Deep Learning–Assisted Ultrahigh‐Cycle Bending Fatigue of Precorroded Completion Strings

This study examines ultrahigh‐cycle bending fatigue behavior of BG2532 after H 2 S precorrosion. At lower stress amplitudes (< 680 MPa), precorrosion has a pronounced effect, leading to increased uncertainty in fatigue life. Fatigue cracks preferentially initiate at the upper and lower edges of the bending section, where H 2 S‐induced surface degradation produces numerous micro fatigue sources. Additionally, H 2 S penetration corrodes the subsurface microstructure, causing localized stress concentration and promoting crack propagation. To describe this behavior, a grouped convolutional neural network was employed, leveraging its capability to model nonlinear relationships efficiently. Only the following four inputs were required: corrosion state, stress amplitude, and two fatigue‐fracture‐related features representing crack initiation and propagation mechanisms. The proposed model achieved accurate prediction of bending fatigue life after precorrosion, with results largely within a factor of two error band.