Adif is responsible for the management and maintenance of an extensive railway network comprising over 12,000 km of conventional tracks and 4,000 km of high-speed lines in Spain. In accordance with its asset management policies, the company maintains a comprehensive inventory database characterizing each element of the infrastructure, tracks, and industrial systems such as power, signaling, and telecommunications. Additionally, adhering to its maintenance policies and current regulations, Adif conducts regular inspections and surveys of the tracks to monitor the network’s quality, enabling the identification of corrective maintenance needs and the planning of preventive maintenance based on the condition. Moreover, for over a decade, the company has been documenting incidents occurring along these nearly 20,000 km of the network, detailing their nature, causes, and impact on assets. This extensive set of historical data becomes a valuable source of information that, when properly analyzed using advanced analytics and machine learning tools, can optimize the management and maintenance of the network.

INES, in collaboration with Ardanuy and with the participation of McKinsey, has worked for a year and a half on developing an advanced analytics tool for predictive maintenance. This system predicts the likelihood of failure for each asset in the network based on historical failures, estimates its criticality, and calculates its risk, aiming to optimize cyclic maintenance tasks and replacements to increase track availability and ensure user safety.

To achieve this, the team collected, cleaned, and processed data from various Adif systems, preparing it to feed machine learning models. Training the models required close collaboration between the data and engineering teams. Optimization algorithms were applied for planning maintenance activities and replacements based on a cost-benefit analysis. The entire process was integrated into a visualization tool that allows Adif to leverage the results and make informed decisions.

The risk-based replacement planning will reduce the network risk by 11% and unavailability by 30%, one of the key objectives of the project. Meanwhile, optimizing cyclic activities will achieve a 38% risk reduction and a 10% reduction in unavailability.

Additionally, knowledge transfer to the Adif designated team was conducted through courses and training sessions to empower them for the future autonomous use of the developed tools.