The invention provides a gearbox fault detection method based on deep transfer learning, and the method comprises the steps: collecting original
signal data of a gearbox, carrying out the
noise reduction and data cleaning of an original
signal, carrying out the extraction through a feature extractor, obtaining a new image, building a model, and carrying out the result analysis and
verification of the model through training data and
verification data. According to the method, the technical problem that the type of gearbox fault diagnosis is single is solved, the image data of the vibration signals, the temperature signals and the oil liquid signals of the gearbox are innovatively combined in a crossed mode, multiple sets of composite image data are obtained, a composite
fault model is finally constructed, and the fault diagnosis accuracy is improved. The obtained
convolutional neural network model is migrated to gearbox diagnosis detection data, the migration diagnosis
fault rate is calculated, the diagnosis rate is 94%-100%, high-precision diagnosis is achieved, and rapid fault detection of the gearbox under the composite fault condition is achieved.