Improving the prediction speed of fatigue life of subsea christmas trees by using Hidden Semi-Markov Models

Aiming at the problems of traditional methods for predicting the residual fatigue life of subsea christmas trees, such as the difficulties in capturing the randomness of multi-state transitions, low model computation efficiency, and insufficient adaptability to multi-stress working conditions, a method for predicting the residual fatigue life based on the Hidden Semi-Markov Model (HSMM) is proposed.Firstly, the group method was used to conduct a five-level arithmetic gradient fatigue test with stress ranging from 360 to 400 MPa on 22 metal specimens. Meanwhile, groups of multi-dimensional monitoring data were collected to construct the raw dataset. By embedding the state transition probability models and the sojourn time function, a dual HSMM model library covering the operation state and the full cycle was established.Secondly, the Forward-Backward algorithm, the Viterbi algorithm, and the Baum-Welch algorithm were used to optimize the model parameters to identify four sub-states and the sojourn time.Finally, the vector autoregressive root test and impulse response analysis were employed to verify the reliability of the model, and the prediction of the residual life of the subsea christmas tree systems was completed.The findings show that the average prediction error of this model is 3 years with 87% accuracy, which is 5.61% higher than that of traditional methods. The prediction duration for a single well is shortened from the average of 58.9 hours in traditional methods to 29 hours, with a 50% increase in the computational speed. The research confirms that HSMM exhibits significant prediction advantages under the working conditions of high stress fluctuation and multi-state coupling. The universality of the algorithm can be improved by dynamically adjusting the working condition parameters, providing efficient support for equipment maintenance.