Rotary machinery fault diagnosis method based on zero trial learning and potential space coding

The invention discloses a rotating machine fault diagnosis method based on zero trial learning and potential space coding, the zero trial learning gets rid of the dependence of traditional deep learning on a large amount of labeled data, fault diagnosis of unknown samples can be realized by using a small amount of labeled data, and a potential space coding model is a new coding-decoding method. According to the method, visual information and semantic information are interacted in a potential space by applying an encoding-decoding thought, and the potential space is shared. According to the method, a large amount of annotation data is not needed for training, and the cost loss of data annotation is greatly reduced. And learning the similarity between unknown class samples in a test set and known class samples in a training set in a potential space to obtain high-level semantic features of the unknown class of the test set to detect unknown samples. According to the method, the detection performance of the netw