Reliability‐Oriented Optimization of Composite Drill Pipes Using Integrated Finite Element and Reinforcement Learning Framework

ABSTRACT This study aimed to optimize the stacking sequence for composite pipes under normative drilling applications through an integrated Finite Element–Reinforcement Learning scheme. In this essence, finite element provides the reward required to train a Deep Neural Network that gradually learns how to minimize the damage experienced by the composite pipe and shifts toward an optimum selection of fiber orientation for all layers. The optimum layup design was then assessed in terms of how it could endure the operational loads. In this way, to understand how different loading scenarios affect the creation of damage within layers, three Machine‐Learning algorithms, Random Forests, Gaussian Process Regression, and Adaptive Neuro‐Fuzzy Inference System, were developed and trained based on finite element simulations performed on a coarse grid of different combinations of operational loads. The best models were employed to enhance the data resolution and, hence, to plot first‐ply failure envelopes and safe operational limits. Furthermore, delamination was analyzed, and it was shown that this phenomenon did not limit the applicability of the optimum pipe. Finally, to account for the variability in mechanical properties, critical loads were calculated based on a population of pipes with statistically distributed mechanical properties. It was shown that such variations in the properties can significantly affect the applicable loads, and stringent precautions should be taken. Appropriate safety factors were proposed based on a thorough analysis of the behavior of composite pipes through statistical studies rather than rules of thumb.