Infrared image monitoring algorithm for converter valve equipment

Traditional converter valve monitoring methods have problems of inefficiency and failure to detect potential faults in time, which may lead to equipment damage or system failure. In this study, according to the requirements of infrared image monitoring of the converter valve, the infrared image moni...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:AIP advances Ročník 15; číslo 3; s. 035127 - 035127-11
Hlavní autoři: Dai, Jiashui, Tan, Fali, Jiang, Yi, Liang, Yunlong, Lin, Kangquan, Tong, Kai
Médium: Journal Article
Jazyk:angličtina
Vydáno: Melville American Institute of Physics 01.03.2025
AIP Publishing LLC
Témata:
Algorithms
Fault detection
Infrared imagery
Monitoring
Target detection
ISSN:2158-3226, 2158-3226
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Traditional converter valve monitoring methods have problems of inefficiency and failure to detect potential faults in time, which may lead to equipment damage or system failure. In this study, according to the requirements of infrared image monitoring of the converter valve, the infrared image monitoring algorithm system of the converter valve equipment based on the improved YOLOv8n target detection model is constructed. The normalized data are combined with the time and temperature data and input into the BiLSTM model for training. According to the high requirements for the target detection algorithm in the converter valve scenario, lightweight PConv convolution is introduced to improve the C2f module of YOLOv8n, and Slim-neck improvement of the feature fusion layer is made. At the same time, the Inner-MPDIoU loss function is introduced so that the overall model is more lightweight and more accurate. In the corresponding converter valve infrared image dataset, mAP can reach 87.8%, which is 2.5% higher than the original model, and the number of algorithm parameters is also lower, reducing by 18.8%.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2158-3226
2158-3226
DOI:10.1063/5.0237028