Depth basis vector coding network and part fault early warning method

The invention discloses a deep basis vector coding network and a part fault early warning method, and belongs to the field of fault diagnosis and equipment health operation and maintenance. The depth base vector coding network is composed of a compression encoder, a depth base vector coding module and an expansion decoder. The depth base vector coding module codes the depth feature vectors into a linear combination of depth base vectors. The invention relates to a part fault early warning method, which comprises the following steps of: training a deep base vector coding network by using a data sample at the early service stage of a part, and combining a trained compression encoder and a deep base vector coding module into a deep linear decomposition machine; in the continuous service process of the part, inputting a monitoring data sample into the depth linear decomposition machine, and representing the data sample as a coordinate point of a depth linear space; and finally, judging whether the data samples ar