Machinery malfunction problems are sources of increased maintenance costs and disturbances in production activity across industry. Data mining technology could extract the unknown natural rules from the amount incomplete noise, fuzzy and random practical information. An Intelligent fault diagnosis framework based on data mining is presented. The framework includes the signal processing, the fault diagnosis and knowledge acquisition based on data mining and the fault diagnosis expert system. The knowledge acquisition method based on I_RIPPER (improvement RIPPER Repeated Incremental Prunning to Produce Error Reduction) is discussed. The predict results of I_RIPPER are compared with that of SVM (support sector machine) and BP neutral network. It's proved that the I_RIPPER is the best way among them because of the rules style and the advantage of the ability to deal with the text fault features.