The researchers of the ITN MAGISTER will study one of the most persistent challenges in aircraft engine development by using a completely new approach: predicting and controlling acoustic oscillations in aircraft engine combustors with machine learning methods. Over the last...
The researchers of the ITN MAGISTER will study one of the most persistent challenges in aircraft engine development by using a completely new approach: predicting and controlling acoustic oscillations in aircraft engine combustors with machine learning methods. Over the last decades the emission regulations for aircraft engines have been steadily tightened.
This project focuses on this new clean combustion technology and its interaction with the increased compression ratio. This development introduces a large risk for reduced reliability and lifetime of engines: pressure oscillations in the combustor called thermoacoustics. Industrial experience shows that often pressure oscillations surface at high TRL, while not detected at low TRL. This is very troublesome when the full engine design has been finalized. The high uncertainty involved in predicting and controlling thermoacoustics in aircraft engines will be addressed in this ITN by a self-adapting approach based on machine learning. Objectives: 1. Develop methods that can predict and control thermoacoustics from TRL 2 to TRL 9. 2. Apply machine learning algorithms to improve models to predict thermoacoustics in aircraft engines and derive combustor hardware design implications from the predictions. 3. Devise and adapt machine learning algorithms to thermoacoustic experiments at the laboratory scale and to industrial scale for aircraft engines.4. Advance acoustic and combustion models to capture the interaction of acoustics with liquid fuel sprays with high accuracy.5.Generate a sophisticated experimental data base for thermoacoustics of liquid fuel combustion for validation of the methods developed in MAGISTER.
ESR 1 performed 1000 experiments on a laboratory-scale thermoacoustic rig containing a turbulent swirling flame. From these experiments algorithms were developed. The algorithms were able to identify the operating points and thermoacoustic decay rates with remarkable accuracy. ESR3 and ESR4 have been pursuing different methods to learn Flame Describing Functions from simulation data. ESR3 is using neural networks to estimate flame dynamics from OpenFOAM simulations of a steady flame. ESR4 relies on quadratic models and simulations from ANSYS Fluent.
ESR8 has created a digital twin of a lab scale thermoacoustic experiment. A thermoacoustic model is built on top of the digital twin and its coefficients are tuned by assimilating data from 100,000 experimental measurements of the frequency and growth/decay rates of thermoacoustic oscillations.
To assess uncertainties resulting from physical parameters and modelling parameters, a LES of premixed turbulent swirl flame has been set up in AVBP. Work was done to assess the interaction of acoustics with droplet formation processes. A 1D analytical model to characterize the dynamic response of droplet evaporation to acoustic perturbations is formulated. A strong correlation between droplet number density wave and spatio-temporal distribution of the vapor formation rate is observed. Work has been devoted to the modelling of two-phase combustion. In particular multi-component evaporation has been implanted in AVBP and its impact on the flame structure has been studied. SU2 code was developed to predict isothermal incompressible flows in the UT designed burner. With ANSYS CFX successfully spray combustion simulations have been performed of the UT burner.
ESR8 has put great effort into characterising an existing thermoacoustic rig. The more accurate measurements reveal that both the viscous and the thermal drag at the heater must be accounted for in order to correctly quantify the influence of the heater on the acoustics. At TUM existing acoustic impedance models for perforated walls of acoustically absorbing liners were reviewed and shortfalls were identified. The most promising model was selected and modifications were made to it to improve the performance of the model with respect to holes-to-hole interaction. To validate the improved model an existing damper impedance measurement test rig was modified to make it compatible with perforated wall segments. The data available so far indicates that the modified model performs better than the standard model. ESR9 strives for employing machine learning to improve the predictive quality of the model. The focus of the ERS10 is on the development of code for the computation of acoustical phenomena in combustor flows. ESR10 starts to get results for the non-reflecting boundary condition treatment using the PML approach in combination with a high order Discontinuous Galerkin discretisation.
The focus of WP4 is on providing experimental data on the response of a cold spray (ESR 11, KIT) and the response of a spray flame (ESR12, UT) to acoustic perturbations and forcing. ESR 11 studies the influence of an oscillating air flow on the atomization process in a burner with airblast atomizer. A test rig was designed and built up, in which the air modulation device, the air ducts, the airblast atomization nozzle, and all the other secondary components were implemented. Shadowgraphy is assembled and tested; PDA is operational but requires optimization. For hot wire anemometry system, the calibration setup is in progress. Experiments on 2 combustion test rigs with very different characteristics are used by ESR 12 to investigate self-exited and external excitation cases. Nonlinear analysis treatment of the pressure time series for the LIMOUSINE atmospheric pressure combustor has been done. Experiments were realized going from a stable to an instable state and showed a trend from a torus shaped phase portrait to a conglomerate trajector
The development of modelling tools is a very important activity in MAGISTER. The goal is to seamlessly combine physics-inspired parametric models with non-parametric flexible models over the spectrum of experimental conditions. Through this approach, the algorithms will learn iteratively from experiments and simulations, build a faithful model of the thermo-acoustic mechanism, and thereby propose the design changes that will eliminate oscillations over the desired operational window. In short: the ITN researchers will create an intelligent digital combustion engineer which will be better informed and therefore more successful than an engineer using traditional engineering methods. The project will develop tools to enable the development and use of the new airplane engine technology, that will help achieve the 2030 climate and energy framework targets.
More info: http://www.utwente.nl/magister.