Processing of huge numbers of images using computer vision for predicting the severity of corrosion or fatigue and recommending preventive actions.
- A pilot was conducted using surface images of the assets and structures, taken at different times during the inspection, maintenance or shutdown/ turnaround in last few years.
- The images and associated data were ingested into tcg mcube and placed in the contextual form based on the classification of assets and structures.
- Part of the images and associated data were used to train the computer vision AI model, while rest was used for the testing.
- It was observed that the condition of bad images and related severity of corrosion or fatigue were predicted with an accuracy of more than 90%.
- A recommendation engine was also developed using associated data points and business feedback/rules.
- The engine provided recommendations like repaint or part/ section repair or on high severity, even replacement.
- Data Used
- 9K+ Images of equipment (tanks, heat exchangers, columns, reactor, separator/ surge drum), pipes, valves, fittings, flare stack, pipe support, structures of plant.
- Technology Used
- Digital camera, user interface to upload images to tcg mcube
- tcg mcube – computer vision (R-CNN, DNN)
Business Impacts & Outcomes:
- Severity of corrosion or fatigue to the assets or structures was reported based on the images
- Low severity (closer to 0.10) means, not much corrosion or damage has occurred to the metallic surface
- Medium severity (closer to 0.70) signifies the asset/structure needs repair
- High severity (closer to 0.99) means replacement is required
- Based on level of severity, insights were provided on factors contributing to the damage (RCA)
- Recommendations like ‘No Action Needed’, ‘Repair’ or ‘Replacement’ was also provided through the system