Exploiting spatial and temporal dependencies to enhance fault diagnosis: Application to railway track circuits

In many practical applications, it is not feasible to measure a large number of variables. Therefore, strategies are required to enhance fault diagnosis, given the available monitoring signals. In this paper, we consider fault diagnosis in networks using a limited number of monitoring signals. We propose to use spatial dependencies between the monitoring data of the subsystems to discriminate between faults. Furthermore, the temporal properties of the monitoring signal are exploited. It is shown that, for a track circuit example, the spatial and temporal dependencies are valuable for diagnosis. Based on these features, an approach is proposed for fault diagnosis in the presence of environmental disturbances.