Fault prediction is a core problem for predictive maintenance. Predictive maintenance is a new branch in maintenance research field, and it can cut down a large number expenses that cost in prevent maintenance, and reduces the time of stop machines. The key problem of fault prediction is to acquire state data of the devices continuously. Build the system state reliability models according to the state data, and to determine the devices whether failure or not in next time via the models. In the paper, the principle of fault models building and the method of state data mining is discussed in details, fault prediction method and the key problems are described in details too.
