Zouhour Bellamine

Hello

English 🇬🇧

✿
Loading
← All work

04 / Computer vision / Industrial inspection

Intelligent Visual Inspection for Laser-Engraved Components

An AI-powered computer vision workflow designed to automatically identify visual defects after laser engraving.

Computer VisionDeep LearningIndustrial InspectionConfidential Project
x: 112 / y: 064scan / verifyCV

Professional context

Valeo · Enfidha, Tunisia

Developed in an industrial automotive environment focused on interior components and manufacturing processes including laser etching.

Learn about Valeo

The problem

Identical-looking component positions→Button interchange→Laser engraving→Manual inspection→Potential defect

Different control buttons can be inserted into visually identical locations. After laser engraving, incorrectly positioned or unengraved buttons may only become apparent during verification, adding manual work and quality risk.

Manual verificationLonger cycle timeHigher scrap riskQuality risk

The AI objective

Can computer vision verify the component automatically?

01

Incorrectly positioned

Detect buttons that do not correspond to the expected configuration.

02

Unengraved

Detect buttons where the expected engraving is missing.

Inspection pipeline

A visual path from component image to defect identification.

01

Component image

A component image enters the inspection workflow.

02

Pictogram detection

Relevant visual information is located for analysis.

03

Button detection

Control-button positions are identified in the image.

04

Configuration verification

The observed configuration is compared with what is expected.

05

Defect identification

Potential positioning or engraving defects are surfaced for review.

CRISP-DM methodology

Iteration is part of the inspection process.

The project evolved as the available data and inspection requirements changed. CRISP-DM offered a disciplined way to move from data through preparation, modeling, evaluation, and back into iteration.

Data→Preparation→Modeling→Evaluation→Iteration↺

My technical contribution

Turning inspection requirements into an AI workflow.

01

Data

Image preparation and dataset organization

02

Annotation

Preparing labeled data for model development

03

Detection

A YOLOv8-based computer-vision workflow

04

Image processing

OpenCV-based processing

05

Model work

PyTorch-based development and evaluation

06

Integration

Flask backend integration

07

Iteration

Evaluating and improving the workflow based on results

Engineering challenge

From visual inspection requirements to a reliable AI workflow.

The challenge was to translate a manufacturing quality-control requirement into a computer-vision workflow capable of distinguishing expected configurations from potential defects while handling variation in component appearance.

Data→Annotation→Training→Evaluation→Improvement↺

Technical layer

PythonWorkflow foundation

YOLOv8Detection

OpenCVImage processing

PyTorchDeep learning

FlaskBackend integration

RoboflowData workflow

Confidentiality by design

Built in a confidential industrial context.

Because this project was developed in a professional environment, proprietary images, datasets, source code, internal system details, and project-specific results are not publicly disclosed.

What is shared

Problem · Methodology · Technical contribution · AI workflow

What remains private

Industrial data · Source code · Internal implementation · Confidential results

From component images
to automated visual verification.

Computer Vision → Deep Learning → Industrial Quality

Status · Confidential Professional ProjectNext project: Olive Oil Export Prediction