Deep-water risers are continuously exposed to currents and turbulence. This leads to fatigue damage accumulation due to so-called Vortex Induced Vibrations (VIV). Consequently, VIV represents a safety risk and is a major design consideration adding notable costs to all stages of the riser system development. Offshore platforms collect data from onboard sensors, but the sensor density along the risers is limited. Consequently, the state-of-the-art models for VIV response prediction cannot provide sufficient precision for on-board monitoring and decision support. The industry standard is to compensate the large uncertainties with expensive safety factors.
The objective of PRAI is to develop a hybrid VIV prediction model combined of physics-based numerical models and data-driven machine learning methods. The model will be used for response estimation in the design of deep-water riser systems, but can also be integrated in future on-board riser management systems. This will lead to reduced cost and increased safety of these systems in both design and operation.
The novelty of PRAI is to combine a recently developed time-domain prediction model with concepts from machine learning. The time-domain modelling allows for intuitive interpretation of the underlying physical phenomena, while the machine learning infers connections directly from available sensor data. The ambition is to pave the road for applying the model in an on-board software system with a limited number of sensors. This will reduce unnecessary replacements and foster more responsible riser-production (UN Sustainable Development Goal #12). The methods in PRAI will be developed for a riser system, but are equally applicable to other offshore slender structures, such as risers for deep-sea mining, mooring lines and power cables for floating wind turbines (SDG #7), mooring lines and food hoses for aquaculture for food production (SDG #2).
Deep-water risers are exposed to waves and currents which cause drag forces and accumulation of fatigue damage. While waves play an important role near the surface, deeper down currents and turbulence around the riser dominate the fatigue accumulation by causing Vortex-Induced-Vibrations (VIV). Consequently, VIV represent a safety risk and are a major design consideration adding notable costs during development and operation.
VIV response prediction tools are traditionally based on semi-empirical methods. Despite huge efforts over the last two decades, the methods still contain significant inherent uncertainties and limitations. As a result, simplifying assumptions are made both during the engineering and operation stages, leading to the application of high safety factors in the VIV response calculation to compensate for the high uncertain in the prediction. This in turn adds costs in terms of VIV suppression devices and/or riser operation constraints. Thus, there is a clear need for developing methods that address the above limitations, providing more accurate response and fatigue predictions, which can be applied in the engineering phase and as a decision support tool for on-board riser management systems.
PRAI will combine a recently developed time domain (TD) prediction model with concepts from machine learning. TD modelling allows for intuitive interpretation of the underlying physical phenomena, while the machine learning infers connections directly from available data. The combination is referred to as 'hybrid analytics', where the existing physics-based response model is improved based on available sensor data, and consequently the final model is alleviated of some of the limitations from the simplifying assumptions. The ambition is to reduce the uncertainties in the riser response predictions and to pave the road for applying the model in an on-board software system with a limited number of sensors.