Eestimation Method of Oil Pipeline Casing Deformation Degree Based on Convolution Capsule Network

At present, there are some problems in the electromagnetic detection of oil and gas field pipeline, such as single quantitative method of pipeline deformation and insufficient accuracy. Aiming at the issue, this paper introduced the idea of deep learning and proposed a quantitative estimation method for the deformation degree of oil pipeline casing based on convolution capsule network. Firstly, several convolutions were designed to extract the characteristics of eddy current signals from different probes. Then the output layer based on capsule network was designed to construct the constraint function based on module length to quantify the minimum arm value. Finally, the degree of casing deformation was estimated. This method considered the relationship between different probes to build a quantitative model, which improved the nonlinear mapping ability of the model, and was suitable for equipment with multiple probes simultaneous detection. Through the verification of the actual downhole pipeline pulse eddy current testing data, this method had better quantization accuracy than the common methods.