Digital and Data-Driven Maintenance of Switching Systems

To facilitate the transition to data-driven maintenance strategies, a holistic sensor concept is being developed that is informative, robust, cost-effective, requires as little maintenance as possible, and is easy to implement during ongoing operations. It combines sensor technology for all three switch maintenance disciplines: acceleration and vibration sensors, as well as acoustic, laser, and environmental sensors for the track superstructure; current, pressure, and limit position sensors for the LST; and temperature, humidity, and power measurements for electrical systems and switch heating.

The development work is carried out iteratively: Based on the FMEA from WP1, an initial sensor selection is made and validated in operation. A baseline is established by comparing the data with conventional inspection data; deviations between sensor diagnoses and actual malfunctions are fed back into the analysis via a feedback loop, leading to continuous refinement of the sensor selection, algorithms, and threshold values. In this way, the system gradually evolves from supplementary monitoring to demand-driven maintenance.

The AXO X1 sensor system, which measures triaxial acceleration, temperature, and humidity, serves as a practical reference solution.

The digital maintenance concept combines the collection of relevant data and metrics with the detection, prediction, and assessment of failures, the optimized planning of corrective actions, as well as the management of their implementation and evaluation of their impact. The solutions can be technological, such as diagnostic and predictive methods, sensor technology, or mobile tools, and organizational, such as process improvements, standardization, and a strategic mix with conventional methods.

The starting point is DB InfraGO’s current, scheduled preventive maintenance program. This approach is highly predictable and compliant with regulations, but it does not always take the actual condition of the infrastructure into account and requires a significant allocation of resources in the event of individual failures.

In the next step, the data-driven approach will be implemented in parallel with the existing regulatory framework: Interval-based maintenance will be fully retained, while continuous condition data will also be analyzed. This reduces unplanned corrective interventions, enables prioritization within the specified intervals, and builds trust among staff through the direct comparison of both approaches. In the long term, the goal is predictive planning, in which the optimal maintenance time is determined based on the remaining useful life (RUL). Trend analyses, anomaly detection, degradation models, and AI-based forecasts provide the foundation; feedback loops after each intervention continuously adjust sensor selection, threshold values, and planning models, thereby creating a robust basis for decision-making in operational, tactical, and strategic planning.

To implement the concept in practice, migration scenarios that vary by location and time are being developed and evaluated. On the high-performance network, fully renovated corridors (Berlin-Hamburg) equipped with new switches are ideal for an early rollout, as extensive sensor systems can be installed there and the condition of the infrastructure can be monitored starting from a defined baseline. In the supplementary network, cost-effective sensors that can be retrofitted quickly hold particularly high potential, because the advanced age of the infrastructure, poor accessibility, and limited budgets there often result in high disruption costs.