This paper proposes a multi-fault diagnosis algorithm combining SOM network with extension theory to meet the requirement that multi-fault modes should be similar and do not contain the standard fault output when SOM network is used for multi-fault diagnosis. First, the training samples are clustered by SOM network, and the fault modes and these clustering centers can be obtained. Second, the dependent function of each feature for each fault mode is set up where the maximum value can be obtained at the clustering center. Next, the evaluation index of multi-fault modes is designed for multi-fault diagnosis, which is based on the dependent function values of features. Finally, the spectrum data of vibration signal of steam turbine generator unit is adopted to verify the algorithm. The results show that both single-fault mode and multi-fault modes can be correctly distinguished by this method, so the algorithm is feasible.

multi-fault diagnosis  /  SOM network  /  extension theory  /  dependent function