This publication was originally written in Chinese. You're viewing an automated translation into English.
Flow patterns are the different forms or flow patterns of fluids flowing through pipes. Depending on the amount of fluid, flow can be divided into two-phase flow and multiphase flow.
This project aims to distinguish different flow patterns using deep learning and other artificial intelligence technologies. This helps identify adverse flow patterns, such as slug flow and annular flow, and enables timely action to prevent pipeline system failures and minimize losses. This project proposes the YOLOv8_1D model and the YI-Net model, enabling real-time flow pattern recognition. This provides a new solution for flow pattern recognition.
Flow pattern refers to the different forms that a fluid presents when flowing in a pipeline. According to the number of fluid, it can be divided into two-phase flow and multiphase flow.
This is a flow pattern identification project aimed at using artificial intelligence technologies such as deep learning to distinguish different flow patterns. Especially to help identify hazardous flow patterns such as slug flow and annular flow, and take corresponding measures in a timely manner to prevent pipeline system failures and reduce losses. This project proposes the YOLOv8_1D model and YI-Net model, which can achieve real-time flow pattern identification. Provided a new solution for the field of flow pattern identification.
At present, YOLOv8_1D and YI-Netv1 are proposed by improving the YOLOv8 classification network and YOLOv10 related modules.
At present, YOLOv8_1D and YI-Netv1 have been proposed by improving the YOLOv8 classification network and YOLOv10 related modules.
Since the collected pipeline pressure data is one-dimensional, to ensure real-time performance and match the model dimension with the data dimension, the YOLOv8 classification network is converted to one dimension and the convolution kernel size is adjusted to obtain the YOLOv8_1D model.
Due to the fact that the collected pipeline pressure is one-dimensional data. In order to ensure real-time performance and match the model dimension with the data dimension, the YOLOv8 classification network is transformed into one-dimensional and the convolution kernel size is adjusted to obtain the YOLOv8_1D model.
In order to further reduce the number of parameters of YOLOv8_1D, the SCDown module proposed in YOLOv10 is one-dimensionalized into SCDown1d, and this module is used to replace the last two convolution Conv1d in the YOLOv8_1D configuration file.
In order to reduce the redundant calculation of YOLOv8_1D, the CIB module proposed by YOLOv10 is one-dimensionalized into CIB1d, and the Bottleneck layer of C2f1d is replaced by CIB1d to obtain the C2fCIB1d module, which is used to replace the last C2f1d in the YOLOv8_1D configuration file.
Finally, a new one-dimensional model was obtained, named YI-Net , which stands for " One -dimensional Intelligent Network".
This is the first version of the YI-Net network, denoted as YI-Netv1.
In order to further reduce the parameter count of YOLOv8_1D, the SCDown module proposed by YOLOv10 is transformed into one-dimensional called SCDown1d, which replaces the last two convolution Conv1d in the YOLOv8_1D configuration file.
In order to reduce redundant calculations in YOLOv8_1D, the CIB module proposed by YOLOv10 is transformed into one-dimensional called CIB1d, and the Bottleneck layer of C2f1d is replaced by CIB1d to obtain the C2fCIB1d module, which replaces the last C2f1d in the YOLOv8_1D configuration file.
Finally, a new one-dimensional model was obtained and named YI-Net , which full name is " One-dimensional Intelligent Network ".
This is the first version of the YI-Net network, denoted as YI-Netv1.