Deep learning-driven optimization design for the configuration of lazy-wave umbilical cables

Abstract

Umbilical cables are essential components of subsea production systems for deep sea oil and gas exploration. Umbilical cables are often set to specific configurations that compensate for the motion of the floating body resulting from harsh environmental loads during operation to ensure structural safety and proper functioning. However, the optimization design of the umbilical cable configuration is a challenging problem owing to highly nonlinear relationship between loads and structural responses as well as the huge computational cost of evaluating numerous sea state combinations. This study proposes a deep learning-driven optimization design framework based on the residual artificial neural network to address the configuration design of lazy-wave umbilical cables. A parallel accelerated computation method based on Adam optimizer was established for the generation of the database. The transfer learning technique was also introduced into the deep learning model to build a configuration design prediction model based on small samples, which significantly reduced the cost of model building and training. Based on the above optimization design framework, a combination of tensile and bending optimization for a lazy-wave umbilical cable configuration was achieved, verifying the effectiveness and practicality of the proposed approach. This study provides an important reference for the rapid engineering design of umbilical cables.