Quantification of pitting corrosion from thermography using deep neural networks

Pipelines serve as the major infrastructure for transmission and distribution in oil and gas and water industries. The health condition of a metallic pipe is determined mainly by the extent to which it is corroded. Thus, the quantification of pitting corrosion in terms of metal loss is required for the understanding of pipe condition. There are different ways to quantify corrosion pit geometry. Direct methods measure the pit depth of pipe samples, which are sand/grit blasted to remove corrosion products, and are often adopted in the laboratory. Indirect methods employ non-destructive inspection techniques to detect and quantify the corrosion without sandblasting, which is preferred for a field test. In this study, pulsed thermography is considered for the quantification of pitting corrosion in a metallic pipe. Thermography testing can generate a sequence of infrared images, which reflect the diffusion process of heat through the pipe wall. This paper proposes a new deep neural network-based approach to quantify the pitting corrosion damages from the acquired thermography images. Through extensive experimental tests, the quantified results demonstrate a reasonably good linear relationship with metal loss of pipe, as the quantification accuracy was better than 98%.