Unsupervised Neural Network for Data-Driven Corrosion Detection of a Mining Pipeline

Abstract

Pipelines failure often caused by corrosion may result in safety, environmental and economic issues. In this study, an unsupervised neural network, Self-Organizing Maps (SOM), is applied to create clusters representing the corrosion impact assessed with ultrasound periodic inspections. Based on this work, it is expected that the new insight into thickness data representation using unsupervised neural network will facilitate planning of corrosion mitigation activities through risk-based inspections of mining slurry pipelines. As a result, SOM led to the reduction of the variables in two-dimensional space nodes. Hierarchical ascending classification (HAC) was then used to classify these nodes regrouping thickness loss measurements. The proposed method by combining both SOM and HAC succeeded in detecting the extent of corrosion in a mining pipeline.