Anda di halaman 1dari 8

Proceedings of ASME Turbo Expo 2010: Power for Land, Sea and Air GT2010 June 14-18, 2010,

Glasgow, UK

GT2010-22239
DEVELOPMENT IN GAS TURBINE REPAIRS
Allan Thomson Advanced Parts Manufacture David A Anderton Advanced Parts Manufacture

Wood Group Light Industrial Turbines Ltd. Kirkhill Drive, Kirkhill industrial Estate Dyce, Aberdeen, UK

ABSTRACT The need for repair rather than replace of gas turbine components is becoming increasingly important to operators in todays economic climate. The use of commercially available numerical analysis software, computational fluid dynamics (CFD) and finite element (FE) have become well established within Wood Group Light Industrial Turbines Ltd. They have allowed the business to be extremely competitive by being able to rapidly respond to a customers request for a repair which may involve a fluid structure interaction and/or conjugate heat transfer analysis. The software has also been used to study critical design limitations and to rapidly enhance reverse engineered parts. Two such cases are presented here: the repair of a compressor rotor blade airfoil and the changes made to an existing design of a cooling passage in a high pressure turbine rotor blade. Each analysis was completed in a very competitive time span. INTRODUCTION In todays economic climate, the repair, rather than the replacement of gas turbine parts during overhaul has become paramount in the drive to reduce costs. If the repair could be done in situ rather than return to shop then many thousands of pounds can be saved in shipping and downtime. Wood Group Light Industrial Turbine Ltd (WGLIT) being a service repair company has had to rely on the original equipment manufacturer (OEM) for a supply of spare parts and experience has shown that part delivery lead times can significantly impact service schedules or parts may no longer

be available for older turbine types. WGLIT has therefore embarked on a reverse engineering programme of a number of gas turbine components to ensure continuity of supply and to manage costs more effectively. WGLIT have invested in both computational fluid dynamics (CFD) and finite element (FE) numerical analysis software to typically determine rapidly whether a repair to a damaged component is possible or not, to determine any impact on component life and the effect on efficiency of the engine. The software has also been used extensively to determine tolerance setting, critical dimensions and the performance envelope of the reverse engineered parts. Two cases are presented here: the repair of a compressor blade trailing edge airfoil and the enhancement made to the cooling capability in a high pressure turbine rotor blade. NOMENCLATURE m T p mass flow rate (kg/s) temperature (K) pressure (Bar)

COMPRESSOR AIRFOIL TRAILING EDGE REPAIR During the assembly of a 36 MW generator set, it was discovered that some of the trailing edge of the 6th stage compressor rotor blade was damaged close to the platform, Figure 1. It was not normal practice to repair in this area, but due to the down time and cost of replacement, a repair option was considered.

Copyright 2010 by ASME

As large financial penalties were being incurred for each day the engine was down, time was of the utmost importance. A system was required that could accurately reproduce the blade geometry in the shortest possible time for analysis. A technique was used that was based on the principle of triangulation, that is, projected fringe patterns were observed with two cameras. The three dimensional coordinates for each camera pixel was calculated with high precision and a triangulated surface mesh of the object was created. Using this method it was possible to create a closed very high quality stereolithographic (stl) triangulated surface in two hours. The accuracy of the model can vary with size, and experience shows that for an airfoil surface the variation was probably less than 0.15 mm. Later to continue the analysis it was necessary to convert to the standard exchange of product (step) format which took a further six hours.

mode shapes. The stl file was therefore converted into a step file which could be read by ABAQUS. The second stage of the analysis was to perform the linear elastic stress analysis in ABAQUS along with a natural frequency and normal mode calculation. The first 10 natural frequencies and normal modes would be compared on the undamaged and repaired blade. The boundary conditions were the same as the first stage of the analysis. Each model consisted of about 56,000 second order tetrahedral elements.

Figure 2: Trailing edge showing proposed repair In order to determine the pressure loading on the blade , the third stage of the analysis was completed as a loosely coupled fluid structure interaction, i.e. the movement of the blade due to aerodynamic loading was small and had little effect on the flow field. A 1/56th of a single row of compressor blades was modeled. The fluid domain inlet and outlet boundaries being set 1.5 axial chord lengths upstream and downstream of the blade respectively. The overall compressor pressure ratio was 11.5:1 and pressure rise across each stage was assumed to constant. Bleeds for cooling and sealing air were unknown and therefore neglected. The polytropic efficiency of the compressor was calculated from field data to be 90%. No geometry information was available with respect to the upstream stator vane therefore the swirl velocity was unknown, as was the static pressure at the exit from the blade row. A number of numerical runs were carried out on a course grid to determine iteratively the inlet swirl velocity and the blade row exit static pressure. That is: (1) Guess average blade passage exit velocity (2) Calculate static temperature and pressure based on this velocity from stagnation values (3) Run the analysis (4) Does average blade velocity equal guess from above? (5) Yes, stop. No repeat the process

Figure 1: Detail of trailing edge damage The analysis was carried out in three stages, only moving on to the next stage if the results from the previous stage were shown to be satisfactory. The first stage of the analysis was to determine whether a repair was possible or not using a linear elastic stress analysis. At that time it was not possible to read a stl file into the finite element software ABAQUS [1], therefore STARCCM+ [2] was used for the initial stress analysis. A worst case approximation of the defect was applied to the model, i.e. some computational cells were removed from the trailing edge. Two numerical runs were carried out, one with the defect and one without. The blade rotated at 5100 rev/min and was held rigid at the root. Each model consisted of about 500,000 polyhedral computational cells. Although STARCCM+ can solve linear elastic stress problems it cannot determine natural frequencies or normal

Copyright 2010 by ASME

The analysis was repeated until the guessed and calculated values were within 1% of each other. Each model consisted of about 550,000 polyhedral computational cells, 350,000 in the fluid region. The near wall region consisted of four prism layers. The flow was assumed to be turbulent and was modeled with the k- turbulence model with all Y+ wall functions. Determination of failure due to high cycle fatigue would be evaluated by a mean and alternating stress and their position with respect to an endurance limit on a fatigue diagram. In order to determine whether the blade would fail due to high cycle fatigue the effect of the repair induced stress concentration factor on endurance limit would have taken into account. Sawyer [3] states that for many engineering materials, if the component lasts to 107 cycles at a given mean and alternating stress then it can be considered to last indefinitely i.e. the endurance limit. The theoretical stress concentration factor and a notch fatigue strength reduction factor can be expressed by a factor q:

Figure 3: Comparison of STARCCM+ (top) and ABAQUS results. The second stage results of the static stress analysis from ABAQUS were compared with the results from STARCCM+. They were encouraging as the model had been solved by two completely different methods, finite volume in STARCCM+ and finite element in ABAQUS. Figure 3 illustrates the similarity in results despite applying about six times more computational cells in the STARCCM+ analysis. The predicted maximum principal stress from the analysis of the damaged and undamaged blade are shown below in Figure 4.

q=

KF 1 kt 1

(i)

Crookson [7] states that KF is always less than kt. Peterson and Pilkey [8] suggest that if KF was unknown, put KF = kt i.e. q = 1.0. This would give a result which was conservative.. The endurance limit modification factor, ke was then found from ke = 1/KF Compressor Analysis Results The results from the first stage of the analyses showed that there was not a dramatic increase in stress in the area of the defect and a repair may be possible. A more detailed analysis would be required with a better representation of the repair, Figure 2.

Figure 4: Comparison of the undamaged and repaired blade

Copyright 2010 by ASME

Frequency Diagram
1000 900 800 700 vibration frequency (Hz) 600 500 400 300 200 100 0 0 1000 2000 3000 engine speed (rev/min) 4000 5000 6000 1st EO 2nd EO 3rd EO 4th EO 5th EO 6th EO 7th EO 8th EO 9th EO 10th EO First Natural Frequency

Figure 5 Frequency diagram The increase in maximum principal stress was about 50%, which was quite considerable. However at that temperature the value of stress was still below 50% of the 0.2% yield stress of the material. The first ten natural frequencies were calculated of the rotationally loaded repaired and undamaged blade. The difference in their values was less than 1% at each natural frequency, the repaired blade being less stiff. A plot of the undamaged blade first natural frequency, and the first ten engine orders are shown in Figure 5. The first and seventh natural frequencies were the only two that may have caused concern as the first was close to the fifth engine order and the seventh close to the stator passing frequency. Only the first natural frequency and engine orders are shown on Figure 5 for clarity. The interaction of the natural frequencies were unlikely as there hasnt been a history within Wood Group of 6th stage compressor rotor blade failures. Up to this point no account had been taken of the dynamic loading on the blade. This was difficult to quantify accurately without mounting strain gauges to a compressor blade and conducting engine run trials. However numerical analysis has shown that the range of normalized alternating stresses in mode 1 had increased by 30% due to the repair. It was likely

that this compressor blade had very low alternating stresses, as, again, there have not been any reported failures of this blade. Figure 6 shows that when the failure line was plotted for a stress concentration factor of 1.68 the alternating stress was still higher than the mean stress, for failure to occur at 107 cycles. This was highly unlikely to have occurred in practice. The unknown alternating stress was one of the limitations of the analysis and would have to be borne in mind during the decision making process. The results of the second stage analyses were encouraging and further work was required to determine the coupled aerodynamic and rotational loads on the blade. The coupled analysis showed that the increase in maximum principal stress in the repaired blade was 68%. The repaired value was 51% of the 0.2% yield stress. Using the theoretical stress concentration factor from this analysis, in equation (i), meant that the specimen endurance limit was reduced by 41%. Since the repair would only be required on three blades out of 56 blades the calculated increase in polytropic efficiency was found to be negligible. The analysis was completed in about four man days and a quantitative engineering solution was provided based on analysis to a complex problem. With the predicted 41% reduction in fatigue life the customer was able to quantitatively

Copyright 2010 by ASME

Figure 6 Fatigue life diagram assess the risk of the continued running to the next service interval. HIGH PRESSURE TURBINE ROTOR BLADE The reverse engineering process of a triple pass air cooled turbine blade was complex. It was relatively straightforward to determine material, casting specification, measurement and checking of the external surfaces. One of the most difficult parts to reverse engineer was the cooling passage as without having access to the original design intent, this must be established from engineering principles. As a starting point, nominal geometry for the internal cooling passage was established by sectioning the blade at a number of radii, Figure 7, measuring the cooling passage profile at various cross sections and assembling the resulting geometry in CAD. During manufacture a number of factors influence blade wall thickness and cooling passage geometry which in turn affect the overall blade thermal response. Typically ceramic core positioning, movement and shrinkage as well as alloy shrinkage during cooling need to be carefully accounted for. These factors can usually be overcome by the experience held by the casting company. Once the reverse engineered blade meets the defined geometrical design intent, it is then flow tested and compared with the OEM blade.

Figure 7: Section being measured on the CMM As WGLIT are not the OEM they are not privilege to cooling flow design data, therefore the range of acceptable cooling flow rates were unknown. This information could only be obtained by flow testing OEM blades. A large batch of new OEM, and used blades that were fit for service were flow tested with air and the pseudo non dimensionless parameter was recorded.

m T p

Copyright 2010 by ASME

Turbine Cooling Passage Analysis A small batch, nine, of reverse engineered blades were produced and flow tested. The results showed that the flow rate was about 8% less than the OEM. This result was totally unacceptable as small increases in metal temperature at elevated temperatures result in large reductions in creep life. A CFD analysis was proposed to determine the cause of the low flow. To modify the original core die was an expensive process and if some dimensions were made too large then it would be costly to recover it. Deadlines for production were also becoming tight therefore it had to be a right first time modification rather than an iterative process. A 600,000 polyhedra cell grid with four prism layers was constructed in STARCCM+ and solved in STAR-CD of the cooling passage only and is shown in Figure 8. The flow was modeled as compressible air, turbulence was simulated by the k- turbulence model with hybrid wall functions in a rotating reference frame.

effect on flow rate i.e. for the same pressure drop the flow rate went down in each case. A closer examination of the flow in the cooling passage suggested that if the flow in the first pass was slowed down and the volume of the plenum chamber at the top of the blade was increased, a more even distribution in flow down the second pass was likely. These analyses were completed and the results are shown below in Figure 9.

Figure 8: Computational grid Initially it was thought that the low flow could be due to a problem with the blade shrinking too much on cooling and the flow was choking somewhere in the cooling passage. This was quickly dismissed as a probable cause as analysis showed that the blade would only choke on the third pass when the mass flow rate was doubled. Attempts were made to increase the flow area of each pass in turn. This can easily be achieved in STAR-CD by applying shell cells to the surface in question then extruding them normal to the surface. Models could be rapidly changed in minutes without the need to revert back to the original CAD geometry. Unfortunately these modifications had the adverse Figure 9: Velocity distribution original (top) and modified core It was clear from Figure 9 that the velocity distribution was more uniform in the modified core, particularly in the third pass. This was borne out by the fact that the flow rate had increased by about 11% due to these modifications. A CAD model of the modified core was produced and a CFD analysis was carried out using the same grid settings and boundary conditions as the previous models. The flow through the core had increased to 12.5% greater than the original. The increase of 1.5% over the CFD modified core was probably due to the smooth transition of surfaces rather than a step change where

Copyright 2010 by ASME

the grid had been modified. A check on stress in the cross section showed that the increase was minimal. A 12.5% increase in flow over the original meant a 4.5% increase in flow of that of the OEM. However this was a calculated increase and it was unlikely that it would be achieved in production. The increase in flow would probably be smaller. It was decided to produce a small batch of blades with this core. Twelve blades were produced and sent off for flow testing along with two blades cast with the original core as a control sample. The increase in average flow rate between the original and modified core was found to be 12%. It has to be borne in mind that the sample size here was quite small. The objective had been achieved in one analysis iteration. The increase in flow could have a significant effect on the life on the blades, and therefore could be a powerful selling point. However it should be borne in mind that there would be a slight reduction in power output, less than 0.5% of the compressor turbine total power output. Further analysis was required. At this point it was decided to determine the design metal temperature at the critical radius using data from Wood [5] and Hannis and Smith [6]. A small spreadsheet was written using the method devised by Fullagar [4] to do this. Hannis and Smith [6] quote a cooling effectiveness of 0.22 for compressor turbine stage 1 rotor blade of similar design and life of 40,000 hours, Wood [5]. A CFD model comprising of 1.6 million polyhedra cells was constructed of the complete blade, cooling passage and surrounding fluid zone. External fluid was modeled as compressible combustion products and the cooling fluid as compressible air. Both fluids used polynomials as a function of temperature for the following properties: specific heat, thermal conductivity and molecular viscosity. Clement and Palfreyman [10] show that when compared with data the k- turbulence model over predicted surface temperatures relative to the k- model, therefore turbulence was modeled using the k--SST model with all Y+ wall functions. Boundary layer transition and unsteady passing wakes from upstream stator vanes were not taken into account due to time and resource constraints. The blade was taken to be Inconel 939 with variable specific heat, thermal conductivity and Youngs modulus. The model was a steady state analysis solved in a rotating reference frame. A full conjugate heat transfer and loosely coupled thermal and rotational stress analysis was done at

what was determined to be the OEM lowest measured flow rate to give the minimum life expected from a blade. Wood [5] gave values for mid-span reaction, stage loading coefficient and flow coefficient. From this data the inlet flow angle was calculated for the rotor for a design running speed of 11085 rev/min. The cooling flow for the first stage compressor turbine was also quoted at 4.7%. The cooling flow split between the stator vane and the rotor blade was kept the same as that of Hannis and Smith [6]. The radial temperature distribution factor was published by Alkabie [9]. Turbulence intensity was taken to be 4% at both the cooling and external inflow boundaries and the length scale was 10% of the corresponding domain width. Boundary conditions have been determined from the best available data and the authors experience. The prediction of average metal temperature at the critical radius was lower than that predicted by the method by Fullagar [4] by about 30 K. A similar analysis was carried out at the WGLIT lowest flow rate and the difference in temperature and radial stress between the two models are shown at the critical radius below in Figures 10 and 11. Reducing the amount of air to the rotor from 2.1% to 1.9% gave an increase in average metal temperature of 3K. As shown below the maximum difference in temperature was found to be 6 K and small increase in maximum principal stress. At the critical radius the difference in average temperature was found to be 5 K. It was known that these models would not be able to make a reasonably accurate prediction of the average metal temperature, however the trend would be correct. To improve the prediction of metal temperature, a low Reynolds number k--SST turbulence model was required to simulate the heat transfer to the blade. A feature of this turbulence model is the requirement to keep Y+ at about 1. 20 prism layers were required next to the surface of the blade both internally and externally which increased the model size to approximately 6.7 million polyhedra computational cells. The predicted average temperature at the critical radius had decreased by about 10 K. This value was well below the temperature predicted by Fullagar [4] which was probably a bit high. A Larson-Miller analysis showed that for a decrease of 5 K at the critical radius it was probable that the creep life of the blade could be extended by at least 25%.

Copyright 2010 by ASME

passage that resulted in a better flow distribution and therefore cooling capability than the OEM at the expense of a slight reduction in performance of about 0.5%. The modification increased the predicted creep life of the blade by at least 25%. The reverse engineered blades will be introduced as part of a monitored development programme and samples will be removed for assessment at planned intervals. ACKNOWLEDGMENTS The support of Computational Dynamics Ltd, London is gratefully acknowledged. REFERENCES Figure 10: Reduction in metal temperature (K) [1] [2] [3] Simulia, 2008. ABAQUS v6.8 www.simulia.com . CD-adapco, 2008. STAR-CD V4.06 and STARCCM+ 3.06. www.cd-adapco.com . Sawyer, J. W. 1972. Sawyers Gas Turbine Engineering Handbook. 2nd ed. Gas Turbine Publications, Stamford, Connecticut, USA. Fullagar, K. P. L. 1973. The Design of Air Cooled Turbine Blades. Design and Calculation of Constructions Subject to High Temperature symposium, University of Technology, Delft Wood, G. R. 1981. The Ruston Tornado. A 6 MW Gas Turbine for Industrial Application. ASME 81-GT-171. Hannis, J. M. Smith, M. K. D. 1982. The Design and Test of Air-Cooled Blading for an Industrial Gas Turbine. ASME 82-GT-229 Crookson, R. A. 1989. Cranfield University short course: Introduction to Gas Turbines, section SME 2198 Properties of Materials: Fatigue. Peterson, R. E. and Pilkey, W. D. 1997 Petersons Stress Concentration Factors, 2nd ed. pub. John Wiley and Sons Alkabie, H. 2000. Design Methods of the ABB Alsthom Power Gas Turbine Dry Low Emission Combustion System. Proc. IMechE vol. 214 Pt. A pp 293-315. Clement, J. and Palfreyman D. 2009. NASA 3cX Turbine Vane Validation. CD-adapco internal validation presentation.

[4]

[5] [6]

[7]

[8] Figure 11: Increase in maximum principal stress (MPa) [9] The above analysis has shown that using CFD it was possible to successfully reverse engineer a cooled turbine rotor blade and improve on its cooling performance through minor modification of the internal cooling passage. CONCLUSION The cases presented in this paper have shown that, with the use of numerical analysis it was possible to formulate a repair to a compressor rotor blade that was previously thought not to be possible. The software allowed a complex analysis to be carried out in a very short time span and a decision whether to repair or not was made based on quantifiable results. The reverse engineering of the turbine blade cooling passage was achieved with one modification to the cooling

[10]

Copyright 2010 by ASME

Anda mungkin juga menyukai