List of Topics:
Location Research Breakthrough Possible @S-Logix pro@slogix.in

Office Address

Social List

AI-Driven Defect Detection in Manufacturing: A Semantic Segmentation Approach using Azure ML and Azure Stack Edge

Defect Detection

Defect Detection in Manufacturing: A Semantic Segmentation Approach using Azure

  • Use Case:

    In modern smart manufacturing, quality assurance is critical to reduce defects, avoid recalls, and ensure customer satisfaction.Manual inspection is time-consuming, error-prone, and costly.

    Semantic segmentation models can classify each pixel of an image, detecting scratches, dents, cracks, or misalignments on manufactured products.

    Deploying AI models on Azure Stack Edge enables real-time, on-premise defect detection close to the production line, while Azure Machine Learning (Azure ML) supports scalable training and optimization in the cloud.

Objective

  • To develop a semantic segmentation model that detects manufacturing defects at a pixel level.

    To train and optimize the model using Azure ML cloud resources and deploy it on Azure Stack Edge for low-latency inference on the factory floor.

    To enable real-time, automated, and accurate defect detection, reducing reliance on manual inspection.

Project Description

  • Data Collection & Preprocessing :

    Collect defect datasets from manufacturing lines (images of surfaces, components).

    Annotate defects at pixel level using tools like Labelbox or VOTT (Visual Object Tagging Tool).

    Preprocess images (resize, normalization, augmentation).

Model Development (Semantic Segmentation)

  • Use architectures like U-Net, DeepLabV3+, Mask R-CNN for defect segmentation.

    Train on Azure Machine Learning Service with GPU clusters.

    Apply transfer learning for improved accuracy on small datasets.

Model Optimization

  • Use Azure ML HyperDrive for hyperparameter tuning (batch size, learning rate).

    Optimize for accuracy, recall (defect detection sensitivity), and inference speed.

Deployment

  • Containerize the model using Docker.

    Deploy to Azure Stack Edge devices placed on the shop floor.

    Enable real-time inference with minimal latency.

    Provide dashboards for operators to view defect detection results.

Monitoring & Feedback

  • Use Azure Monitor & Application Insights for telemetry.
  • Retrain models periodically with new defect data collected from production.
  • Key Technologies & Azure Services :
    Azure Service Purpose in Project
    Azure Machine Learning Service Model training, experiment tracking, and pipeline automation
    Azure Blob Storage / Azure Data Lake Storage of annotated defect images and datasets
    Azure Compute Instances / GPU VMs High-performance compute for training semantic segmentation models
    Azure ML HyperDrive Hyperparameter optimization for segmentation networks
    Azure Stack Edge On-premise, low-latency deployment of defect detection models on the factory floor
    Azure Container Registry (ACR) Storage and management of Docker images for deployment
    Azure Monitor & Application Insights Monitoring, telemetry, and logging of deployed models