Could AI prevent train crashes ?

© ikku Sama/Pexels

AI is rightly said to have many flaws. ;nbsp;: monstrous consumption of resources, opacity of large language models, data confidentiality issues or ethical considerations regarding its autonomy. However, it should not be denied that it can be of great help in many areas: ultra-fast drug design, natural disaster prediction, fraud detection in the financial sector or supply chain optimization.

One sector in which this could prove very useful; even if at first glance it seems niche; is the railway sector. Researchers at Stony Brook State University (New York) have just developed a system combining artificial intelligence and guided ultrasound waves (or Longitudinal Ultrasound Guidance) to examine switch rails. These are key points on the networks where trains change tracks. To learn more, their work can be found in the October edition of the journal NDTE & E International.

The rapid expansion of high-speed networks: a major challenge

According to recent data from the International Union of Railways, the global high-speed network now stands at 59,000 km. The latter has experienced exponential growth over the past few decades, driven by increasing demand for fast travel.

Growth necessarily means increased demand on infrastructure, particularly at the level of switch tracks. These areas are a bit special since they are subject to particularly intense mechanical stresses due to their complex geometry and their central role in guiding trains. These rails are subjected to high dynamic loads (see video below) and are more vulnerable to fatigue phenomena and structural deterioration.

Subscribe to Presse-citron

Their maintenance is therefore of capital importance for the viability of the networks; hence the need today to develop innovative inspection solutions, capable of early detection of anomalies likely to compromise traffic safety.

AI and physics: the winning combination?

Faced with the limitations of traditional inspection methods, Stony Brook researchers have developed a radically new methodology. For Prof. Zhaozheng Yin, an associate researcher in biomedical informatics at the American University's AI Innovation Institute, the inspection of these rails must be done non-invasively. “The use of guided waves has emerged as a relevant solution. These waves propagate over relatively large distances and are particularly sensitive to defects, allowing the inspection of large areas in short time frames ».

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

Guided wave rail maintenance involves sending ultrasonic waves (mechanical vibrations that propagate in a medium at a frequency higher than what the human ear can perceive) into the rails. These waves then propagate in the rails and are reflected by the faults present, thus making it possible to locate them and assess their severity by analyzing the signals received.

This technology is perfectly suited to the specific operational constraints of the railway sector, where maintenance interventions must be carried out in limited time slots, generally at night. A technique that guarantees a ratio inspection speed/reliability of optimal results, while ensuring exhaustive coverage of critical areas. However, it is not new since it dates back to the 1960s; what is new here is its alliance with artificial intelligence.

The geometric complexity of switch rails, combined with their intrinsic mobility, required the development of algorithms capable of analyzing complex acoustic signatures. This is precisely what these researchers have managed to do: develop a neural network that is very capable of carrying out this task.

The model decomposes ultrasonic signals into distinctive features, allowing the identification of acoustic variations characteristic of a structural failure. It particularly excels in its ability to ignore the noise inherent in the railway environment (mechanical, environmental or electrical noise), accurately isolating the signatures specific to the anomalies being sought.

The system's performance is exceptional, with a fault detection rate greater than 91%. This approach outperforms all existing methodologies in all aspects evaluated during the rail analysis.

If this technology were integrated on a global scale, the high-speed network could achieve an extremely high level of safety and optimization. Drastic reduction in the risk of derailments and accidents, more precise maintenance planning and increased network availability: three essential factors for a reliable rail network. The thorny question remains as to the economic viability of this solution, which would necessarily require substantial investments. However, the long-term benefits could offset these initial costs; everything will depend on the ability of rail sector stakeholders to quantify them precisely in order to convince political and financial decision-makers to reach into their pockets.

• Researchers have combined ultrasonic waves and artificial intelligence to efficiently detect defects in railway tracks.
• This method allows for rapid and precise inspection of critical infrastructure, including switch tracks subject to high stress.
• This technology could greatly improve the safety and maintenance of railway networks if adopted on a large scale.

📍 To not miss any Presse-citron news, follow us on Google News and WhatsApp.

[ ]