Abstract: For the knowledge acquisition from the data table recorded faults cases with continuous-value features, the discretization-mapping scheme translating the table into the generalization information system of rough set theory is investigated. The direct k-means algorithm of the clustering analysis is improved. The amount of faults confirmed by experts in the table is set as the classification amount k and the set of indexes of mean clustering centers ordered by sort ascending is set as mapping function. On the basis of the partition universe with k, the discretization scheme to the table is proposed. The discrete symbol denoting a continuous value is the same as the symbol of the center provided with the nearest distance between the value and a group of clustering centers. By the process, a normal table accord with the pattern of rough set theory is acquired and a few decision-making rules are extracted. The rules show that the scheme has the performances optimizing the partition points and rejecting the outside fluctuation. They reveal the generalization characters of the faults and can be used to construct and to extend the knowledge database of the fault diagnosis devices of a rotor-bearings system.

Key words: Attributes discretization, Cluster analysis, Fault diagnosis, Knowledge acquisition, Rough set