Deep prototype network-based few-sample bearing fault diagnosis method

Abstract

The invention discloses a few-sample bearing fault diagnosis method based on a deep prototype network, and relates to the technical field of bearing fault diagnosis technologies, and the method comprises the steps: support set prototype calculation: employing an improved k-means + + clustering algorithm to calculate a prototype of each fault category in a support set; discriminating and classifying the query set: applying the support set prototype to discriminating and classifying the query set; aggregating query set samples; constructing prototype loss; and testing a test set sample: after the test set sample is processed by the updated feature extractor, carrying out distance measurement on the test set sample and the prototype of each known fault category. According to the method, the prototype of each known class is calculated by using a feature clustering algorithm k-means + +, and the marked target sample is allocated to the closest prototype class by calculating the Euclidean distance from the marked target sample to the prototype. The prototype loss is constructed to enhance the compactness between the fault marking sample and the corresponding prototype, so that the known fault category sample can be identified, and the fault sample from the unknown category can be effectively eliminated.

Description

technical field [0001] The invention relates to the technical field of fault diagnosis of bearing vibration signals, in particular to a method for diagnosing faults of bearings with few samples based on a deep prototype network. Background technique [0002] In modern industrial production, bearings play the role of power transmission. As the main rotating components, they are prone to failure under continuous working conditions, causing damage to equipment, economic losses or casualties. In order to ensure the stable operation of machinery and avoid the occurrence of faults, intelligent fault diagnosis methods are widely used in bearing health monitoring. [0003] In traditional fault diagnosis methods, the training and test samples generally come from the same working condition, and the fault types are known. In actual situations, due to the influence of uncertain factors such as changes in working conditions, sometimes unexpected and unknown failures of machinery occur, ...

AI Technical Summary

Problems solved by technology

In addition, in practical situations, it is difficult to obtain a large amount of labeled data for training models

Images