Prior Training of Data Mining System for Fault Detection

Many approaches have been used to fault discovery in complex systems. Model based reasoning; data mining analysis; rule base methods are the few among those approaches. To be successfully applied, these approaches all have to have some knowledge about the system prior to faults detection during the system run. Fault Tree Analysis shows the possible causes of a system malfunction by enumerating the suspect components and their respective failure modes that may have induced the problem. The rule based inference build the model based on the expert knowledge. Those models and methods have one thing in common; they have presumed some prior-conditions. Complex systems often use fault trees to analyze the faults. Fault diagnosis, when an error occurs, is performed by engineers and analysts performing extensive examination of all data gathered during the mission. International space station (ISS) control center operates on the data feedback from the system and decisions are made based on threshold values by using fault trees. Since those decision-making tasks are safety critical and must be done promptly, the engineers who manually analyze the data are facing the challenge of time limit. To automate this process, this paper presents an approach that uses decision trees to discover faults from data in real-time and capture the contents of fault trees as prior knowledge and use them to set the initial state of the decision trees.