The complexity of turbo-generator set's structure and it's vibration make its fault to have multigradation and randomness

as well as the fault information being not complete.For this

a method to diagnose fault of turbo-generator set's vibration based on rough set theory and naive Bayesian classificatin algorithm has been put forward.The minmum attribute reduction set is sought by using the rough set

and then the bigger area in which the faults may occur has been diagnosed by using the naive Bayesian classification algorithm

and finally

directing against the concrete fault settings

the said method being validated.Results of practical calculation examples show that the the said method can get better diagnosis result when the fault information isn't complete

and even the kernel attribute being lost

enhancing the fault-tolerance capability of the diagnosis system.