Autonomous Modelling Solutions for Operational Structural Dynamic Systems

Identifying and understanding how engineering structures respond over time to different loads is a key task which supports many other activities. For example, the creation of digital twins for design, production and management of engineering systems is currently a key area of development across academia and industry. This project aims to extend the range of operational scenarios in which this is possible and to further automate the process.

The offshore energy sector is a good example of an application area where the technology being developed in this project can support enhanced insight into an engineering system. The nature of the environment often means direct measurements of structural loads is infeasible. For example, the forces generated by the wind on a turbine blade or by the waves on a mono-pile structure are difficult to measure but important in the structure's operation. This difficulty motivates the development of methods which can infer, not only the dynamic properties of these systems, but the unmeasured inputs (loads) from the available measurements. The project will also extend the identification task further to include the ability to make autonomous modelling decisions. In other words, the information from collected data is fused with prior engineering judgement to learn appropriate ways to represent the system automatically.

The technology developed in this project will support operations and maintenance activities in two key ways. By providing improved estimates of the properties of engineering structures and the conditions they have operated in, decisions can be made with an increased level of confidence. Secondly, automating the way engineers build models reduces subjectivity in the process and frees up engineers to focus on where their intervention will bring greatest benefit.

The key outcome of this project will be allowing engineers better understanding of the behaviour of structures of interest, developing better representations of them - such as digital twins - and improving their ability to make operations and maintenance decisions based on measurements. Overall, this increased insight into engineering systems will allow for more targeted and effective management. The result of which is reduced unnecessary maintenance, hence a reduction in costs and reduction in risk to the staff that have to perform maintenance in these harsh environments.

There are challenges in this proposal; the first is to enhance existing methodologies for learning the inputs and parameters of a dynamic system to include more complex loading scenarios. These scenarios include modelling multiple correlated forces, distributed loads and loads at unknown locations. The second challenge is to bring further automation to the process of deciding an appropriate model for the dynamic system. It is hypothesised that by addressing these two challenges, determination of multiple/distribution unknown loads and automated learning of the dynamic model structure, the range of situations which can be robustly considered is increased. This enhanced understanding will greatly aid the incorporation of dynamic models in virtualisation of engineering systems and the development of smart infrastructure.