A vision-based AI system for detecting corrosion in metal alloys

A vision-based AI system for detecting corrosion in metal alloys

The Challenge

A leading specialist in water treatment needed a method to identify tank maintenance at customer sites to increase proactive maintenance of their infrastructure. The current process was time consuming and often led to costly errors. The solution was a mobile application to capture images of metal coupons planted in the water purification systems to detect their composition and corrosion using AI that matched the precision of laboratory results. Tatras was tasked with building this AI solution.

Hypothesis

  • A deep learning-based vision model utilizing transfer learning can accurately detect and classify corrosion in metal alloys from images, providing rapid assessment and recommendations comparable to traditional laboratory methods.
  • This approach significantly reduces inspection time and improves efficiency in corrosion management across various industrial applications.

Execution

  • Collect diverse corrosion images from various sites and alloy combinations.
  • Prepare and standardize images for training.
  • Implement transfer learning using pre-trained vision models.
  • Develop algorithm to generate risk scores based on corrosion intensity.
  • Evaluate model accuracy and risk score reliability.

Outcomes

  • Test accuracy of 98%+ achieved.
  • Created interpretable reports linking visual cues to corrosion severity.

Project Highlights

98%

Accuracy on test data

7Days to 7Minutes

REDUCED OVERALL TURNAROUND TIME FROM THE TRADITIONAL 1 WEEK TO MINUTES