A study of neural network-based evaluation methods for pipelines with multiple corrosive regions
Zhiwei Zhang,Songling Li,Huajie Wang,Hongliang Qian,Changqing Gong,Qiongyao Wu,Feng Fan
DOI: https://doi.org/10.1016/j.ress.2024.110507
IF: 7.247
2024-09-21
Reliability Engineering & System Safety
Abstract:In recent years, significant developments have been made in methods for assessing the remaining strength of corroded pipelines. However, existing methods have limitations as they mainly focus on the local impact of corrosion defects. This study explores evaluation methods using neural networks to predict the ultimate resistance of pipelines containing multiple corrosive regions. Firstly, based on the validated method, the study generates a dataset comprising 3,000 corroded pipeline models and pixelates the corrosion information of these models via digital images. Then, three neural network evaluation frameworks are constructed: a Multilayer Perceptron (MLP) using the overall corrosion matrix, an MLP based on corrosion feature parameters, and Convolutional Neural Networks (CNN) based on corrosion images. Following this, the study analyzes the relationship between various corrosion parameters and failure pressure, compares the training effectiveness of the three neural network methods, and validates the accuracy and applicability of the proposed approach. The results indicated that various corrosion features should be considered when evaluating corroded pipelines, particularly depth. In addition, all three neural network-based methods show improved applicability and reliability compared to traditional evaluation methods, with CNN-image having the highest evaluation accuracy (correlation coefficient = 0.9564, average error = 3.46%).
engineering, industrial,operations research & management science