Utilizing Natural Frequency Monitoring and Machine Learning to Monitor and Predict Structural Integrity and Minimize the Cost of Fixed Offshore Platform Intervention

Anthony Falsetta,Elaine Whiteley,Craig Dickinson,George Zhou,Shankar Sundararaman
DOI: https://doi.org/10.2118/202645-ms
2020-11-09
Abstract:Abstract The objective of this paper is to present a method for monitoring the natural frequency of a fixed offshore platform and to utilize the raw data to proactively monitor and predict structural degradation of the platform using machine learning. This can enable operators to predict critical member failure and guide timing of operational decisions and interventions, thus helping reduce cost of ownership, particularly for ageing assets. A Natural Frequency Response Monitoring (NFRM) system utilizes accelerometers and Fast Fourier Transform (FFT) to measure the sway and torsional frequencies of a fixed offshore platform. A significant structural event is detected by a shift in the natural response of the platform. A machine learning model (or digital twin) can be trained on this accelerometer data to replicate the platform response and determine the fatigue damage accumulation in platform members. The model, which can be re-trained to maintain accuracy, acts as a near real-time fatigue tracker with the NFRM system acting as a verification tool. A proof of concept case study is presented and demonstrates how Finite Element Analysis (FEA) simulations and measured data can be used to develop a machine learning model. The combination of an NFRM system with machine learning has the potential to assist operators in the prediction and identification of structural changes prior to them becoming critical. Knowledge of the actual fatigue damage in key platform members on a near real-time basis allows for proactive planning of inspections or repairs, as well as confidence in the structural performance of the platform, which can justify life extension.
What problem does this paper attempt to address?