Abbreviations

CAD

computer-aided drawing

FEA

finite element analysis

HP

high pressure

HPT

high-pressure turbine

IP

intermediate pressure

IPC

intermediate pressure compressor

IPT

intermediate pressure turbine

LHM

latent heat of melting (energy/volume)

LP

low pressure

LPC

low-pressure compressor

LPT

low-pressure turbine

NGV

nozzle guide vane

Symbols

A

contact area

Cp

specific heat

F

force

R

radius

T

temperature

V

velocity of disk

β

heat partition ratio, relative flow angle

ρ

density

μ

friction coefficient

ω

angular speed

2.1 LS-DYNA verification against welding experiments

‘Continuous-drive’, friction welding experiments of two steel bars were published by Bahrani and Duffin⁽ Reference Bahrani and Duffin ^{16 )} and Bahrani et al⁽ Reference Bahrani, Healy and Mcmullan ^{17 )} in terms of axial shortening rate, equilibrium torque, coefficient of friction at the welding interface and width of plastically deformed region. They also presented the derived interface temperatures and metal viscosity according to their calculations.

Continuous-drive friction welding is a metal bonding process including four phases where the speed of the rotary specimen is controlled by the direct drive. In the first phase, a rotating specimen is accelerated by a direct drive and pushed against a stationary piece. The temperature of the friction interface increases during this phase. In the second, axial shortening of the welded specimens starts. The removed material is centrifuged in the radial direction and this is called radial upsetting. The third phase is the equilibrium phase where the rotational speed, axial force, friction torque at interface, interface temperature, axial shortening and radial upsetting ideally remain constant. In the fourth phase, the rotating specimen is brought to rest and the axial forging continues to ensure the weld quality.

In this study, the LS-DYNA FEA verification is done at the equilibrium phase, and the simulation results are compared against reported experimental values. The welded rods have an inner diameter of 12.7mm and an outer diameter of 19.0mm⁽ Reference Bahrani and Duffin ^{16 ,} Reference Bahrani, Healy and Mcmullan ^{17 )}. The axial length of the each welded rod is 40.0mm in the FEA model. Three models with different mesh densities are used, namely, Mesh 1, 2 and 3. Total numbers of elements are 15,360, 19,200 and 28,800 for each rod, respectively. The different parameters in the mesh size are the length of the elements in the axial direction near the welding interface. Mesh 1 has four units of element axial length. Mesh 2 has two units, and Mesh 3 has one unit of element axial length. After a predefined distance from the friction interface, the element length increases as illustrated in Fig. 3(a). The mesh density in the radial direction is the same for all three cases and is given in Fig. 3(b). Figure 3(c) describes the process where one element is rotated and forged into a stationary element.

Figure 3

(Colour online) Finite element model of the steel bar friction welding experiment. (a) Three different meshes from front view, (b) side view of all three meshes and (c) Description of the friction welding.

The material model used in the FEA computations is the ‘Johnson–Cook’. This material model is composed of two parts, namely the constitutive and the damage model. The constitutive model provides an approximation to the yield strength of the material, and the damage model provides the effective plastic strain at failure. The failure strain represents a cumulative damage on a structure. The material model also includes the effects of the temperature and strain rate in the constitutive and damage models, respectively. The properties of steel are reported by Johnson and Cook⁽ Reference Johnson and Cook ^{18 )}.

A ‘thermal–structural’ calculation is invoked in LS-DYNA. The rotating bar is modelled with rigid solid elements, and the static bar is modelled with deformable solid elements. Before the simulation of the equilibrium phase in the steel bar welding experiment, the one-dimensional transient temperature distribution across the rod is calculated by an explicit finite difference calculation. The obtained temperature distribution across the static steel rod is given as an initial condition to both rods at the beginning of the equilibrium phase simulation. A prescribed rigid body rotation is given to the rotating bar, with an initial angular speed, and this is then pushed against the static bar with constant axial force. Between the rotating and the static components, the ‘eroding surface to surface’ contact is defined. Metallographic examination of the interface region for the investigated experimental conditions showed that the deformation of the rotating and static parts is symmetrical in the axial and radial directions⁽ Reference Bahrani and Duffin ^{16 )}. Using the rigid formulation for the rotating bar, the ‘symmetry’ condition dictates that the simulated axial shortening rate will be half the experimental value, since only the deformable elements will erode. This is done to simplify the model and reduce the associated computational cost.

During the FEA simulation, the eroded elements are removed from the computation and there is no formation of radial upsetting. The axial displacement of the rigid rotating bar is plotted in Fig. 4. The slope of the axial displacement is the axial shortening rate. In Fig. 4, the slope of the straight full lines represents the experimental axial shortening rate at 16,000N axial force and 416.4 rad/s angular speed. The dashed lines are the obtained trajectories in time of the simulated rotating specimens. The trajectories of Mesh 2 and Mesh 3 are shifted in the y-axis direction for illustration purposes. The effect of mesh size on displacement in the axial direction can be clearly seen. Every step movement occurs with the deletion of a mesh row. Mesh 3, which has the highest density and friction interface, shifts from the experiment value by about 5% at around 0.2s. Since the mesh is only increased in the axial interface, the aspect ratio becomes higher having a value of 4. The other meshes have aspect ratios of 2 and 1 for Mesh 2 and Mesh 1, respectively. The numerical procedure can be improved by keeping the element density high, while keeping the element aspect ratio close to 1. The reproduction of the results and the good agreement with the experimental observations allowed for the simulation methodology to be subsequently used in the post shaft failure simulation of unlocated turbines.

Figure 4

Axial displacement of the left steel bar for the three meshes.

3.1 Performance model: Axial load and gas torque on turbine

The axial load on the turbine wheel depends on various aerodynamic and mechanical factors during the post shaft failure. These factors can be listed as compressor performance, secondary air system response and turbine performance.

Handling of the complete event requires a sophisticated transient model and in this study, for design purposes, a reduced order approach with simplifying assumptions is used to calculate the turbine torque and the axial force acting on the wheel. Therefore, an atmospheric discharge of the gas at constant turbine entry temperature is considered. Then, the turbine performance calculations are carried out using a ‘mean-line’ analysis.

Using a time-varying total pressure at the turbine inlet, the axial load and the gas torque are calculated for each time step. The total pressure evolution at the inlet of the turbine rotor is given in Fig. 5. The decay also includes the presence of compressor pulsation which can be seen as the difference between the full and dashed lines. It is possible that there will be partial or full recovery of the compression system after a shaft failure event which might continuously provide HP gas at the turbine inlet at a reduced level^{( 1 )}. The assumptions for the performance calculations are listed below:

• ∙ Discharge of the gas from combustor at constant temperature.

• ∙ The turbine total to total pressure ratio and efficiency are constant.

• ∙ The effect of cooling mass flow rate is not taken into account in continuity.

• ∙ The flow enters to the rotor row with zero incidence and exits with zero deviation.

• ∙ The upstream turbine disk cavity is at the same pressure with the upstream gas path.

• ∙ The downstream turbine disk cavity pressure is equal to the downstream gas path.

Figure 5

Discharge of gas from upstream combustor reservoir.

During intact operation, the gas path is sealed from the disk cavities in order to minimise leaks. However, with the dislocation of the turbine, the seals open at the front of the disk to equalise the pressures between the gas path and the disk cavity. At the rear of the disk and with the movement of the disk, the air will be kept sealed and will maintain its HP level or the damaged seals will allow the equalisation of the pressures between the gas path and the rear disk cavity. This condition requires a transient modelling of the secondary air system. According to the conditions mentioned above, the disk cavities are assumed to be at the same pressure with the gas path without loss of generality.

Haake et al⁽ Reference Haake, Fiola and Staudacher ^{15 )} plotted the change in turbine inlet total temperature as a function of time during turbine dislocation. Assuming a reference temperature of 1,000K, the difference between the turbine inlet temperature at the beginning and at the end of the event will be in the order of 100–300K. However, the turbine inlet temperature is also a function of the engine control system which shuts the fuel flow after the detection of the shaft break. Additionally, the turbine performance is affected by the dislocation which may drastically increase tip clearance losses if the flow path has a high hade angle. Moreover, during the event, the axial velocity of the flow entering the stage should be expected to be either constant or decreasing. However, with the increase in the rotational speed, the velocity triangles are heavily distorted to result in the decrease of turbine expansion and efficiency. The simplifying assumptions in the model can lead to discrepancies from a real event; however, these assumptions can be considered as a worst-case scenario making the methodology applicable for a structural design study.

3.2 Friction model

The contact between the rotating turbine wheel and stationary components involves disk rubbing and blade interference. The frictional torque at the contact interface is calculated by a friction clutch approach. The pressure is assumed to be constant across the contact surface and the corresponding frictional torque is calculated.

The calculation of the axial dislocation rate of the turbine is based on the application of a solid melting process, as introduced in shaft failure modelling by Gonzalez⁽ Reference Gonzalez ^{19 )}. The solid melting process was proposed by Carslaw and Jaeger⁽ Reference Carslaw and Jaeger ^{20 )}.

By neglecting radiation and surface convection, a 1-D heat transfer is assumed (Equation (1)). The flux term in Equation (2) includes a conductive and a convective term for the heat flow. The heat is generated at the interface by friction. Due to the high turbine end load after the shaft break and the high rotational rubbing speed, the large frictional power heats up the frictional interface to the melting temperatures rapidly. Based on this condition, a quasi-steady temperature distribution is assumed. By inserting Equation (2) into Equation (1) and removing the transient term (Equation (3)), the steady temperature distribution is given in Equation (4) with the boundary conditions which are given by Equations (5)–(7). The frictional power per unit area is given in Equation (8), where the first term is the friction torque which acts against the gas torque to arrest turbine overspeed:

(1) $${\rm \rho }.c.{{\partial T} \over {\partial t}}{\plus}{{\partial ({\rm flux})} \over {\partial x}}{\equals}0$$

(2) $${\rm flux}_{x} \;{\equals}\;{\minus}k{{\partial T} \over {\partial x}}{\plus}{\rm \rho }.c.T.V$$

(3) $${{\partial T} \over {\partial t}}\;{\minus}\;{k \over {{\rm \rho }.c}}.{{\partial ^{2} T} \over {\partial x^{2} }}\;{\plus}\;V.{{\partial T} \over {\partial x}}\;{\equals}\;0$$

(4) $$T\;{\equals}\;T_{s} \;{\plus}\;\left( {T_{m} \;{\minus}\;T_{s} } \right).{\rm e}^{{{{{\minus}V.x.\rho .c} \over k}}} $$

(5) $${\minus}k{{\partial T} \over {\partial x}}\;{\equals}\;q_{{\rm \rmu }} \;{\minus}\;{\rm LHM}.{\rm V}\,@\,x\;{\equals}\;0{\rm }\left( {{\rm friction}\,{\rm interface}} \right)$$

(6) $$T\;{\equals}\;T_{m} \,@\,\,x\;{\equals}\;0{\rm }\left( {{\rm friction}\,{\rm interface}} \right)$$

(7) $$T\;{\equals}\;T_{{{\rm solid}}} \,@\,x\to\infty$$

(8) $$q_{{\rm \rmu }} \;{\equals}\;\left( {F.R.{\rm \rmu }} \right).{\rm \romega }.{1 \over A}$$

With the continuous centrifugation of the molten material from the interface, the flux into the solid is zero. The analytical expression is obtained by inserting Equation (4) into Equation (5) to calculate the movement of the frictional interface which is given in Equation (9). The term β in the velocity equation represents the heat partition between the surfaces that are in contact. Barber⁽ Reference Barber ^{21 )} provided an expression for the heat partition based on the thermal properties of contacting structures. The latent heat per unit volume (J/m³) is given with the LHM parameter. The frictional torque and the gas torque are then used to calculate the rotational speed by explicit time integration with the information of rotor inertia and the turbine gas torque (Equation (10)):

(9) $$V\;{\equals}\;{{\left( {1{\minus}{\rm \beta }} \right)} \over {\left( {{\rm \rho }.C_{p} .\left( {T_{{{\rm melt}}} \;{\minus}\;T_{{{\rm solid}}} } \right)\;{\plus}\;{\rm LHM}} \right)}}.{{F.R.{\rm \rmu }.{\rm \romega }} \over A}$$

(10) $${\rm \romega }^{{t\;{\plus}\;1}} \;{\equals}\;{{\left( {{\rm Torque}_{{{\rm gas}}}^{t} \;{\minus}\;{\rm Torque}_{{{\rm friction}}}^{t} } \right).{\rm Inertia}} \over {{\rm \Delta }t}}\;{\plus}\;{\rm \romega }^{t} $$

Assuming the rotor wheel rigid and considering 100% erosion of the static part, contact area and contact radius values can be obtained at each axial position by translating the rotor body in a CAD tool. The analytical friction formulation starts with establishing contact radii and contact areas for each axial displacement. These values are then interpolated during the analysis. The axial force and the torque acting on the turbine are calculated through a ‘mean-line’ analysis. During the implementation of the velocity equation, the disk is assumed to move at a speed that matches that of the contact regions where there is disk rubbing. In the case of multiple contacts, the net force across the turbine wheel is distributed proportionally to the areas, and the minimum velocity coming from the simultaneous contacts is considered as the disk velocity. By calculating the frictional force and the torque, the new rotational speed for the next time step can be calculated from momentum and inertia. The axial load and the gas torque are taken as boundary conditions from the analytical formulation. They are applied with a time curve in the FEA simulations.

3.3 Turbine finite element model

A ‘thermal–structural’ solution is applied in LS-DYNA for the simulation of the contact between the rotor and the downstream components. During the event, the temperature of the contact interfaces increases. The failure strain limit is reached due to the nature of the impact and also due to the reduced material strength at high temperature. The material model used in the FEA computations is the ‘Johnson–Cook,’ and Inconel 718 material properties are defined for all parts. The four main contact regions between the turbine wheel and the downstream stationary components are illustrated in Fig. 6(a). In Region 1, blade to NGV flow path contact and blade shank platform to NGV flow path contact occur. Region 2 shows the contact between rotor retention plate and the stationary knife seal component. In this region, the rotor has an axial seal arm extension. The knife edge seal and the seal arm provide sealing of the gas path from the lower space. Region 3 shows the contact between the seal elements and downstream support structures. The seal arm in this area has two extensions which point vertical and horizontal directions. Region 4 includes contacts at the shaft assembly. Due to the radius, contacts in this region have the lowest frictional torque capacity in terms of overspeed reduction.

Figure 6

(a) Four main contact regions in the turbine assembly and (b) front view of the finite element model.

The finite element model of the turbine assembly has 288,000 solid elements. The rotating turbine wheel is composed of blades, disk, upstream drive arm, retention plates, seals and downstream shaft. All these parts are modelled as rigid bodies. The downstream components are NGV inner annulus platform, knife seal and seals, all modelled as deformable bodies. The rigid assumption of the HPT wheel is consistent with the full downstream component erosion assumption of the CAD analysis and friction model. This assumption brings an advantage in friction model – FEA consistency. Additionally, the assumption reduces FEA computational cost since there is no ‘stress–strain’ calculation in rigid elements. Considering a similar simulation duration, finite element model aspect ratios are kept close to 1 for the turbine simulations using friction welding analogy.

For the ‘thermal–structural’ analysis, a ‘time-dependent’ axial force and gas torque are applied to the turbine wheel, a uniform initial temperature is applied to all parts and an initial angular speed to the rotating parts. The turbine assembly FEA model is illustrated in Fig. 6(b). An ‘eroding surface to surface’ contact is defined between the rotor and the downstream components.

3.4 Turbine overspeed simulation results

Figure 7 compares the LS-DYNA computations against the results obtained from the integrated analytical tool. Results are plotted in terms of axial dislocation and rotor acceleration. The axial dislocation of the turbine rotor alone is not enough to prevent overspeed. The effect of this is indirect. It is an indication of how fast the turbine wheel reaches a certain axial position. If the contact path during the axial movement has an increasing radius, a fast turbine dislocation will generate a larger frictional torque to prevent excessive acceleration. Additionally, if the turbine moves faster, blade interference will take place earlier. Blade interference is important since it promotes friction, and the destruction of the blades will reduce the efficiency and torque generated by the turbine.

Figure 7

Turbine post shaft failure simulation. (a) Rotor displacement and (b) rotor acceleration.

As mentioned before, in this study, mechanical damage effects are ignored, and the turbine performance is considered unaffected throughout. The effect of friction on overspeed can be seen in Fig. 7(b) where the actual acceleration is compared against frictionless acceleration. A very good qualitative and quantitative agreement can be observed between analytical and numerical results, with the lower-order model producing very similar results to the higher-fidelity, 3D FEA tool.

Results are plotted using a normalised time scale of 1. This is the time when the axial mechanical friction methods should take place effectively as the axial load drops below 15% of the initial load. Within the time period until the turbine end load drops to 50% of the initial, modern engine protection methods like fuel shut off, which are based on sensor detection, have enough time to operate. However, since the analytical model does not consider the mentioned control system effects, it simply tracks the overspeed curve. At two time units, the acceleration curve smoothly reaches a plateau of around 1.55 angular speed units.

3.5 Design space exploration of a stationary seal element

After the comparison of the analytical and finite element models, a design space exploration was conducted. The coupled analytical performance and friction tool allows for a faster and easier study of the various design attributes, compared to the full 3D FEA simulation.

According to the 2nd region as illustrated in Fig. 6(a), the contact radius and area of the knife seal, as it rubs against the turbine wheel, are defined as variables. These variables are changed parametrically, and the obtained geometries are compared against each other in terms of turbine terminal speed, turbine dislocation rate and the stationary component mass. Figure 8 illustrates the entire design space considered. The horizontal axis shows the mass change of the redesigned knife seal component with respect to the baseline geometry. The vertical axis shows the change in contact mean radius between the rotor and the knife seal. The plotted contact mean radius increase corresponds to the value at the edge of the knife seal for each individual design. So it is the value at the initial contact point. However, during the entire analysis, the contact radius changes with the dislocation. Every line plotted represents an isoline of change in terminal speed with respect to baseline.

Figure 8

Design space of rotor post shaft failure behaviour by speed isolines with sample geometries.

Apart from the parameters which are given as the radius at the initial contact point, mass and the rotor speed, the arrows are given for qualitative information for the contact area, thickness and the dislocation rate. With increasing contact area and radius, the mass of the knife seal geometry increases. Additionally, the material removal rate, which is also the dislocation rate, decreases. Due to reduced velocity, the contact radius does not increase enough, so the terminal speed with respect to the baseline increases in this case. For the same contact mean radius, if the contact area decreases, the rotor dislocates faster and the contact interface sees higher radii values. This condition also implies that the stationery component becomes thinner. For this reason, the terminal speed is reduced moving towards the left-hand side (on the x-axis).

The reduction in terminal speed is a desired outcome. However, the decreased thickness of the stationary component may raise integrity problems during nominal engine operation. So the reduced area and the dislocation rate act in the opposite sense, and hence, the component needs to be optimised considering the different ‘trade-offs’. Illustrative thick and thin configurations and their respective positions within the design space can also been seen in Fig. 8.

An interesting observation, based on the results obtained, is the location of the baseline configuration on the ‘performance chart’ shown in Fig. 8. The speed lines start to become more vertical after a certain value of knife seal mass, moving from right to left. The baseline configuration is located at a position where it is not possible to have a further reduction in mass while moving on the same terminal speed isoline. This implies that if an integrity issue arises related to the thickness of the knife seal, any further thickness increase for the same terminal speed, will require a drastic change (increase) in terms of knife seal radius. Also, the radius of the knife seal has a counterpart on the disk side which is the retention seal arm, as sketched in Region 2 of Fig. 6(a). The increase in radius of the knife edge seal will require significant changes to the disk design. However, when the first line from the right in Fig. 8 is considered, a thickness increase is quite possible without a change in terminal speed. So according to the parametric study, the location of the baseline design on the performance chart can be considered to be at the limit of any further potential improvement from a terminal speed reduction point of view.

About the use of the same material in the contact analyses, one should note that the use of different materials can change the frictional behaviour and the wear characteristics. A lower strength material can fail sooner than a higher strength material or difference in thermal properties may lead to a temperature distribution to soften one material more than another. The honeycomb seals or other stationary components can be considered to be weak when compared to rotating parts, however, Soupizon⁽ Reference Soupizon ^{22 )} gives an example of struts to be capable of cutting through the blades of a dislocating rotor to secure the disk from bursting. When different design features of the assembly are considered, assuming same materials provides a reasonable starting point for design purposes.

An experiment to characterise this extremely remote event is expensive, and an engine test would result in the complete destruction of the whole engine. On the other hand, modelling the dynamic response of the engine requires handling of the control systems, turbomachinery aerodynamics, combustor aerodynamics, secondary air systems and axial friction. With the verification of the individual sub-systems at intact engine conditions, a model that incorporates each of those sub-systems can be applied for the fully characterisation of the post failure event. Under these conditions, the experimental comparison becomes relevant; however, the integrated approach may still need an experimental correction since a real unlocated event also includes distinct characteristics from the intact conditions like

• ∙ radial rubbing between parts due to severe vibration,

• ∙ ingress, purge or gas accumulation in the secondary air system and in the flow path which can potentially change the direction of the turbine end load, and

• ∙ deterioration of the turbine performance due to dislocation and the loss of aerodynamic shape by fracture.

For the sub-system verification, the friction modelling approach in this manuscript is supported by finite element modelling with a friction welding analogy which is similar to the structural interactions after a shaft break. As a future application from a modelling point of view, the friction can be calculated in an integrated approach which takes into account the whole engine dynamics. But primarily, the method can be used as a stand-alone design approach to exploit the turbine end load, which is the highest available source of load after a shaft break, in order to get a desired fail-safe outcome in the engine. This may include

• ∙ A redesign of parts to maximise friction up to the time when the control system takes action.

• ∙ A redesign of parts to prevent reduction of the turbine end load by potential gas accumulation at the rear of the turbine.

• ∙ A redesign of parts to maximise turbine dislocation for maximum turbine performance deterioration.