Research on Ferrographic Image Fault Diagnosis Based on Channel Overlapping Technique and Information Fusion MechanismSource: Journal of Tribology:;2024:;volume( 146 ):;issue: 007::page 74601-1DOI: 10.1115/1.4064858Publisher: The American Society of Mechanical Engineers (ASME)
Abstract: Utilizing computer technology to realize the application of ferrographic intelligent fault diagnosis technology is a foundational investigation to oversee the operations of mechanical equipment. To continuously improve the accuracy of artificial intelligence recognition, the complexity and computation of the model will be increased. The proposal of the transformer model (the core technology of chatgpt) has fundamentally changed the intelligence level of artificial intelligence, but it has also greatly increased the demand for computer computing power. What's more, it is difficult to equip industrial quality inspection sites with high computing power computers. The channel overlapping technique developed in this paper is a technology to segment the three channels of image information and reserve overlapping areas for an information communication mechanism. With this mechanism, the model location channel overlapping convolutional neural network can obtain high recognition accuracy by using only one-half of the original training computing power. When channel overlapping combines with no position information, information fusion is formed. The model channel overlapping technique fusion convolutional neural network established by the information fusion mechanism will get a higher prediction accuracy through joint training with the original image. However, the computation consumption is nearly one-third of the pure traditional convolutional neural network algorithm.
|
Collections
Show full item record
contributor author | Fei, Xie | |
contributor author | Haijun, Wei | |
date accessioned | 2024-04-24T22:47:44Z | |
date available | 2024-04-24T22:47:44Z | |
date copyright | 3/25/2024 12:00:00 AM | |
date issued | 2024 | |
identifier issn | 0742-4787 | |
identifier other | trib_146_7_074601.pdf | |
identifier uri | http://yetl.yabesh.ir/yetl1/handle/yetl/4295889 | |
description abstract | Utilizing computer technology to realize the application of ferrographic intelligent fault diagnosis technology is a foundational investigation to oversee the operations of mechanical equipment. To continuously improve the accuracy of artificial intelligence recognition, the complexity and computation of the model will be increased. The proposal of the transformer model (the core technology of chatgpt) has fundamentally changed the intelligence level of artificial intelligence, but it has also greatly increased the demand for computer computing power. What's more, it is difficult to equip industrial quality inspection sites with high computing power computers. The channel overlapping technique developed in this paper is a technology to segment the three channels of image information and reserve overlapping areas for an information communication mechanism. With this mechanism, the model location channel overlapping convolutional neural network can obtain high recognition accuracy by using only one-half of the original training computing power. When channel overlapping combines with no position information, information fusion is formed. The model channel overlapping technique fusion convolutional neural network established by the information fusion mechanism will get a higher prediction accuracy through joint training with the original image. However, the computation consumption is nearly one-third of the pure traditional convolutional neural network algorithm. | |
publisher | The American Society of Mechanical Engineers (ASME) | |
title | Research on Ferrographic Image Fault Diagnosis Based on Channel Overlapping Technique and Information Fusion Mechanism | |
type | Journal Paper | |
journal volume | 146 | |
journal issue | 7 | |
journal title | Journal of Tribology | |
identifier doi | 10.1115/1.4064858 | |
journal fristpage | 74601-1 | |
journal lastpage | 74601-12 | |
page | 12 | |
tree | Journal of Tribology:;2024:;volume( 146 ):;issue: 007 | |
contenttype | Fulltext |