The state signal of the fuel regulator in the weak state presents strong fuzziness in the space, it is difficult to effectively extract its state characteristics and carry out accurate fault diagnosis. Therefor, based on the combination of unsupervised clustering algorithm and multi-source information fusion technology, a centroid scale change method for state feature extraction was proposed. The unsupervised clustering and information fusion algorithm were used to extract the feature of the state signal aiming at the pressure fluctuation of the fuel regulator outlet assembly. The scale change criterion was designed to reconstruct the extracted features. In Linux environment, Java language was used to write feature extraction algorithm, feature reconstruction algorithm and K-NN classification algorithm, and these algorithms were tested. The results show that the reconstructed feature data after centroid scale change presents independent distribution in the new feature space, and by using K-NN classification algorithm, the input data can be effectively classified.The proposed method provides reference for the fault diagnosis system construction of the fuel regulator.