Life Extension of Offshore Structure Using Machine Learning

Abstract

Abstract The objective of the paper is to demonstrate the Machine Learning (ML) based Structural Integrity Management (SIM) Methodology and its application for the life extension of the offshore structure. This paper also illustrates how the sensor data are used to generate an ML based predictive model and how it will be used to minimise the inspection cost without using the traditional Risk Based Inspection(RBI) methodology. Structural assessment, real-time monitoring and predictive maintenance are the three main aspects of the life extension process for the offshore structure. Usually, the structures are designed for a fixed design life, but during life extension process it is assessed through FE analysis and various inspection methods whether the structure will have adequate fatigue life for another 5-10 year. But running fatigue analysis is computationally expensive as well as inspection also increase the operational cost. Sensors are installed on the offshore structure, and the stress, acceleration, wave, current etc. are measured and transmitted through wired or wireless sensor network and stored in cloud computing. This data is used for predicting the new wave and current data for 1 and 10-year return period. The acceleration data is used to get the modal frequencies and calibrate the FE model. Also, the measured stress value is compared with the FE model generated stress value, and the FE model is further calibrated. Machine Learning Algorithm (Recurrent Neural Network) is used to generate the predictive maintenance schedule based on the data-driven fatigue prediction model created from the measurement data. The case study shows the life extension of the offshore jacket structure with proposed machine learning based life extension methodology. The data-driven fatigue predictive model generates the remaining fatigue life, and it is compared with the fatigue life calculated from the FE model. It shows a good match and within 5-10% inaccuracy limit. The predictive maintenance schedule is developed based on the remaining fatigue life. ML-based model significantly reduces the computational cost as well as the real-time data also improves the fatigue life calculation accuracy. Hence, though predictive maintenance, the overall operational cost will be reduced significantly.