A method of combining the rough sets and neural network based on the condition attributes discretization and reduct ion algorithm is proposed to fault diagnosis. Firstly the method for optimizing Naive Scaler breakpoint set is presented to discrete the decision table

and then the discernibility matrix and function are used to get the minimum attribute reduct ion set. Finally

the neural network is applied to fault diagnosis on JZQ-250 gearbox

and comparing the diagnosis results of the characterist ic set before reduction with that after reduction

the experiments show that the rough-neural network can reduce the network structure

and has the powerful fault tolerance and antijamming capability with the feature of less iteration

faster convergence rate

higher diagnostic accuracy

which is an effective method for the gearbox fault diagnosis.