Fault Diagnosis of Bearings Using MSCPCNN with Limited Samples

In recent years, convolutional neural networks have been widely used in the field of fault diagnosis. However, the features extracted by convolutional neural networks have positional information, and under small sample conditions, the diagnostic accuracy will decrease when the signal shifts. A multi-scale combined pooling convolutional neural network (MSCPCNN) is proposed to address this issue. Firstly, the process of fault diagnosis in convolutional neural networks was analyzed, and a combination pooling layer was designed; Then, a multi-scale convolution method was used to fuse the feature information extracted by convolution kernels of different sizes; Finally, the proposed method was validated on both the Xi'an Jiaotong University dataset and the constructed fault simulation test bench. The results show that the proposed method has better diagnostic performance than the comparison algorithm in limited sample situations, achieving average accuracy of 98.54% and 98.76% respectively in the case of only 10 samples in each class.