QualiChip_Accenture

An AI that recognizes whether a manufactured chip is a "good" or "bad" chip (wehther it is damaged or not). Also provides a user dashboard for monitoring the manufactured chips in real time.

Comment

Inspiration

Real Time access of chip quality: Misproductions will be detected in real-time enabling the possibility for direct intervention so quality aspects can be ensured. Sheds some light into a highly automated blackbox production line.

What it does

By using the power of Artificial Intelligence, broken chips are detected during production. A Power BI-Dashboard visualizes KPIs concerning the production's overall quality and displays images of currently produced chips.

How we built it

  • We built two pipelines, one for training and one for inference
  • Training pipeline: Initial dataset -> Machine Learning Service Workspace -> Preprocessing & Augmentation (python scripts) -> Training in Custom Vision
  • Inference pipeline: Image file -> Azure Blob Storage -> Preprocessing-Script (Python) -> Azure Blob Storage -> Inference (1. Bounding Box 2. Classification) -> Azure Cosmos DB -> Power BI

Challenges we ran into

  • Data-Labeling (dust particles problem). Also, labeling some data correct was also a challenge, since some broken chips could not be detected that easily even with the human eye.
  • We planned to use Azure ML Pipeline for the image preprocessing and transformation, but the input and output format are not compatible, which is why we had to preprocess the data with a python script and by using tools such as OpenCV, Tensorflow, or Pillow.

Accomplishments that we're proud of

  • Having a functional prototype that classifies chips with high accuracy.
  • Accomplished image cropping with object detection to reduce noise and improve performance

What we learned

  • Azure (Custom Vision, Machine Learning, Blob Storage, Cosmos DB)
  • Power BI
  • Computer Vision (Preprocessing, Augmentation)
  • Labeling of Bounding Boxes

What's next for QualiChip_Accenture

  • Direct connection to IoT via Azure IOT Hub, Scaling and Testing within a real production environment
  • Using Azure DataFactory for pipeline orchestration

Built With

Share this project:

Updates