In view of the lack of interpretability and weak anti-noise ability of current intelligent diagnosis models in fault diagnosis of rotating machinery

an interpretable diagnosis model based on the discrete multi-wavelet transform and convolution layer fusion is proposed. Firstly

the discrete multiwavelet layer (DMWL) is constructed by using the discrete multiwavelet filter

and the input vibration signal is decomposed into multiple frequency component signals by using the multiwavelet filter

and then the information of each frequency band is learned with the one-dimensional convolution layer to realize the fault information location of the model. In order to enhance the ability of discrete multi-wavelet transform convolutional neural network (DMWT-CNN) model to learn the information of each frequency component

the frequency attention mechanism (FAM) was introduced. Finally

it is verified in two kinds of rolling bearing data sets and a variety of diagnostic models. The research results show that the model can locate and detect fault information

and the diagnostic accuracy reaches 90% under the interference of –6 dB Gaussian white noise

which is higher than other diagnostic models

and has interpretability and good robustness.