Self-Organizing Map-Based Feature Visualization and Selection for Defect Depth Estimation in Oil and Gas Pipelines

Magnetic Flux Leakage (MFL) sensors are commonly utilized to detect defects in oil and gas pipelines and determine their depths and sizes. As a preprocessing step, MFL data are often reduced into a representative feature set that is capable of accurately estimating pipeline defect depths. However, this estimation capability may vary depending on the features used, which necessitates the need for selecting the most relevant ones. In this paper, self-organizing maps (SOMs) are used as feature visualization tool for the purpose of selecting the most appropriate features. First, a self-organizing map (SOM), i.e., A two-dimensional discretized representation of the input space of the training samples for the features, is produced. The SOM weights for each individual input feature (weight plane) are displayed then visually analyzed. Irrelevant and redundant features can be efficiently spotted and removed. The remaining "good" features (i.e., Selected features) are then used as an input to a feed forward neural network for defect depth estimation. Experimental work has shown the effectiveness of the proposed approach. For instance, within ±5% error-tolerance range, the obtained estimation accuracy, using the SOM-based feature selection, is 93.1%, compared to 74% when all input features are used (i.e., No feature selection is performed), and within ±10% error-tolerance range, the obtained estimation accuracy, using the SOM-based feature selection, is 97.5%, compared to 86% when all the input features are used (i.e., No feature selection is performed).