Research on bearing fault diagnosis based on improved dense connection network

In the fault diagnosis of rolling bearings, a limited number of samples can lead to insufficient training of deep learning models, making it difficult to accurately diagnose the fault types in a timely manner. Data augmentation on one-dimensional vibration signals often suffers from issues such as generating non-intuitive and low-quality data, making it challenging to ensure model performance. To address the problem of small sample-based fault diagnosis in deep learning, this paper proposes a rolling bearing fault diagnosis method that combines conditional deep convolutional generative adversarial networks (cDCGAN) with an efficient dense connection network (Efficient Densenet). Firstly, vibration signals are transformed into two-dimensional grayscale images as input to the model. The sample data is fed into the conditional deep convolutional generative adversarial network, and the trained G generator is used to generate reliable new grayscale image samples to augment the original dataset, ensuring dataset diversity. Finally, the augmented dataset is input into the efficient dense connection network for fault diagnosis, and the diagnosis results are visualized to obtain a confusion matrix. The classical Case Western Reserve University dataset (CWRU) is used as the validation dataset, and a small sample of data is extracted for experiments. The experimental results demonstrate that the dataset augmented by the cDCGAN network can effectively improve the fault diagnosis performance of the model, achieving high accuracy even with a small sample size. Comparative experiments validate the enhancement of the proposed method on the model’s diagnostic capability.