Crack Identification is an important challenge in estimating the reason for the cause of the failure. Onboarding of a defective machine into the insurance ecosystem leads to big losses being incurred. Most of the heavy machinery susceptible to fatigue loading is located in remote locations, where the connectivity may be poor.

What it does?

The Cracke is an end-to-end automatic report generation tool that identifies cracks to pixel-level accuracy. It is computationally inexpensive and so can be deployed on even a Raspberry Pi. Cracke leverages the Paris Equation for crack propagation to estimate the number of cycles of load applied. The tool compares the number of cycles of load with a standard loading condition for the material to identify the root cause of the defect. Then a report is generated in an HTML format.

How We built?

Cracke has 3 main components:

Image Processing Pipeline

  • Median Blur Filtering
  • Canny Edge Detection
  • Morphological Segmentation
  • Bounding Box Generation

Crack Estimation Pipeline

  • Approximate Crack Length
  • Use Paris Equation for crack propagation
  • Compare with standard loading for a given material

Report Generation Pipeline

  • Status of the material
  • Analysis of the image
  • Estimate crack start and cause
  • Generate report

Hurdles we overcame?

  • Integration of the pipelines with the frontend for the app
  • Deploying the pipelines on the Raspberry Pi

What we learned?

That the Occam's Razor is still valid, that a simple solution can be very effective in solving huge problems and saving a ton of money.

What's next for cracke?

  • Improve material database
  • Incorporate more domain knowledge
Share this project: