Train Crash Energy Management Optimization via Machine Learning

The patent pre-processes crash simulation data and trains machine learning models in advance to establish prediction capabilities. By performing the computationally intensive finite element simulations beforehand to create training datasets, the system can then quickly predict crash responses for new scenarios without running full simulations, thus resolving the time consumption issue while maintaining accuracy through the pre-trained models.

The patent replaces the traditional mechanical simulation system (nonlinear finite element analysis) with a machine learning-based prediction system. By training neural networks on simulation data, the system substitutes the computationally expensive mechanical simulation process with a faster data-driven model that can predict crash responses in real-time while preserving the accuracy benefits of detailed simulations.

The patent implements a nested modeling approach where simplified dynamical models are embedded within a more comprehensive framework. The simplified models provide initial predictions, which are then refined using data from eight-marshalling train simulations that capture the full complexity of real train crash behavior. This nested structure allows the system to maintain computational efficiency while progressively improving accuracy through multiple levels of modeling fidelity.

The patent segments the crash analysis process into multiple stages: initial prediction using simplified models, intermediate refinement using partial train configurations, and final adjustment using complete eight-marshalling train data. This segmentation allows computational efficiency at each stage while cumulative accuracy improvement ensures the final results accurately reflect real train crash responses.

The patent applies local quality optimization by adjusting energy absorption characteristics at different crash interfaces independently. Rather than using a uniform CEM approach, the system modifies structural parameters, material properties, and energy management device configurations at specific locations along the train to achieve optimal energy distribution. This allows each crash interface to have tailored energy absorption properties that collectively balance the overall energy distribution across all interfaces.

The patent introduces dynamic adjustment of energy absorption characteristics across different crash interfaces. By making the energy management system adaptive and configurable, the system can dynamically allocate energy absorption capacity to different locations based on real-time or scenario-specific requirements, enabling balanced energy distribution while maintaining effective front impact energy absorption through programmable or adjustable energy management devices.