AI can help the rail industry to be more flexible, efficient and automate the railway. For Siemens Mobility, they have been using AI for over a decade to support their customers.
While Artificial Intelligence (AI) seems like an emerging technology, especially with the emergence of ChatGPT, global technology company Siemens has been using AI across its technology solutions for more than a decade to help analyse the complex ecosystems they operate in.
Dr Claus Bahlmann, head of research and development for Artificial Intelligence, and principal expert for AI and computer vision in Siemens Mobility, explained that it is all about getting smarter and automating as much as possible.
“With automation – powered by AI – we can increase flexibility to deal with varying peak travel times, reduce the downtime caused by maintenance and staffing issues, and even carry out upgrades without affecting rail operation so much. All of this benefits the passengers,” he said. “There’s a limit to how much new railway infrastructure we can build, and investment is also limited, which is why we need to take this digital approach. The technology is there to increase capacity while making rail travel more reliable, more comfortable, and more inclusive.”
AI can also make maintenance more effective and efficient.
“Technicians in the field have traditionally relied on a service log and manual, but now they can point a tablet at the area of concern and AI will identify what part they are looking at,” Bahlmann said.
“AI will then provide all the contextual information they need: the manual, the circuit diagrams, or a catalogue of spare parts.”
Inspection is another area where AI is being deployed into Siemens’ products.
Bahlmann said sensors are mounted on the infrastructure where it is possible to inspect the rollingstock, because vehicles are scheduled to pass that location, and sensor data of the rollingstock can be collected and inspected.
When sensors are mounted on the vehicles, they will acquire sensor data from the infrastructure and the AI will recognise specific areas that need inspection or maintenance.
“For example, we can inspect the condition of rail tracks from images captured by cameras mounted on trains and looking into the track bed,” he said.
Gonzalo Martinez Delgado, head of customer service for Australia and New Zealand at Siemens Mobility, recalled how AI helped a Swiss rail operator who wanted to improve its network availability and minimise impacts to passengers.
“Delayed minutes of trains caused frustration to the end customers, the passengers, and impacted the network operation,” Martinez Delgado said. “It was realised that much of the delays were due to point machine faults and the aim was to reduce that delay to improve the operation and customer satisfaction, and at the same time, save money for the operator.
“We saw the opportunities of using AI that could predict some of these failures to avoid delays.”
Martinez Delgado said the customer defined the targets and Key Performance Indicators (KPI’s) to measure the success of the initiative at the beginning of this joint co-operative journey.
The project started in 2019 and through the following four years, both parties worked together expanding the AI capabilities based on customer needs, starting with a pilot project, followed by a proof of operations phase and finally incorporating the AI tools into the daily maintenance process of the customer, with the aim of helping the customer to achieve 100 per cent system availability.
Optimising maintenance
The use of AI technology on the Swiss network meant it could not only track the condition of infrastructure but could also anticipate potential faults and alert the maintenance team, improving efficiency.
“The team then know in advance where a failure might occur and how critical the situation is, to properly plan in advance when best to go to site, the tools needed, etc,” Martinez Delgado said. “I use the example of how much easier it is to go to the supermarket with the shopping list for your groceries than blindly walking around for your shopping wondering what you need at home.”
It is not just about having the tool of AI itself, but knowing how best to provide the inputs, implement the tools, and analyse the data that makes it beneficial.
Martinez Delgado said it’s important to understand the problem the customer is trying to solve and how the end user consumes the data.
He spoke about the importance of the three integrated core elements of the golden triangle, people, processes, and products that need to be addressed together when implementing innovation to drive productivity and efficiency successfully.
“AI is not enough on its own, at the end of the day, the person still needs to be there,” he said.
“AI will not replace people. We can have the most wonderful tool but if the people don’t believe in it and don’t want to use it, then it is pointless.
“The technology is simply a tool, and we still need the person to go out and decide whether the work is needed and then carry out the work.”
Working with customers
Martinez Delgado emphasised the importance of listening to each customer and understanding their specific needs to find the right solution.
He described it as a journey of collaboration between the Siemens team and customers.
“I don’t think AI is a plug and play system, it needs a lot of adaptation to meet customers’ needs,” he said.
“What we are doing is listening to problems they may be having and systematically bringing in ideas to fix those issues.
“We are introducing this technology across all our new projects as well as legacy systems. All these projects, both new and legacy, give us an opportunity to learn and improve the digital platform Railigent X, where all the different use cases are implemented in reusable apps.”
The prediction paradox
Martinez Delgado said one of the difficulties with this new predictive maintenance technology system is that it is difficult to prove that it is working.
He uses the term “prediction paradox” to describe the fact that it is difficult to prove that you have a valid tool to predict failures in a running system, because once implemented, the failures will be avoided and then obviously you cannot prove a failure would have indeed occurred. In other words, how can you measure something that does not happen?
The term highlights the difficulty in validating the success of a prediction tool when its effectiveness prevents the very events it aims to predict.
At the beginning of the project the system was monitored by the technicians, but no special interventions were made beyond the prescriptive preventive maintenance tasks. The aim was to confirm the correlation in between AI failure warnings and actual failures.
“It was an opportunity for us to confirm the system was working and give a clear indication of its effectiveness compared to the KPIs defined by the customer,” Martinez Delgado said.
“Once the customer trusted the technology the next step was to make it available to the end users in the way they need it. We learnt that the technicians were not necessarily interested in how the AI tool was doing the predictions but more in the specific action required. They expected a simple input and not a sophisticated engineering tool. This is where the relevance of the people and process implementation is crucial. And the other key factor was the low number of false positives because otherwise the technicians would have discarded the tool. They did not want to be alerted unnecessarily.
“Initially this was a challenge, but as we had more and more data come through, we found ways to prove the success of our system and the benefits to the customer.”
Martinez Delgado said that as more data came in, the team could graph the number of operational faults compared to the number of predictions, which resulted in an interesting correlation – as the number of predictions went up, the number of operational faults in the network would decrease, whereas when predictions went down the number of faults would climb. “For both parties, this was the best indication that the system was effective and working,” he said.
What AI means for Australia
Siemens Mobility is already using AI as part of its service offering for projects in Australia, and Martinez Delgado believes the country is well positioned to capitalise on the technology across its varying networks.
“At the end of the day it doesn’t matter how different the networks are, every train has bogies, wheels, doors and we can use the data coming from the different diagnosis systems to harness AI and support our customers,” he said.
“Often the problem is the same, it is simply the details of the problem that can be different.”
Martinez Delgado said that the basis of Siemens’ AI systems remains the same but will be customised across networks. Once the system is fine-tuned to a network’s specific needs it is set up for success.
He said that from the recent record levels of investment into rail projects across Australia, it is important to ensure rail operators are maximising the effectiveness and availability of their networks.
“Budgets are reducing, and governments cannot be spending billions on new trains. They need to ensure they are extending the life of the network and the trains running on them,” he said.
“AI ensures greater efficiency in maintenance teams as well. We work with a freight customer to help monitor and assess their critical wayside infrastructure in real time. This saves manual inspection and reporting, where previously it was taking days for an issue to be investigated.
“AI will make all of our networks more efficient and productive.”