A Tier-1 supplier to aerospace and automotive OEMs needed to automate quality control of newly manufactured components, including engine cowls, airframes and instrument cluster panels. The goal was to minimize human involvement in QC checks and streamline production-line operations by detecting non-conformities and defects with AI/ML-based computer vision and traditional image processing.

A team of three ALTEN engineers in India designed and developed the solution over six months. It uses two techniques: a deep learning model trained to identify defects in newly manufactured components, and image comparison against an approved reference (“golden”) image.

The scope included:

• Design and development of an automated QC application
• Deep learning model development and training for defect detection
• Golden image comparison for instrument cluster panels
• Traditional image-processing pipeline development
• Android application development for a production-line tablet

Challenge:

Manual Quality Checks

• QC of aerospace and automotive components relied on human inspectors, which added time and cost to production
• Manual inspection was less accurate than the client needed

Hard-to-Spot Surface Defects

• Painted surfaces on engine cowls and airframes could carry flaws such as blisters, embedded foreign objects and poor surface preparation
• These had to be identified before components left the production line

Inspection on the Production Line

• The solution had to run on a device installed directly on the production line

ALTEN Solutions

Deep Learning Defect Detection

• Developed a deep neural network model trained to identify defects in newly manufactured components
• Detected flaws in the painted surfaces of engine cowls and airframes, including blisters, embedded foreign objects and poor surface preparation

Golden Image Comparison

• Compared the live camera feed of instrument cluster panels against an approved golden image
• Highlighted pixel-level differences beyond a set threshold to identify defects

Traditional Image-Processing Pipeline

• Built an image-processing pipeline in C/C++ and GStreamer to support defect analysis alongside the deep learning model

Android Inspection Application

• Deployed the model in an Android application on a high-end tablet installed on the production line
• Used the application for QC of engine cowls and airframes, with detected defects displayed on screen

Business Benefits

Reduced Inspection Time and Cost

• Automated QC cut the time and cost of inspecting each component

Higher Inspection Accuracy

• AI/ML-based inspection proved more accurate than inspections carried out by humans

Less Human Involvement

• Automating defect detection reduced the manual effort needed for routine QC checks

Improved Operational Efficiency

• Streamlined QC helped the production line run more efficiently

Tools, Technologies & Expertise

• Engineering and Development Tools
• Java (Android application development)
• Python (ML model training)
• PyTorch and TensorFlow (deep neural networks)
• C/C++ and GStreamer (image-processing pipeline)

Engineering Capabilities

• Computer vision development
• Deep learning model training
• Automated defect detection
• Golden image comparison
• Android application development
• Image-processing pipeline development

Domain Expertise

• Aerospace manufacturing
• Automotive manufacturing
• Smart factory applications
• Production testing, repair and overhaul
• Quality monitoring solutions

Where ALTEN Adds Value

• Applying both deep learning and traditional image processing, choosing the right technique for each component
• Taking AI/ML models from training to a working application on the production line
• Reducing human involvement in QC while improving inspection accuracy

Why ALTEN
ALTEN combines computer vision, deep learning and manufacturing know-how to automate quality control for aerospace and automotive suppliers. By delivering a production-line application that detects defects more accurately than manual inspection, ALTEN helps manufacturers cut inspection time and cost while improving operational efficiency.