With the rapid pace of innovation and development in the wind energy industry, the necessity of computer modelling as a substitute for physical experiments is increasing apace. The use of computer modelling can both decrease costs associated with physical experiments and enable rapid prototyping of components. With the increased reliance on computer models has come an associated focus on the complexity and fidelity of these models resulting in modern FE models containing possibly millions of degrees of freedom along with material and geometrical nonlinearities. This increased complexity has obvious impacts concerning the computational resources required for analysis. Whilst this may not pose a particular problem for one-off analyses, problems can rapidly arise when considering model updating or uncertainty quantification both of which being essential model validation procedures.

Traditional dynamic substructuring methods such as the Craig-Bampton were originally developed for greatly reducing the complexity of linear finite element models, whilst preserving response fidelity within frequency regions of interest [1]. Such linear methods have been widely studied and implemented in many commercial FE packages. When a nonlinear system is considered however, these linear methods cannot maintain fidelity of the reduced model. As a response to these failings, there have been numerous recent efforts to develop a nonlinear dynamic Substructuring framework. Some techniques focus on the enriching of standard linear methods to capture nonlinear effects with an example being the use of modal derivatives for geometrically nonlinear systems [2]. Whilst other methods create entirely different reduction procedures such as the use of nonlinear normal modes or metamodeling [3].

In this paper the potential of nonlinear normal modes for reduction of nonlinear finite element models is presented. The extraction of nonlinear normal modes using output only data through autoencoder neural networks is also discussed,

[1] Craig, R., R. and Jr. (1985) ‘A review of time-domain and frequency-domain component mode synthesis method'. Available at: https://ntrs.nasa.gov/search.jsp?R=19860042139 (Accessed: 13 May 2019).

[2] Wu, L. (2018) ‘Model order reduction and substructuring methods for nonlinear structural dynamics'. doi: 10.4233/UUID:F9736E8B-D00F-4E25-B456-48BD65B43788.

[3] Apiwattanalunggarn, P., Shaw, S. W. and Pierre, C. (2005) ‘Component Mode Synthesis Using Nonlinear Normal Modes', Nonlinear Dynamics. Kluwer Academic Publishers, 41(1–3), pp. 17–46. doi: 10.1007/s11071-005-2791-2.