The invention discloses a motor fault diagnosis method and
system fusing order analysis and sparse sensing technologies, and aims at solving the problems that an existing model is poor in generalization ability and lacks
interpretability, a vibration
signal is resampled into an angular domain stationary
signal, and an angle-order spectrum is generated; the method comprises the following steps of: constructing a
gate control order Transform diagnosis model consisting of a Transform
encoder containing an information
bottleneck order
gate control module (IB-OG) and a fault classifier, namely embedding the IB-OG module in a self-attention mechanism of the
encoder, carrying out sparse screening on input order characteristics by utilizing a built-in microstochastic
gate control mechanism, removing redundant
noise and reserving key fault characteristics. In addition, class condition domain invariance constraints are introduced into the
loss function, and the model is forced to align feature distribution of different working conditions in the same
fault class. In conclusion, the diagnosis precision, the generalization ability and the physical
interpretability of the model under the complex working condition are remarkably improved.