FFCNN-SVM transfer learning fault diagnosis method based on feature fusion under small sample

Abstract

The invention discloses an FFCNN-SVM transfer learning fault diagnosis method based on feature fusion under a small sample. The method comprises the following steps: migrating a mature model in a source domain to a target domain through a model migration method in migration learning to form a preliminary model of the target domain; then, by utilizing a convolutional layer and extracting characteristics of picture features, adding the convolutional layer on the preliminary model, then training the preliminary model by utilizing a small amount of scarce sample data provided by the target domain, and forming a target domain shallow model after fitting; replacing a full connection layer of the CNN with the SVM to achieve a classification effect. The method is advantageous in that: through a bearing fault data set, the new fault diagnosis performance of the method can be well verified; experimental results show that compared with other transfer learning methods, the method provided by the invention has a better fault diagnosis effect.

Description

technical field [0001] The invention belongs to the field of fault diagnosis, in particular to an FFCNN-SVM migration learning fault diagnosis method based on feature fusion under small samples. Background technique [0002] As a key component in industrial production equipment, rolling bearings will directly affect the safe operation of the entire mechanical equipment and lead to high maintenance costs and economic losses once they fail. Therefore, early and accurate detection of these faults is critical to the operational safety of modern manufacturing systems. In addition, it can also guarantee the reliability and safety of the control system by maintaining it accordingly, properly maintaining the strategy, etc. Due to the complexity of fault types, noisy signals from field sensors, and lack of fault samples, early and accurate fault diagnosis is still very challenging and is an active research area. [0003] In the past few years, deep learning based methods have achie...

AI Technical Summary

Problems solved by technology

However, the features learned by a pretrained network from a large amount of unlabeled data may not be so easily distinguishable, because it is difficult to reveal the intrinsic characteristics of faulty samples when they are so scarce and overwhelmed by normal samples

Images