Abstract

Pipe infrastructure systems in service are aging and continue to degrade with passage of time. As the defects grow with time, for safety of mankind they have to be introspected continuously. Due to regular usage the inner surfaces can be inundated with tiny cavities which are not harmful and do-not need immediate repair. However, due to continuous usage these defects have to be monitored continuously so that whenever they are about to reach the pre-fixed threshold of being harmful we can repair them without delay. Here, we have developed a novel method for identification of large obtrusive defects using magnetic flux leakage (MFL) based nondestructive evaluation (NDE) technique and dynamically updated transfer learning. Running the pipeline inspection gauge (PIG) within the pipeline to collect very accurate, low noise readings for defect detection is expensive and time-consuming. The objective is to automatically detect the defective areas at the beginning and data obtained via fast inspection full of noise is estimated by mixture regression which produces posterior probabilities of the defects at each scan point. We can use transfer learning perspectives by leveraging the defect probabilities and location from the previous days, and then consequently update those probabilities based on current data by applying a dynamically updated transfer learning for properly detecting the size of the defect.

How to Cite:

Mukherjee, S., Huang, X., Udpa, L. & Deng, Y., (2019) “Dynamic tracking of defects in pipelines via NDE based transfer learning”, Review of Progress in Quantitative Nondestructive Evaluation .