TransCityRail South: Using AI with Assets

Last month, I had the pleasure of taking part in a panel discussion at TransCityRail South alongside Gareth Evans (Network Rail), Anna Saunders (High Speed One), Richard Graham (KeolisAmey Docklands) and Yung Loo (Arup), fantastically moderated by Helen Fospero.

We were looking at the use of AI in asset management. The basis of this is around condition monitoring, where a variety of physical sensors can be used to build up an empirical understanding of an asset over time. For example, large civil infrastructure assets, such as a bridge, can be fitted with movement (gyroscopes and accelerometers), temperature and vibration sensors to understand its performance in real-time. This capability has been available for many years but the use of AI over such large quantities of data means we can now start to predict its future condition and, critically, how long we can continue to safely use it.

But you need good quality data first – and a high volume of it. A few spreadsheets won’t suffice. Start with the data collection first, get it extracted from your asset base into a cloud-based data lake and then others can start to help you take advantage of it.

Beyond asset safety, the principle use case here is preventative maintenance. The phrases “sweat the asset” or “fix on fail” will be very familiar to asset owners, but knowing exactly when maintenance or replacement is needed can become highly subjective. Intrusive, expensive inspections undertaken by workers are required first, assuming that your asset is easy to reach without (a) closing it from general use, (b) doesn’t necessitate specialist equipment (e.g. to work at height) to reach it and (c) can be accessed without weather dependence (such as offshore assets). Each time, you’re looking for a delta in the data; a kind of tangible difference that then requires consultation, opinion and then a decision.

However, it was clear from questions put to us in the room that there is still concern about the use of AI in our sector, which is why in one of my answers I mentioned the human-in-the-loop methodology. This is where one can take advantage of the scale and performance of AI while using a human to mark its homework, so to speak. In my experience, this approach has brought about significant savings while avoiding the introduction of risk. We can repeatedly apply various algorithms over the data to get different views of its condition, and then, subjectively or not, use that to make rational decisions as to an asset’s safety and functional ability.

The conference included superb keynotes on passenger experience, sustainability, strategic delivery and Network Rail’s High Speed division, as well as a rail supplier showcase nearby. Well worth looking out for next year’s event.