P网络模型。 For the sake of fault diagnosis and maintenance problems of automated manual transmission (AMT), an online fault diagnosis system was established based on CAN bus and PSA (PCA-subtractive clustering-ANFIS) model. The original sample data of AMT fault diagnosis was derived from sensors of electronic control utilities (ECU) and transmitted on CAN bus. Principal component analysis (PCA) was used to reduce correlations and dimensions of input variables. An original fuzzy inference model was built by using subtractive clustering algorithm, on which an adaptive-network-based fuzzy inference system (ANFIS) was based and adopted to build a new type fault diagnosis model. Experimental result showed that combined with CAN bus，PSA model is effective, and both of its fitting ability and convergence speed are superior to PCA-BP network model.