Existing test models for hierarchical modeling of complex equipment require not only failure mode， effects and criticality analysis （FMECA） information for each layer of equipment，but also the relationship between failure modes， which increases the actual workload and modeling difficulty， and is disconnected from the actual fault diagnosis. In order to solve the above problems，a modeling method based on Petri nets is proposed，which combines testability models with fault diagnosis models. Stochastic Petri nets are used to establish test systems at the equipment system level， and fuzzy Petri nets （FPN） are used to establish subsystem fault diagnosis model，completing the transfer of system to subsystem. Fault statistics are processed according to FMECA information，the parameters are studied and optimized by neural network. The forward inference is used to predict the fault accurately，reverse inference and minimum cut set are used to locate the fault. Finally，the turbofan engine fan component model is taken as an example to perform inference analysis，and the correctness and rationality of the inference model is verified through fault trees and statistical data.