Knowledge representation is the necessary process to build an autonomous fault diagnosis expert system for satellites. In order to enhance the knowledge representation ability of time-related rules and r-of-n rules, an improved fault Petri net model for knowledge representation is proposed on the basis theory of fault Petri net model. The alarming threshold function and the fire threshold function is introduced, the alarming conditions and result is given in the form of definition, similarly, the fire conditions and result is redefined. A transition gives an alarm for some time and then could fire, the improved fault Petri net model could present the time-related rules while the fault Petri net model couldn t. Further more, the improved fault Petri net model could present the r-of-n rules much more conveniently than the fault Petri net model, it needs much fewer transitions to present the same rule. The improved fault Petri net model upgrades the knowledge representation abilities and provides basic conditions for the knowledge representation of the diagnosis expert system.