5 Takeaways on AI Inspections from the UIIC Symposium 2024
Concept X recently sponsored the UIIC Symposium in Richmond VA, USA, generously hosted by our friends at Dominion Energy. The focus of the entire symposium was around the use of visual data for utility inspections, which perfectly aligns with what we do at Concept X! It was a superb two days of learning and networking, including a demo of Skydio’s latest drones for grid inspections.
It was inspiring to hear from all the utilities about their own initiatives around drones, imagery, LiDAR, and most of all AI. While the industry as a whole is excited about the potential of AI in inspections, utilities are at different stages of technology maturity and deployment. Each of them is working on their own pathway to AI inspections, but along the way there are common challenges, opportunities, and learnings. So here are our 5 key takeaways from all the sessions and our many discussions –
AI is here, but it’s still early days
While Large Language Models (LLMs) like ChatGPT and Gemini have been all the rage for the last year, AI is gradually making its way into visual inspections within utilities. We heard test cases and deployments using drones, ground patrols, and satellites by various utilities. While all of them agree with the potential of AI as a value multiplier for remote inspections, most are still in early stages of testing and rollouts. Utilities need to be operationally conservative to manage risks, but at the same time need to balance this against opportunity cost of maintaining status-quo and not innovating. With the growing pressures of an ageing grid and increasing climate risk, the balance is tilting in favour of innovation and automation.
Data remains key, but we can do more with less
The age-old mantra of GIGO (garbage in garbage out) remains true as ever for all Machine Learning (ML) deployments. Good thing is that with constantly improving drones and cameras, it’s easier than ever before to get high quality data cheaper. We certainly saw examples of this from multiple utilities who have started to accumulate decent volumes of data through their inspections. However, this data distribution is non-uniform across utilities (i.e., not everyone has same volume) and across use cases (eg., detection of rare defects). Also, as one presenter put it, “every image of an asset is an asset”, but the quality of these images is sometimes not ML appropriate. Thankfully ML has evolved to a point where training new models does not require thousands of images, which opens up possibilities for utilities of all sizes. At Concept X we have mastered the art of low sample ML training using our pre-built ML library, transfer learning, and unique synthetic data generation approach. So, give us a shout if this is of interest!
Capitalisation of AI inspection technology could be a game changer
Utilities continue to be hard-pressed on Opex, which means budgets are tight for inspection innovation, even though that’s the very thing needed for long-term O&M cost reductions. This catch-22 has led utilities to be more creative in their accounting and operationalisation of AI for inspections. The idea is to consider inspections as “structure assessment or continuous data modelling surveys”, which enable deeper understanding of the current state of all assets and in-turn improve long-term asset health. With this lens, initial deployment of AI technology which creates these detailed and accurate asset assessments can be potentially capitalised. We definitely agree, but also believe flexibility is needed at the supplier end to make sure their software costs can indeed be capitalised.
The whole is greater than the sum of its parts
Not surprisingly most utilities are trying to solve for very similar inspection use cases using broadly similar approaches. While it’s true that T&D structures vary across utilities, we believe there are enough similarities to create significant collaboration opportunities. While events like UIIC are a great platform for knowledge sharing, we think it can go further than this. This is especially true for smaller utilities and co-ops which can benefit from joint inspection protocols, underpinned by shared AI models and common foundational data. The benefits from such an approach can be immense – from time and cost savings to more accurate AI detections. Concept X is currently working on revolutionising ground patrols using this framework, so please reach out if you would like to participate in the program!
Building the right business case: speed, quality, cost, safety
While it’s easy to get carried away with AI and the latest remote inspection technologies, basic business fundamentals ultimately remain key for successful deployments. Not all inspections need the latest drones, not everything can be detected by satellites, and sometimes a phone camera may just be enough. Right tool for the right job! The right tech should provide measurable uplift on the most critical inspection value levers – speed, quality, cost, safety, and anything else the business values. Additionally, operationalisation cost, dependencies, and user adoption need to be considered and planned before deploying. At Concept X we regularly work with our utility partners to help them maximise the value from AI inspections, and it all starts with defining the right business case.
If you have been considering AI to improve your inspections or are interested in learning more about any of the above points, please reach out!