Engineering-Constrained Bidirectional LSTM for Structural Integrity Prediction and Optimization of Offshore Jumpers

In short

Jumper design normally means running a finite-element model for every candidate geometry and load case. The paper treats the jumper as an ordered sequence of pipe and elbow elements and trains a bidirectional LSTM to predict longitudinal and combined stress unity for every element at once, conditioned on the geometry, pipe section, operating conditions, hub tolerances and end displacements.

Reading the sequence in both directions lets each element's prediction depend on the boundary conditions at both hubs. An embedding for the element type and a position channel let the network learn the different behaviour of bends and straight runs.

The trained surrogate is then placed inside a constrained configuration search. Candidate geometries are generated for a fixed span, hub elevation difference, seabed offset and belly clearance. Infeasible ones are discarded, and the rest are predicted and ranked against a target unity, optionally by where the critical section sits. Two optimized configurations from the paper were checked against finite-element runs, with the predicted peak unity and its location matching closely.

The method reduces the time to evaluate a configuration from hours of finite-element work to milliseconds per prediction, so a design space of thousands of candidates can be screened in minutes during front-end engineering design. The paper is explicit about its scope: static code checks, the M-shape topology and the training ranges.

Cite

Chi Yang (2026). Engineering-Constrained Bidirectional LSTM for Structural Integrity Prediction and Optimization of Offshore Jumpers. Proceedings of the Thirty-sixth (2026) International Ocean and Polar Engineering Conference, Orlando, Florida, USA, May 31 to June 5, 2026, pp. 745–750. Paper ISOPE-I-26-108.

@inproceedings{yang2026bilstm,
  author    = {Yang, Chi},
  title     = {Engineering-Constrained Bidirectional LSTM for Structural Integrity Prediction and Optimization of Offshore Jumpers},
  booktitle = {Proceedings of the Thirty-sixth (2026) International Ocean and Polar Engineering Conference},
  address   = {Orlando, Florida, USA},
  publisher = {International Society of Offshore and Polar Engineers (ISOPE)},
  year      = {2026},
  pages     = {745--750},
  number    = {ISOPE-I-26-108},
  isbn      = {978-1-880653-72-2},
  issn      = {1098-6189}
}