A unified diagnosis model based on probabilistic graphical theory was studied. The model constructing method and its variations in different scenarios were discussed. Complex problems such as fault diagnosis in multimode systems, diagnosis with coupling faults, dynamic faults and fault prognosis were solved by using the framework. In order to combine the advantages of the model-based method and the data-driven method, a model learning algorithm was proposed, by means of which the diagnostic result was improved. In the end, the possible model developing directions and research focus were discussed, which can provide a reference for the follow-up theory research.