Intelligent fault diagnosis is important for improving the reliability of smart manufacturing. Fault diagnosis methods based on deep learning have been very successful in industry, but there are some differences in the features extracted by different models. A fusion network model (CLOD) based on deep learning is proposed for problems such as incomplete data feature extraction. Firstly, a time-frequency analysis of the fault signal is carried out by Fourier transform to obtain a time-frequency spectrum sample, then the sample is fed into a convolutional network model (CLOD) trained and learned after the fusion of LSTM model and improved CNN model, and finally the model performance is improved by updating the network parameters to achieve accurate and intelligent diagnosis of bearing faults. Compared with the traditional method, CLOD greatly increases the fitting speed and stability of the model on the basis of guaranteed accuracy.