Graph Convolutional Networks (GCNs) are a novel approach to deal with interconnected data in the form of arbitrary structures. GCNs can be thought of a generalization of Convolutional Neural Networks, which are only able to process fixed data structures. Applications of GCNs have been made in biology and chemistry such as the exploration and identification of molecules structures with state-of-the-art results. This opens possibilities in reliability and maintainability of complex systems such as plants or equipment with different components that interact and affect each other's degradation processes. In this context, we propose a novel GCN based model for the health state diagnostics of trains (in railroad transport system) via the prediction of whether a specific wheel is under excessive load. We introduce a specific adjacency matrix for the model (which defines the connections between nodes of a graph) where connections represent the physical relationship between wheels and let the GCN based model exploit correlations between different parameters encoded in each node characterizing a particular wheel in terms of 23 parameters such as its side of the shaft, current load and train velocity. Based on the results, the proposed GCN based model seems to be a competent and promising approach for health state diagnosis of complex systems and worth further development.