Challenge

This major utility provider primarily uses traditional ground crew inspectors, supplemented by digital aerial imagery, to maintain 70 000+ transmission assets. Inspections are rotational (not targeted) covering approximately 20% of assets per year. Their current processes are impacted by substantial challenges:

• Volume of work requires up to 5 years duration between inspections on individual assets
• Process is tedious, inconsistent, & not applied to all assets annually
• Higher operational transmission structure maintenance costs related to manual inspections
• Manual effort necessitates significant labour input from valuable resources
• Worker safety concerns arise when deployed in the field
• Over 1 million images have been collected since 2012 & are not currently fully utilized

Working together to understand these challenges, we’ve developed a solution leveraging AI visual recognition. Visual recognition models were created to detect damaged assets present in yearly maintenance imagery, & to flag these instances for further review. The solution is modular & scalable, enabling additional value with future development.

‍

Solution

The solution is actively being delivered to detect damage on self-weathering steel transmission structures throughout the client territory. This optimizes the existing process used to inspect 157 000km of transmission lines including the collection of 150 000 images annually.

Delivered using Microsoft Azure (via Platform as a Service), featuring high performance & scalable architecture for training & running AI models. This enables automated analysis of transmission asset images, flagging images that contain damaged assets for human inspection.

Result

The initial proof of concept was successful, resulting in the requirement for a full enterprise solution implementation, to be completed by June 2021. The expected results of the enterprise solution include:

• Reduction of operation transmission structure inspection costs year over year as the solution continues to mature
• Exponential reduction of time taken for first-pass image analysis
• Significant reduction of manual image inspection by returning only images that contain damage
• Implementation of consistent asset evaluation, mitigating the risk of subjective human assessment
• Efficient & targeted ground crew tasking, deploying teams into the field only where the need for physical maintenance is confirmed
• Reduced risk of outages & asset failure, due to rapid damage detection & repair
• Improved safety for workers, with a drastic reduction in ground inspections required in remote/dangerous locations
• Leveraging the investment in >1 million still images collected since 2012, this solution utilizes every image on file in the creation of the AI Visual Recognition solution
• Capture & retention of Subject Matter Expertise through model training

Delivery of a successful proof of concept was vital to demonstrate the effectiveness of this new approach. Increasing familiarity with AI concepts drove an organizational desire to apply this technology to existing workflows. With similar challenges being faced by other utility providers, this scalable solution can be applied for any organization that manages physical assets through imagery.