| Summary: | Current methods to predict hydrodynamic loads rely either on oversimplified and semiempirical methods or the use of numerical simulation and analysis techniques such as Finite Element Analysis (FEA) or Boundary Element Analysis (BEA). These methods are conservative which results in the over-design of these craft so they are heavier and slower than they could otherwise be. Better understanding of load intensities will inform the design process of marine structures and could result in lighter and more efficient designs. This research investigates the possibility of solving these problems employing artificial intelligence (AI) as an alternative to the current methods. Few studies have applied Artificial Intelligence to the design of marine structures. Detailed review of the past and present research shows that AI and in particular Artificial Neural Networks (ANN) can be used as an inverse problem solver when there are no closed form relationships that exist between the input and the output. An inverse approach is defined as the problem where response of the structure is known but the load that caused that response is unknown. In real problems/structures the response to a point load is experienced throughout the structure with different levels of intensities which is the link between the external load and these differential intensities. Determining this relationship will result in a unique solution without the knowledge of material constitutive laws, material properties and structure size or thickness. The aim of this investigation is to develop a real time in-service load measurement tool using an inverse approach. To achieve this, ANN, experimental techniques and FEA analysis are combined to form a hybrid inverse problem solver that can be trained to use structural response, such as strains at various locations, to predict the loads that caused them. The main objective of this research is to investigate the suitability of the proposed methodology for real time in-service load monitoring on large marine structures. The proposed system must be able to measure both steady-state as well as transient load such as equivalent slamming load. The outcome of this investigation was successful prediction of the external loads in terms of their approximate location and load intensities. The only disadvantage of this method is that the solver requires training and can only learn from cases that it has been subjected to. However, once the system is trained it can predict both static and dynamic loads quickly and accurately.
|
|---|
