Multidimensional big data processing for damage detection in real pipelines using a smart pig tool

Abstract

The history of the hydrocarbons business in Colombia dates back to the early twentieth century where mining and energy sector has been one of the principal pillars for the its development. Thus, the pipelines currently in service have over 30 years and most of them are buried and phenomena like metal losses, corrosion, mechanical stress, strike by excavation machinery and other type of damages are presented. Since it can generate social and environmental problems, monitoring tools and programs should be developed in order to prevent catastrophic situations. However, the maintaining of these structures is very expensive and it is normally developed by foreign companies. In order to overcome this situation, recently the native research institute “Research Institute of Corrosion - CIC (Corporación para la Investigación de la Corrosión)” developed an in-line inspection tool to be operated in Colombian pipelines (especially gas) to get valuable information of their current state along of thousand kilometres. The recorded data is of big size and its processing demand a high computational cost and adequate tool analysis to determine a certain pipeline damage condition. On other hand, the author from UPC and UIS have been bringing its expertise in processing and analysing this type of big data by using mainly Principal Component Analysis (PCA) as an effective tool to detect and locate different damages. In previous papers, multidimensional data matrix was used to locate possible damages along the pipeline, however most of activated points were considered false alarms since they corresponded to weld points. Thus, in this paper it is proposed no considering piecewise weld points (tube sections) and an extension of PCA named Multiway PCA (MPCA) is applied for each each one of the tube sections that form the pipeline. Therefore, if a tube section is found outside from overall indices found by using the MPCA model, an alarm activated in that section and a precise location can be obtained by analyzing only data from that specific tube section.