The increasing demand on railway transportation due to social, economic and demographic factors has triggered an evolution in preventive maintenance strategies towards more optimized and cost-effective processes. Within this context related to the increasing interest in predictive maintenance strategies, commercial trains are being equipped with both positioning and communication systems as well as onboard intelligent sensors monitoring various subsystems all over the train, thus providing real-time flow of information that is transferred to centralized data servers via wireless technology. This information consists of georeferenced alarms, called events, along with their spatial and temporal coordinates. Alstom Transport has concieved TrainTracer TM, a state-of-the-art software which allows to collect and process real-time data sent by fleets of commercial trains equipped with onboard sensors monitoring various subsystems such as the auxiliary converter, doors, brakes, power circuit and tilt systems. This data consists of series of timestamped events where each event is identified by a numerical code in addition to context variables describing the physical, geographical and technical framework of the train at its exact time of occurrence. This paper aims to investigate how a data mining approach can be applied to discover significant co-occurrences between pairs of events. Once identified and scrutinized by various metrics, these co-occurrences are then used to derive temporal association rules that can be used to perform an on-line analysis of the incoming event stream in order to predict and alert the imminent occurrence of severe failure events, i.e., failures requiring immediate corrective maintenance actions, also called target events. The proposed methodology is based on different randomization null models applied on temporal data sequences.