THE use of artificial intelligence (AI) has accelerated in recent years due to the falling cost of data storage and processing, rapidly expanding data availability, and improved data storage and modelling techniques. Analytical AI works with historic data to make numeric predictions, while generative AI enables machines to produce new output similar to human-generated content. Generative AI has been building momentum and reached a turning point at the end of 2022 when applications such as ChatGPT became publicly available.

A third of respondents to McKinsey’s annual global survey on the state of AI in 2023 said that their organisations regularly use generative AI in at least one business function, while 60% of those who have adopted analytical AI said they are also developing generative AI use cases. While the railway industry has in the past faced challenges in adopting digital technology due to limited data availability and quality, regulatory requirements and lack of a standardisation, both analytical and generative AI provide an opportunity for the rail sector to progressively embrace digitisation.

According to The journey toward AI-enabled railway companies, a report produced by the International Union of Railways (UIC) in partnership with McKinsey, there are over 100 potential use cases for AI in the rail sector. Based on research that included a survey of 11 railway companies across Europe and Asia, and in-depth interviews with 15 leading companies and original equipment manufacturers (OEMs), the report says that AI implementation is focusing on 20 key use cases, using analytical AI to improve train punctuality, customer engagement, safety, and operational performance.

Of the 11 companies surveyed, 38% are embarking on their first at-scale deployment, building up internal capability. With that in place, 25% are undertaking at-scale deployment of several AI use cases, while 13% have several ongoing pilots and 25% are at the early development stage with no pilots as yet underway.

Use cases deployed at scale by operators include crew and shift optimisation, where 40% of companies interviewed have deployed AI to allocate staff more effectively, reducing labour costs. Cutting energy consumption is another priority, enabled by the deployment of driver advisory systems incorporating AI-enabled software. Systems such as EcoRail, which has been adopted by Via Rail Canada, are expected to deliver a 10-15% reduction in energy consumption.

Other AI use cases are still in the pilot phase, such as using the technology for predictive fleet maintenance, which has now been adopted by 50% of the operators interviewed. These interviews confirmed UIC research that predictive maintenance has enabled a 15% increase in reliability, a 20% reduction in maintenance costs and a 30% reduction in train failures. Around 30% of railway companies said that they are pursuing the use of algorithms in service scheduling, which assess customer demand, determine priority train paths and define the optimal output within constraints such as station capacity, staff availability and the profitability of the services concerned.

All of the major infrastructure managers interviewed for the report are now using predictive maintenance to prioritise work on the most critical assets with the highest probability of failure. This AI use case has been deployed for some time, with the more mature solutions now using both internal and external data to identify the optimum maintenance cycle.

Austrian Federal Railways (ÖBB), for example, combines internal data with Lidar scanning to monitor trackside areas and identify hazards such as trees that might fall on the railway. The UIC observes that predictive infrastructure maintenance can typically reduce unplanned downtime by 15-25%, cut maintenance costs by 15-30%, deliver an increase in the capacity to detect failures of 100% or more, and reduce delays per train by 20%.

Use cases in the pilot phase include passenger flow management to optimise the movement of passengers in stations, avoid bottlenecks and improve security.

Infrastructure managers are also using deep learning to identify the best use of network capacity, taking into account operator needs, maintenance requirements and external factors. German Rail (DB) is using AI to identify upcoming peaks in demand at stations and on long-distance services, enabling high-capacity trains to be dispatched to relieve potential bottlenecks. Of the infrastructure managers interviewed for the report, 60% are using AI in real-time traffic management, identifying the most efficient routes, coordinating services and reducing overall disruption.

By changing the way that services are planned and delivered, AI has the potential to unlock benefits of between $US 13bn and $US 22bn a year worldwide, according to the report. Based on the cost structure of European railway companies, for a company with an annual revenue of €5bn AI could deliver €700m a year in value. This combines reductions in staff, maintenance and corporate costs with increased revenue from using AI for passenger revenue management, such as dynamic pricing, and from optimising use of network capacity. The greatest savings come from using AI for predictive infrastructure maintenance, cutting costs by 25%, followed by predictive rolling stock maintenance (20%) and improving energy efficiency (15%).

To capture this value, the reports says that all stakeholders should prioritise the development of AI use cases that serve a clear need, rather than developing technology for technology’s sake. Alongside the technology and a data management strategy, successfully deploying AI in rail will require AI-specific skills and roles, as well as a strategic roadmap for implementation.