Exhaust Jet Analysis

Abstract

The aim of project A3 of the Collaborative Research Centre (CRC) 871 is to develop a novel method for automated condition assessment of aircraft engines. The method is based on a combination of numerical simulations, Background Oriented Schlieren (BOS) measurements, and the use of pattern recognition algorithms. The density distribution in the exhaust gas stream of the engine is first measured using the BOS method and then compared with a damage library which was generated numerically through CFD. The automated classification of damage cases can be realized using pattern recognition algorithms. In the third funding period of the CRC, existing reconstruction algorithms used for the calculation of the density distribution in the exhaust gas were improved to increase the accuracy of the BOS measurements. Subsequently, the method was tested experimentally. First, experiments with a model combustion chamber show that 98.5% of all damage cases can be correctly classified by a suitable choice of integral parameters to describe the density distribution in the exhaust gas jet. In a second experimental test, the method was applied to a research engine in order to take realistic impacts into account.

1 Introduction

Due to the growing connectivity and mobility in society, a global increase in air traffic has been observed. According to a forecast made by the International Air Transport Association (2017) (IATA), passenger traffic is expected to double in the next 20 years, which will lead to an increasing demand for commercial aircraft. This will also lead to higher cost of maintenance. Aircraft engines are among the most maintenance-intensive components of an aircraft, as they are subject to particularly stringent safety criteria and have a decisive influence on aircraft performance. It is estimated that 30% of operating costs are attributable to aircraft engines, of which one third is due to their maintenance (Rupp 2001). This makes it clear that there is a potential for cost savings in the maintenance process, which is the incentive for the development of new methods. For the maintainers, precise knowledge of the engine's condition is necessary to enable them to provide targeted repair and, at the same time, to reduce the risk of unexpected damage occurring during the maintenance process. For this purpose, engines are nowadays equipped with Engine Health Monitoring (EHM) systems, which can monitor the engine's performance on the basis of thermodynamic data within the framework of a Gas Path Analysis (GPA) and are used for detecting deviations. Despite the continuous improvement of the monitoring systems, a full condition assessment cannot yet be carried out, since such systems obtain their data from the engine control system. The condition assessment is therefore limited to the measurement of pressures and temperatures at discrete locations of the engine, which represents a small database for a detailed analysis of the engine condition. According to Volponi (2014), an increase in the number of sensors used is not a solution to this problem, as it entails an increase in engine weight and maintenance effort for the additional sensors. By contrast, a non-contact engine inspection offers the possibility of identifying damaged components before the repair for a precise estimation of the maintenance effort. Nowadays, borescopes are inserted through small openings in the engine to assess the damage condition of the engine. However, since the number of borescope openings in engines is limited, a full damage inspection cannot be carried out here either, which leads to a residual risk of the occurrence of unexpected sources of damage.

For these reasons, there is an unsatisfied demand for a condition assessment of the engine prior to disassembly for reliably planning engine maintenance. Within the scope of the present project, a novel method for the detection of defective components in the hot gas path of an aircraft engine has been developed and tested. By combining numerical flow simulations, optical measurements, and pattern recognition algorithms for machine learning, this approach allows detecting component damage before disassembly. The method proceeds in 3 steps:

• 1. The influence of characteristic component damage on the density distribution in the exhaust gas jet is predicted using numerical flow simulations.

• 2. The simulations are then used to build a damage library in order to train pattern recognition algorithms.

• 3. The engine condition can then be analysed and evaluated using these pattern recognition algorithms. For this purpose, the density field in the exhaust jet of an engine is reconstructed using the BOS method and compared with the library to identify the damage case.

With the aid of numerical simulations, Adamczuk et al. (2013a) investigated the influence of defective components on the density distribution in the exhaust gas jet and were able to show that both, a burner failure and different defects in the high-pressure turbine have an influence on the density distribution downstream of the turbine. Building on this, Adamczuk (2014) and Adamczuk and Seume (2016) were able to show that the influence of selected defects can be observed as far downstream as the exhaust gas jet of the engine, which is a basic requirement for a damage analysis with the BOS method. In the work of Hartmann (2012), Adamczuk et al. (2013b) and Adamczuk (2014), the density distribution in the exhaust jet of a helicopter engine was reconstructed using the BOS method. Density non-homogeneity that occurs in the form of a cold air stream in the exhaust gas jet were considered. The applicability of the BOS method to a civil aircraft was also demonstrated by Adamczuk (2014) and Hartmann et al. (2015) by measuring the exhaust jet using a non-tomographic setup consisting of one camera. This set up identified cases of damage such as oil leakage in the bearing seal air.

2 Objectives

The quality of an automatic failure classification by using the BOS method depends on the reconstruction accuracy of the BOS algorithm. There already exist algorithms such as the filtered back projection (FBP) and the algebraic reconstruction (ART), which can be used for reconstructing the density distribution in the exhaust gas flow. Preliminary investigations have shown, however, that especially in regions with large density gradients artefacts are formed, which impair the quality of the reconstruction. Therefore, one objective of this work is to extend already existing algorithms in order to increase the reconstruction accuracy of complex density structures.

Within the framework of an experimental test on a model combustion chamber, it will be examined whether the combination of achievable reconstruction accuracy and the properties of the pattern recognition algorithm are sufficient for an automated classification of combustion chamber failures. For this purpose, a model combustion chamber is used, which offers the possibility of adjusting the parameters of individual burners in order to simulate a damage case.

As a next step towards an industrialization, the applicability of the presented method to an aircraft engine has to be demonstrated. For this purpose, numerical simulations are first used to test whether it is possible to automate the detection of defects and defect combinations in the engine.

By using synthetic BOS measurements based on the numerical simulations, various parameters of the BOS setup such as the number of cameras and the distance to the exhaust gas jet are investigated in order to find the optimal choice of parameters. The results of the synthetic measurements are then used to determine the probability of success of an automatic classification by the chosen pattern recognition algorithm.

In a second experimental test, the applicability of the presented BOS method in a real research engine will be investigated as part of the scope of two measurement campaigns. For this purpose, a special measuring ring was designed and manufactured for the attachment of the BOS measuring equipment. The measurement data from the engine test will be used to evaluate the accuracy of the numerical simulations and the automated defect detection on the basis of a real database.

3 Tomographic Reconstruction of Density Fields

The background is observed by the cameras through the whole density field. The displacement seen by a single camera only gives the integral of the density gradients from the camera. To reconstruct the density field, the projections from several cameras are needed. To evaluate the number of cameras needed and to test existing tomographic reconstruction methods, an artificial density field was used. The artificial density field was created with RANS simulations of a low- pressure turbine. To simulate cold streaks, an inhomogeneous temperature field was set as the inlet boundary condition. The three main cold streaks with decreasing intensity are marked (1, 2, 3) in Fig. 1.

Fig. 1

Artificial density field from RANS simulation of a low pressure turbine (LPT) with inhomogeneous inlet boundary condition

The advantage of using an artificial density field is that the reconstruction can be directly compared to the original density field. Also, the number, location, distance, and angle of view of the cameras can be chosen freely. For the following comparison of reconstruction algorithms, 16 virtual cameras, placed with equidistant spacing on a 180° arc, are used. Special attention was paid to the reconstruction of large density gradients resulting from the cold streaks.

Two reconstruction algorithms, the filtered backwards projection (FBP) and the algebraic reconstruction technique (ART) are compared. The FBP reconstruction algorithm was applied to BOS by Goldhahn and Seume (2006) based on the work of Radon (1917). The three main steps are:

• 1. calculation of the Fourier transforms of the measured projection integrals

• 2. back transformation of the high-pass filtered signal

• 3. back projection with coordinate transformation.

The implementation and use for BOS measurements can be found in Goldhahn (2009). The ART is based on solving a linear system of equations with an iterative method. Compared to the FBP reconstruction, the ART algorithm is slower, but needs less projections for a good reconstruction (Kak and Slaney 2001; Guan and Gordon 1996). This can be seen in Fig. 2a and b, comparing the FBP and ART with 16 virtual cameras each. The ART shows less reconstruction artefacts and is closer to the artificial density field (Fig. 1). The relative density difference is shown in Fig. 2c and d. Here, the artefacts can be seen as a large relative differences. With the ART, the first and second cold streak can be well detected while the third cold streak cannot.

Fig. 2

Comparison of FPB and ART reconstruction of artificial density field (from Hartmann 2020)

To improve the quality of the reconstruction, Hartman (2020) developed a method combining the FBP and ART algorithm into an initialised algebraic reconstruction technique (iART). In this method, the gradient field from the FBP reconstruction is used to initialise the ART. In Fig. 3 the results of the reconstruction with 16 virtual cameras are shown. The artefacts, seen in the reconstruction with the FBP and ART method are further reduced and the third cold streak can be detected. Using the root mean square deviation as a measure for the accuracy shows, that the quality of the measurement is improved from RMS_FBP = 0.1, over RMS_ART= 0.0547 to RMS_iART = 0.0257 (Hartman 2020).

Fig. 3

Reconstruction with the initialised algebraic reconstruction technique (iART) algorithm

4 Application of the BOS Method to a Model Combustion Chamber

In the first experimental test, the combination of the BOS measurement with the support vector machine (SVM) described by Hartmann (2020) is investigated for automated defect detection on a model combustion chamber. First, the BOS measurement is tested for sufficient accuracy to achieve automated classification. Subsequently, integral parameters of the density distribution are introduced to parameterise the defect influence on the density distribution. The parameters are then used to automate the classification using SVM.

Experimental Setup

For the investigations, a model of a combustion chamber was used in the laboratory of the partner institute for combustion, ITV (Fig. 4). This consists of eight swirl-stabilised premix burners distributed over a circle with a circumference of 210 mm. Each burner consists of a 28 mm diameter combustion tube mounted with a swirl generator and a turbulence grid to simulate realistic combustion chamber flow conditions. The experimental set-up allows individual burners to be manipulated in order to simulate defects such as clogging of the fuel nozzles. Furthermore, it is possible to investigate geometric defects by adjusting the position of individual burners. A detailed description of the test rig can be found in Hennecke et al. (2015), von der Haar et al. (2016), Hartmann et al. (2016) and Hartmann et al. (2018).

Fig. 4

Model of the combustion chamber. Figure according to von der Haar et al. (2016) and Hennecke et al. (2015)

For the reconstruction of the density distribution, 16 cameras of the type Allied Vision Technologies Manta G201b with 25 mm lenses were used. The cameras are positioned on a semicircle with a radius of 1950 mm around the burner. The angle between two cameras is 11.75°. In front of the cameras a dot pattern with a dot size of 3–4 px is placed. For each operating point, 500 images are generated per camera, at a frame rate of 50 Hz. The exposure time is 150 μs due to the low beam speed.

Results

In the experiments conducted by Hartmann (2020), several parameters of a burner were varied in order to investigate the influence on the density distribution of the exhaust gas jet behind the combustion chamber. A detailed description of the results is documented in Hartmann's dissertation (Hartmann, 2020). At the beginning, the output of a single burner was reduced to simulate a burner failure. FIG shows the exit density together with the marked positions of the burners. The investigations have shown that there is a visible loss of rotational symmetry in the density distribution of the exhaust gas. Due to the local momentum deficit caused by a burner failure, cold air flows into the flue gas jet, which leads to a local increase in density (Fig. 5). A similar behaviour occurs when varying the air ratio of an individual burner. A decrease leads to a reduction of the volume flow at constant burner output and thus to a rich combustion. The reduction of the volume flow causes a momentum deficit which allows fresh air to enter the exhaust jet.

Fig. 5

Influence of the reduction of the burner power on the normalised density distribution in the exhaust gas jet. a Reference case b Measurement with burner failure (from Hartmann 2020)

As already shown in Hartmann et al. (2018), geometric defects of burners also have the potential of influencing the density distribution in the outlet of the combustion chamber. The circumferential position of a single burner was varied, resulting in localized regions of higher density, similar to the other two parameter variations. In the case where the burner power was simultaneously reduced, an increase of the region of higher density was observed. It was also possible to show that a burner offset in combination with an increase of the air ratio has no influence on the density distribution. This shows that a mutual compensation of defects can occur, which makes detection difficult.

Subsequently, Hartmann (2020) investigated the possibility of an automated classification of defects using the combustion chamber model. Integral parameters were introduced to describe the density and the density gradient at the outlet of the combustion chamber. For this purpose, various statistical moments were introduced and tested for their ability to parameterise the influences of defects. Some parameters are shown as examples in Fig. 6. The measurement data from the case studies with different combustion chamber defects were used as training data for a support vector machine (SVM) algorithm to form a separation plane between the different classes as shown in Fig. 6. A total of 400 data sets from the measurements were used. Of these data sets, 100 were used for the reference class and 300 for the defect class. Two thirds of the latter data sets were used for training the SVM algorithm and one third as a test data set.

Fig. 6

Selection of statistical moments to describe the normalised density distribution and separation plane between different classes with optimised selection of parameters (Hartmann 2020)

To find the required number of parameters, the number of parameters used was increased gradually and then the result of the classification was calculated with the test data. Figure 6 shows that a good separation between reference and damage cases can be achieved with only a few parameters. By using two parameters, a classification accuracy of 99% of the defective cases and 87.8% of the reference cases can be achieved. Thus, an overall accuracy of 96.2% can be achieved. The best result can be achieved with five parameters. Here, 98.5% of all cases are correctly assigned, of which 99% are defective cases and 97% are reference cases.

5 Application of the BOS Method on an Aircraft Engine

Next, the BOS method was used for the investigation of an aircraft engine. First, numerical investigations of the entire hot gas path (HGP) are performed, to check whether combustion chamber damage has an influence on the density distribution at the outlet nozzle of an engine.

Numerical Setup

The finite-volume solver TRACE 9.1.538 of the German Aerospace Center (Franke et al. 2005) is used for the numerical simulations of the turbine and the exhaust jet. The numerical setup (see Table 1) of the turbine is resolved with 93 million cells and the exhaust jet with 26 million cells. For all calculations, a second-order accurate Fromm scheme (Darwish 1993) using no limiter is used for spatial discretisation. Turbulence is modelled using the shear stress transport model k-w-SST by Menter (1994) with the correction by Kato and Launder (1993) to correct the overproduction of turbulent kinetic energy at the stagnation region. Rotational effects are considered using the modification by Bardina et al. (1985). Boundary-layer transition on the blades is modelled by a multi-mode model (Kozulovic et al. 2007) and all walls are resolved with the wall-bound cells adhering to y⁺  ≤ 1.

Table 1 Parameters of the numerical simulation model

For unsteady calculations, in addition to the previously specified settings, an Euler backward 2nd order time discretisation scheme with 1080 timesteps per period and 30 sub iterations was used. The Courant-Friedrichs-Levy number is set to CFL = 50. These parameters were derived by a timestep study of the mid-span flow path.

Two-dimensional boundary conditions at the high-pressure turbine (HPT) inlet are specified as obtained from a combustion chamber simulation. At the exit guide vane (EGV) outlet, the back- pressure is taken from a one-dimensional performance calculation. For the inter-row coupling between rows of HPT and low-pressure turbine (LPT), as well as between both components, direct interfaces are used for both steady and unsteady simulations, i.e., the steady-state simulation is a frozen-rotor calculation. Unsteady simulations are restarted from the steady results and were performed for nine sectional revolutions after which they achieved convergence. Cooling mass flows in the HPT have been neglected to reduce the complexity and the duration of the simulation. From this turbine simulation, the core stream boundary conditions for the exhaust jet simulation are provided. The exhaust jet is simulated in a free stream condition using the data of a performance simulation for the boundary conditions in the free and bypass stream.

Experiments on an aircraft engine

In addition to the numerical simulations, experiments with an aircraft engine are carried out to validate the results of the numerical simulation and to test the BOS method under real conditions. In the second experimental test, the research engine V2500 of the Institute of Jet Propulsion and Turbomachinery (IFAS) at Technische Universität Braunschweig is used for the investigation of the BOS method in the test facility of MTU Maintenance in Langenhagen. Within the scope of an engine test, the fuel supply of a burner is gradually throttled in order to simulate a burner defect, similar to the tests with the model combustion chamber. The resulting inhomogeneous density field at the engine outlet is then reconstructed using the BOS method, Fig. 7. A special measurement ring was designed and manufactured for the test.

Fig. 7

Investigation of the BOS method on a real aircraft engine

The ring has a weight of 509.75 kg and an average diameter of 4.15 m. The entire ring construction consists of two halves assembled together. One half is used as a carrier for 15 DALSA GENIE NANO-M4060 cameras with RICOH FL-BC2518-9 M lenses of 25 mm focal length. The distance between the cameras is 8°, equidistant. A retroreflective foil with a defined dot pattern is applied to the second half. The dot diameter is 2.8 mm and the dot density 35%. The cameras are mounted to the central ring strut of the BOS ring by means of a manually adjustable U-profile support and a holding device. The orientation of each camera can be adjusted so that it coincides with a fixed target position. Each camera position is marked by means of attached calibration crosses on the opposite retroreflective foil. Before the engine test, the alignment of each camera is checked and, if necessary, corrected by means of a calibration in order to avoid measurement errors. Six mounting plates were designed with which the ring can be fixed to the eyelets in the exhaust duct, thus mounting the BOS ring in the exhaust duct of the test cell. The experimental setup in the MTU test cell with the instrumented engine is shown in Fig. 8. To prevent axial movement of the ring, retaining clips were used on all 6 plates.

Fig. 8

Experimental setup consisting of the V2500 research engine of the IFAS and the BOS ring in the MTU Maintenance test cell. (Picture used with kind permission of MTU Maintenance and IFAS)

A scheme of the setup is shown in Fig. 9. A 16 Amp three-phase connection was used for the power supply. A power distributor supplies 5 power supply units per phase; each phase is connected to a camera and a ring light. During each measurement, the recorded images were transmitted via a switch to the PC, which is located in the transition room next to the test cell. The cameras can be controlled via the PC by means of a measuring program in order to trigger the flash lights synchronously. Before each measurement, a setup file must be loaded into the program which defines camera settings such as exposure time, the region of interest (ROI) and frame rate of the image recordings. For the present experiment, 1000–1200 images were taken at each operating point with a frame rate of 14 Hz. The exposure time is 110 μs and the ROI 4112 × 900 px is used.

Fig. 9

Schematic of the experimental setup

In addition to the BOS measurement system, a separate monitoring system is used to observe safety-relevant variables on the BOS ring during the test and to determine the pressure and temperature boundary conditions near the ring for the BOS measurement. For this purpose, thermocouples type J were used on the 6 mounting plates to monitor the temperature and vibration sensors from Conplatec (Brüel & Kjaer 4508) to observe the ring vibration caused by the unsteady aerodynamic load from the exhaust jet and the engine acoustics. Prandtl tubes were used to determine the pressure and temperature boundary conditions near the ring.

For the test, the temporal profile of the engine’s thrust was determined in order to reach the desired operating points of the engine (Fig. 10). A distinction is made between the four load bands A, B, C, D, whereby load band A with 102.7 kN represents the highest engine thrust and D with 42.75 kN the lowest. The dwell time on all load bands except A is 6 s. On load band A the dwell time was set to 3 s to reduce the risk of damage to the test structure due to the high thermal and mechanical stresses. In order to ensure thermal equilibrium of the engine despite the short dwell time, the engine is first stabilised on load belt B for 3 s before it is moved up to load belt A. Before the test, the engine runs through a system check to ensure that all systems are functioning correctly. Then, the reference thrust curve is first run without burner defect to generate the reference images for the BOS measurement. After the reference measurement is completed, the damage case with 100% closed fuel spray nozzle is first initiated by replacing a single burner. At the same time, a mechanical check of the BOS ring is carried out to assure the integrity of the bolt connections. Finally, the damage case with 50% throttled fuel nozzle is examined.

Fig. 10

Thrust curves of the engine during the experimental test (provided by IFAS)

Results

The main goal of the BOS measurements on the aero engine is to detect the cold streak introduced by the throttled or closed fuel spray nozzle. To investigate the propagation of the cold streak through the engine, first RANS simulations were performed. Results of the simulations are shown in Fig. 11. The normalized density is defined by

$${\bar{\rho}} \, = \,\frac{{\rho - \rho_{\min } }}{{\rho_{\max } - \rho_{\min } }}$$

(1)

Fig. 11

Results of RANS-simulation of exhaust jet of aero engine

High temperatures result in a low normalised density; cold streaks result in an increase of the normalised density. To illustrate the changing shape of the cold streaks, three cross-sections are extracted from the simulation:

• A. nozzle inlet

• B. nozzle outlet

• C. BOS measurement plane.

At the nozzle inlet plane (A), the cold streak can be seen as a round shape in the upper right part of the normalised density field. A swirl is introduced by the exit guide vanes (EGV), deflecting and deforming the cold streak in the nozzle’s outlet plane (B). The influence of the EGV can also be seen by the flower shape of this plane. Further downstream at the BOS measurement plane (C), the cold streak is even more deformed and less distinguishable. The area of the low normalised density plane is also reduced, because the hot gases mix with the bypass and ambient air.

Figure 12 shows the measurement results from the BOS measurements with the aero engine. The normalised density field is shown for the reference case with all fuel spray nozzles opened and with one fuel spray nozzle 50% and 100% closed respectively. All density fields have the same shape. This is expected, as the cold streak is predicted not to alter the shape of the density field. Contrary to the prediction by the RANS simulation, a cold streak is not visible for the measurements with the closed fuel spray nozzle. The measured density field is not as symmetric as the simulation results, and all density fields are shifted upwards. Four main aspects were identified to explain the discrepancies between the simulation and the experiment:

Fig. 12

Reconstruction of normalised density field from aero engine test

• 1. Structures in the test facility, like the aero engine’s ceiling support, influence the exhaust gas jet and the flow around the engine. These effects are not modelled in the RANS simulation and were underestimated before the experiment. The uneven flow around the aero-engine can explain the upwards shift of the exhaust gas jet and might lead to a deformation of the density field and an underestimation of the mixing behaviour of the exhaust gas jet. A better insight into the aerodynamics of the test facility will be gained by a simulation of the complete test facility currently being carried out by the Institute of Jet Propulsion and Turbomachinery, IFAS at Technische Universität Braunschweig.

• 2. The upwards shift of the exhaust gas jet is leading to issues concerning the reconstruction of the density field. The ambient temperature and pressure are set as boundary conditions on the boundary of the reconstruction. Because of the upwards shift, the density field is no longer in the middle of the reconstruction area and the reconstruction boundary is intersecting with the density field. This leads to errors at the reconstruction boundary and can deform the reconstructed density field. These issues will be addressed in the upcoming second measurement by a rearrangement of the cameras. In addition, more temperature and pressure measurement points will be added, reaching further into the hot gas stream, to decrease the uncertainties of the boundary conditions.

• 3. The missing cold streak in the measurement might be explained by an underestimation of the mixing of the exhaust gas in the RANS simulation or by the reconstruction errors. To reduce the influence of the mixing of hot and cold gas and of the upwards shift, it would be preferable to decrease the distance between the engine’s outlet and the measurement plane. However, this is not possible in the test facility used for these measurements, due to constructive restrictions.

• 4. The exhaust gas jet can have an uneven distribution of exhaust gases. The composition of the exhaust gases influences its optical properties. To consider the influence of the inhomogeneous gas, constant values need to be defined locally for upcoming reconstructions.

We are looking forward to addressing as many of the mentioned aspects in the upcoming measurement campaign. Unfortunately, the date for the measurement is after the publication date of this report. We hope, that the interested reader will follow our upcoming publications, where we will present results of this upcoming measurement campaign.

6 Conclusions

During 12 years of research many steps towards the detection of hot gas path defects by background-oriented schlieren (BOS) measurements have been completed. Initially, the algorithms for tomographic reconstruction were improved for better accuracy of BOS measurements. The filtered backwards projection (FBP) and the algebraical reconstruction technique (ART) combined yielded the initialised algebraic reconstruction technique (iART).

iART’s mean deviation in the density compared to numerical simulations is 0.0257, which is a significant improvement over FBP and ART, for which the corresponding deviations are 0.1 and 0.0547.

Following the improvement of the reconstruction algorithm, the BOS method was used to detect defects in a model combustion chamber, in which the influence of characteristic burner damage on the density distribution in the exhaust gas jet was investigated. For this purpose, the fuel supply to a single fuel spray nozzle was gradually reduced. This resulted in a momentum deficit, which allows dense ambient air to enter into the exhaust jet of the combustion chamber. The measurement data was then used to train a support vector machine (SVM) algorithm for the automated classification of burner damage. It was shown that by choosing suitable integral parameters to describe the normalised density distribution, a high probability of successfully classifying burner failures can be achieved even with a small number of parameters. The best result was achieved with a choice of 5 parameters. Thus, 99% of all defect cases were correctly classified and 97% of all reference cases.

After the successful application of the BOS method on a model combustion chamber, a real engine was considered. First, numerical investigations were performed using RANS simulations to investigate the influence of combustor damage on the density distribution of the exhaust gas jet. The results showed that a cold streak forms at the combustion chamber outlet as a result of a burner failure. Despite mixing with the surrounding gas in the hot gas path (HGP) of the engine, the cold streak could also be observed at the exit of the engine, which indicates an existing influence on the density distribution in the exhaust gas jet.

Further experiments on the research engine V2500 validate the numerical simulations of the reference configuration. For the application of the BOS method a specially designed measurement ring was used for mounting the cameras. The experimental data also showed that, in contrast to the RANS simulations, there is an asymmetry in the density distribution. Furthermore, the existence of a cold streak in the exhaust gas jet could not be confirmed for the investigation case with 50 and 100% throttled fuel spray nozzle. Finally, four possible causes for the occurrence of this deviation were identified and solutions were described which will be implemented in the upcoming measurement campaign.