Multidimensional Edge Perception Model for Rail Vehicle Operational States Based On Artificial Intelligence of Things

In the domain of Prognostics and Health Management (PHM) for intelligent rail vehicles, real-time multidimensional perception is crucial for vehicle monitoring. However, achieving such a perception of low-cost, computationally limited Internet of Things (IoT) devices presents a significant challenge. Given the lack of effective IoT multidimensional perception models and the surging demand for PHM data analytics, this study proposes a non-invasive multidimensional Artificial Intelligence for Internet of Things (AIoT) perception model to improve vehicle performance and predictive maintenance. The model uses the Tiny Machine Learning approach to deploy a lightweight model on the edge devices of rail vehicles, which intelligently recognizes the vehicle operational states in real-time by monitoring multidimensional data such as acceleration and tilt angle, and transmits the resulting data to the IoT cloud for fusion and classification statistics. Experiments conducted in a metro environment show that the model can recognize nine complex operational states in both real-time and offline modes with an accuracy rate of more than 97%, which is significantly better than the traditional multilayer perceptron (MLP) model. The model's two-axis recognition outperforms single-axis and three-axis methods and exhibits strong robustness under vibration conditions. Its versatility allows extension to different sensors and fault state detection and can be applied to intelligent condition monitoring in various transportation and machinery systems.