With the advent of the era of big data， the mechanical equipment fault diagnosis method based on deep learning has at? tracted more attention. However， the traditional deep network model seriously limits its application in practical engineering due to the excessive amount of parameters and calculations. Based on this， a GhostConv lightweight network model is proposed and used for fault diagnosis. GhostConv generates a small part of the feature maps through conventional convolution， and performs multiple feature extraction on the generated feature maps to generate the remaining feature maps. Contact the feature maps of the two parts to obtain a complete feature map. GhostConv structure saves the cost of generating redundant feature maps in conventional convolu? tion to the maximum extent， and reduces the model parameters to ensure the performance of the model. In the experiment， the con? tinuous wavelet transform is used to transform the vibration signal to generate a two-dimensional time-frequency diagram， and then the designed GhostConv is used to establish a lightweight fault diagnosis network model. The original dataset and noisy dataset of Case Western Reserve University are used for experimental verification， and compared with the conventional convolution structure network model and depth separable convolution structure model in terms of parameters， calculation and recognition rate. The ex? perimental results show that the GhostConv lightweight network model still has high recognition accuracy and strong anti-noise abil? ity under the condition of fewer parameters and calculations with good robustness and generalization ability. The parameters of the model are only 6% of the conventional convolution model and 56% of the deep separable convolution model. Under the condition of strong noise interference， the fault diagnosis and recognition rate is still higher than that of the conventional convolution model， which confirms its engineering application value.