Weld crack detection in spiral-welded pipes by direct current potential drop method and backpropagation neural network

Abstract

Pipelines are essential for transportation, and fractures can lead to severe accidents. Accurately detecting weld cracks is vital for supporting the safe operation of large-diameter spiral-welded pipelines. The direct current potential drop method detects cracks by observing the discontinuity of the electrical potential field caused by cracks inside a current-carrying body. The variation in crack lengths and positions significantly affects the measured potential drops. Traditional calibration curves focus on the relationship between crack length and potential drops, but detecting crack position is also essential. This research introduces an innovative method to identify the position and length of weld cracks in spiral-welded pipes by combining the direct current potential drop method and the backpropagation neural network. Finite element models of spiral-welded pipes with varying crack positions and lengths were created, and extensive simulations were conducted to collect potential drops across the weld seams. A backpropagation neural network model is constructed and trained through deep learning technology. The well-trained backpropagation neural network can precisely predict the position and length of the weld crack by scanning the potential drops of the entire weld seam. Several experiments have been conducted to validate the proposed method for detecting weld cracks.