Researchers have developed a novel multimodal method for diagnosing faults in train bogie motors, combining physics-inspired regularization with an enhanced convolutional neural network (ConvNeXt) architecture. This advancement is crucial for the safety and efficiency of railway transportation, as it enables the detection of anomalies in traction motors, critical components that operate under demanding conditions and are prone to complex, multifactorial failures.

The proposed method addresses the limitations of traditional approaches, which often lack the ability to effectively integrate data from multiple sources or to capture the inherent complexity of physical systems. By incorporating physical principles into the regularization process, the model not only improves its generalization capability but also imbues the neural network with an intrinsic understanding of motor behavior. This results in more accurate and robust diagnostics, even in scenarios with incomplete or noisy data.

The enhanced ConvNeXt architecture, adapted to process multimodal data (such as vibration, current, and temperature signals), allows for more efficient extraction of relevant features. The integration of physics-inspired regularization acts as a bridge between deep learning and physical models, optimizing the detection of subtle patterns that indicate the onset of a fault. Preliminary results show a significant improvement in diagnostic accuracy and reliability compared to existing methods.

This development has direct implications for predictive maintenance in the railway industry, enabling earlier interventions and reducing unplanned downtime. The ability to predict and locate faults with greater anticipation and precision not only optimizes operational costs but also raises safety standards for passengers and cargo. Future research is expected to explore the application of this approach to other complex mechanical systems and its validation in large-scale operational environments.