Aimed at the puzzle of knowledge acquisition in intelligent decision-making technologies, the Rough Set Theory is used as a tool to implement the faults knowledge discovery and a decision-making table with symbols data is designed as a discussed case. By the data operation to the table, both a set of knowledge particles and a group of knowledge rules are obtained orderly. And the model of knowledge acquisition by means of data-driven way is built. It shows out the essence of knowledge discovery that the knowledge in an original knowledge base is combined into many equivalence classes. By experiment simulated the rub-impact faults on a rotor system with two spans, the FFT spectrum characteristics of the rubimpact responses are put forward to a new demand for how to get the original faults knowledge. The data structure of a decision table is a key factor. Because the table is used to save the original fault knowledge from plant site, therefore its performance must be very good to suitable for knowledge discovery. So,it should be designed out firstly. The experiment indicates that to build an effective decision table is a hard task with certain difficulty. To define the exact universe of discourse and build the attribute sets to describe the original faults knowledge as well as discrete the data resources into symbols data set are included in the task. The conclusions are that it should be resolved to save the original faults knowledge with data mode in the guidance of the ideas of knowledge discovery system engineering. And the mode must be structured for a specific rotating machinery unit as objects in categories.