Knowledge-guided CNN-based small sample similar abrasive particle identification method

Abstract

The invention discloses a knowledge-guided CNN-based small sample similar abrasive particle identification method, which comprises the steps of marking key features of an abrasive particle height graph in a binary graph form according to an abrasive particle generation mechanism; on the basis, constructing a U-net network of a VGG16 model to automatically extract typical characteristics of the abrasive particles; fusing the output of the U-Net network with the convolution layer of the full convolution CNN network through a weighting mode, guiding the training of the full convolution CNN network, and enabling the full convolution CNN network to quickly locate the distinctive features of similar abrasive particles. According to the constructed network model, the weighted sum of Focal loss and binary classification cross entropy loss is adopted as an overall loss function, parameter training is conducted through an SGD optimization algorithm, a final similar abrasive particle classification model is obtained, and identification of typical similar abrasive particles is achieved. According to the method, the abrasive particle knowledge experience and the CNN network are effectively combined, and the problems that in the existing abrasive particle analysis field, the number of similar abrasive particle samples is small, and the recognition accuracy is low are solved.

Description

[0001] technical field [0002] The invention belongs to the technical field of wear particle analysis in the field of machine fault diagnosis, and in particular relates to a small-sample similar wear particle identification method based on knowledge-guided CNN. Background technique [0003] During the operation of mechanical equipment, the relative movement of the friction pair inevitably causes friction and wear. Over time, the original design function of the parts will be damaged or even become invalid. Abrasive particles, as the direct product of wear, record their generation mechanism with complex morphological features, which is an important basis for wear mechanism analysis and wear state monitoring. After years of accumulation, researchers have accumulated a lot of knowledge and experience about abrasive grains, and can accurately identify different types of abrasive grains. With the demand for condition monitoring of intelligent equipment, the traditional wear partic...

AI Technical Summary

Problems solved by technology

However, due to the inherent defects of the method itself, such as: two-dimensional images cannot reflect the three-dimensional shape of abrasive grains; there are many three-dimensional shape parameters; the number of typical abrasive grain samples is small, etc., which reduces the identification accuracy of similar abrasive grains

Images