Home Blog AI-based risk analys ...

AI-based risk analysis for hydrogen-fueled transportation

Roberto Mascherona

The transition to more sustainable energy sources has become a global priority. In this context, the use of hydrogen as a fuel is a promising option for the transportation sector.

However, due to its relative novelty and chemical instability, the introduction of hydrogen as an energy source in rail transport requires careful assessment of the associated risks and the design of appropriate safety measures.

To address this challenge, the Aramix team developed a framework based on the use of artificial intelligence to analyze and manage the risks of hydrogen-fueled rail infrastructure. This framework covers the entire life cycle of the element, including production, storage, refueling, and transportation. 

You can read the full case study here.


Hydrogen, rail transport and data

One of the main obstacles in addressing the risks associated with hydrogen in rail transport is the lack of comprehensive data and practical experience on these facilities. Therefore, we integrated traditional risk analysis techniques with the application of artificial intelligence models to identify unknown events and risks, also known as “unknowns.” This integration allows us to structure cause and effect scenarios, identify safety barriers, and evaluate their effectiveness on paper, i.e., on paper.

The framework developed by Aramix has 3 essential components and steps:

  1. First, traditional and robust methodologies such as HAZID analysis, HAZOP analysis, and FMECA analysis are used to identify relevant risks, evaluate process parameter deviations, and analyze possible failures and their causes and effects on the infrastructure. In addition to these traditional analyses and in order to better understand phenomena, we have also developed a complementary activity, developing a Computational Fluid Dynamics (CFD) based model to simulate the physical behavior of hydrogen in the event of a leak. This model, combined with the available expert knowledge and partial scientific literature available on the subject, allows us to evaluate the adoption of safety barriers and to confirm or remove the risks initially assessed. It also allows for general considerations that can be applied to different types of infrastructure;
  2. in addition, an innovative approach is integrated, using Systems Theoretic Accident Modeling and Processes (STAMP), a qualitative framework for accident assessment that considers the system as a whole that allows the identification of possible accident scenarios and their possible impact (in this specific case having a focus on the presence of Hydrogen) to be analyzed in relation to hazards, possible system losses, and control management. Also considered in the identified framework is the development of the Goal Tree Success – Master Logic Diagram (GTST-MLD), a hierarchical goal-oriented model that quantifies the impact of hydrogen-caused accidents in the system. This other model, although starting from different assumptions than STAMP can allow for a qualitative-quantitative view complementary to it focusing on reliability and system availability.
  3. Finally, the Modeling and Simulation (M&S) model uses artificial intelligence to study and analyze accident behavior. Through the adoption of causal inference models, the M&S phase quantifies process deviations and identifies cause-and-effect relationships and interactions among various components of the system. In essence, this module allows us to simulate and evaluate how the system will respond in emergency situations, enabling a deeper understanding of potential risks and possible preventive measures to be taken.


Conclusions and outputs: how to depict complex probabilistic models 

The output of this multifaceted risk analysis framework includes a quantitative risk analysis and probabilistic information about the reliability of the entire system.

Beyond what we were able to detect-the conclusions of the study that were delivered to the Client and are not releasable to date-the point is HOW we were able to graphically represent the results of a truly complex model in an immediately intelligible way.

Aramix implemented a causal model based on a Bayesian network within the STAMP framework to address uncertainties related to the lack of data and real-world scenarios in the context of hydrogen infrastructure. But what are Bayesian networks?

A Bayesian network is a statistical model that represents dependency relationships between multiple variables: it is used to analyze and infer conditional probabilities between variables, allowing uncertainties and causal relationships in a system to be evaluated. It is widely used not only in the world of artificial intelligence and risk analysis, but also for medical diagnostics and prediction of events, such as natural disasters.

In this specific case, the causal model allowed the study of hydrogen behavior in an enclosed environment and the still unknown dynamics related to fire and explosion prevention, for example, in tunnels and nonlinear sections of the line.

In particular, the framework used made it possible both to study the possible impact of unknowns on the scenarios under analysis and to depict an “on paper” risk assessment that allows for better optimization and planning of any additional in-depth studies (simulations, studies, test sets, etc.), which are typically very expensive computationally, economically and in terms of timing.

This demonstrates how artificial intelligence can be applied even in situations where data are missing or partial. This innovation in the use of artificial intelligence supports human vision and planning, going beyond what has been known so far and opening up new perspectives, both for the world of logistics and transportation and for the general progress of society.