|
|
|
| Identification of Pipeline Inner Wall Geometry Based on the PCA-BP Neural Network |
|
|
|
|
Abstract Pipelines, such as those used for petroleum and natural gas transportation, are subject to various factors such as temperature, pressure and corrosion during long-term operation. These factors can lead to structural defects within the pipelines, including collapse, deformation, rupture and corrosion. Thus, a pipeline inner wall geometric identification method based on principal component analysis (PCA) and back propagation (BP) neural network is proposed to tackle these problems. Firstly, an automated modeling technique for the adaptive inner wall of the pipeline is adopted using CAD secondary development technology, enabling the rapid acquisition of a large number of randomly generated geometric models. Subsequently, these models are subjected to two-dimensional constant magnetic field finite element analysis to obtain magnetic field response data of specific measurement points. This adaptive sample library generation system reduces the cost required for modeling and simulation while providing necessary data foundation for the identification of pipeline inner wall geometry. Meanwhile, the PCA method is utilized to reduce the dimensionality of the magnetic field response data. This dimensionality reduction technique effectively eliminates redundant information and improves the efficiency of the subsequent training process. Finally, a mapping relationship between the magnetic field responses of the measurement points and the geometric parameters of the pipeline wall is established using a BP neural network which optimized by the Levenberg-Marquardt algorithm. This mapping relationship enables the identification of the geometric shape of the pipe wall based on the magnetic field response data. The results of numerical examples demonstrate that the PCA-BP neural network-based pipeline inner wall geometric inversion analysis model can accurately and rapidly predict the geometric shape of the pipeline inner wall. Even for complex geometries with varying degrees of random errors, the proposed method exhibits strong identification performance. This work provides a new way to address structural defects within pipelines and offers an effective tool for pipeline safety assessment and health management.
|
|
Received: 24 March 2023
Published: 23 October 2023
|
|
|
|
|
|
|
2023, Vol. 44 
