WP1 activity focused on improving the current contact modelling techniques for nonlinear dynamic analysis to ensure accurate predictive capabilities over the lifetime of an aeroengine by:
- developing solutions to allow engineers and researchers alike to assess the position and extent of the zones of relative cyclic motion along the interface.
- developing efficient non-linear model reduction strategy and multi-scale approaches to reduce the size of the problem and enable more complex computations.
- developing and validating prediction tools to study the effect of wear on the dynamics of structures with friction contacts.
- improving the understanding of the friction mechanisms and their effects on the dynamic response of jointed structures, leading to an upgraded and fully validated state-of-the-art modelling approach for nonlinear dynamic analysis.
In WP2, different linear joint identification techniques have been compared and some existing methods were improved further to have better performances. In particular, some guidelines to choose an interface model, useful in practice, are derived. In addition, a new nonlinear identification technique is developed, and the accuracy and robustness of the method when applied to a bolted connection are studied. The main advantage of the method is that it does not have any restriction on the measurement locations. The methods are verified by using case studies, where simulated experimental results are used.
In WP 3, novel methods to compute efficiently the dynamic behavior of modern turbines have been developed and tested, tackling two main challenges:
1. an accurate model of the friction in the interface between blades and disk cannot be built because of the complex physics taking place on the component surfaces in contact,
2. the numerical models of modern turbine components have huge dimensions and simulations can take several hours or more.
The research effort allowed to reach the goals by:
- developing a strategy where the vibration of an assembly of blade and disk is first measured experimentally, leading to a hybrid representation where the contact and friction properties are provided be the experiment whereas the global vibrational behavior is obtained from the numerical model.
- proposing new algorithms to compute the vibration from models with a very large umber of unknowns, by dividing the large model into small portions of the turbine that can be efficiently handled by the many processors of a high performance computer.
- developing new strategies to speed up the iterative processes in parallel computing approaches.
In WP4, the activity focused on enabling new methods from High-Performance Computing for typical engineering simulation codes, by addressing the topics of I/O and task-based programming models. One project developed a system for auto-tuning of I/O parameters using machine-learning techniques which operate transparently in the background without user intervention. A second project developed a library for managing data layout and data-movement across the whole memory hierarchy. Finally, the well-established task-based programming model PyCOMPSs was extended to support new task types for I/O and communication over MPI, respectively Knowledge of the task type is exploited to take better scheduling decisions and overlap these operations with regular compute tasks whenever possible. The software arising from these projects is being published under an open source license.
