejected automatically and operators review the uncertain cases.
Quality engineers finally have detailed defect data to fix root causes.
- 100% inspection
- Automatic rejection
- Operator review loop
- Defect analytics
Handed over to the client
- 01Production AI serviceDeployed in the client's own cloud tenant
- 02Source code and modelsIn a repository the client can access
- 03Test set and resultsThe evidence behind the launch decision
- 04MonitoringAccuracy, usage and cost
- 05Six-month warrantyFree bug fixing after delivery
Success metrics
How success is measured
| Measure | Outcome | How it is measured |
|---|---|---|
| Defect detection on validated defect types | 95%+ | Target for defect types with enough examples |
| Of parts inspected | 100% | Where the camera station covers the line |
| Customer returns for visual defects | Lower | Measured over several months after go-live |
Verified reviews
What clients say about our custom software work
Verified reviews from our clients on the same kind of work, published on Clutch.
-
Custom Software Development
Web-Based Coupon App Development
The work produced by the team was of exceptional quality.
Jon Thies
CTO, Model Rocket Feb 2022 – Mar 2023
-
Custom Software Development
Mobile & Web Platform Development & Design
Silver Scintilla was responsive, respectful, and enjoyable to work with, quickly addressing any questions or requests.
DeShawn Brown
CEO & Founder, Lithios Jun 2021 – Nov 2022
-
Custom Software Development
Web-Based HRM System Development
Silver Scintilla provided the best solution with scalable functionality, personalized features, and robust security.
Suheb Khan
Director, Collaborative Insight Technologies Jul 2020 – Jun 2022
Free discovery call
Have an idea? Let's turn it into AI-powered software.
Book a free discovery call with our experts. Share your idea and we will help you shape the scope, timeline and budget, under NDA.
- Free consultation
- NDA before we talk
- Transparent estimate