Knowledge Driven Preprocessing for Weld Simulations

(1)

M A S T E R’S T H E S I S

2006:144 CIV

PETER THOR

Knowledge Driven Preprocessing for Weld Simulations

MASTER OF SCIENCE PROGRAMME Mechanical Engineering

Luleå University of Technology

Department of Applied Physics and Mechanical Engineering Division of Computer Aided Design

2006:144 CIV • ISSN: 1402 - 1617 • ISRN: LTU - EX - - 06/144 - - SE

(2)

Knowledge driven preprocessing for weld simulations - Peter Thor

1

Preface

This report is my final thesis for the degree of Master of Science in Mechanical Engineering at Lule˚a University of Technology in Lule˚a, Sweden. The thesis work has been conducted at the department of Design Methods & Systems with close corporation with Advanced Material & Manufacturing Technology, both stationed at Volvo Aero Corporation in Trollh¨attan, Sweden. The duration was between August 2005 and February 2006.

I would like to acknowledge the follow individuals, in alphabetical order, for their support throughout my time in Trollh¨attan:

Henrik Ahlberg, for general interest and ideas Petter Andersson, for support with software issues Patrik Boart, for supervision and help during the thesis Henrik Gustavsson, for supervision and ideas during the thesis Ola Isaksson, for support and guidance

Malin Ludvigson, for general help and friendliness

Peter ˚Astr¨om, my supervisor at Lule˚a University of Technology All other final year thesis students, for laughter outside the work

I would especially like to thank Sara Tors˚a for your loving support - without you I would not be where I am today.

Trollh¨attan, 2006-04-03

Peter Thor

(3)

Abstract

Being a supplier of products and services in numerous military and civilian fields, Volvo Aero Corporation develops and manufactures advanced components for aircraft engines.

By being aware of and reusing knowledge in the form of expertise and engineering know-how throughout the whole organization large time-savings can be made. The knowledge allows for alternative studies where different configurations can be compared and balanced. A method called Knowledge Based Engineering (KBE) can act as a valuable tool to help capturing this knowledge and guaranteeing quality of a product.

The work described in this thesis focuses on developing methods to support the generation of a weld simulation model used for a jet engine component.

The geometry of the model is treated in a general manner that allows for substitution in the future. Subcomponents within the model can be individually treated or combined with other subcomponents to create a complete system.

An automated method for mesh generation is developed and implemented.

The methods not only result in shortening lead-times but also assures consistency and quality every time they are asked for. They can be reused and further expanded to support other types of configurations in the future.

Keywords: Knowledge Based Engineering, Knowledge Fusion, Finite Element Method, Weld Simulation

(4)

List of Figures

1.1 Volvo Aero Corporation business areas . . . . 5

1.2 Weld simulation of a Turbine Rear Structure . . . . 6

1.3 Reducing lead-time by capturing and reusing knowledge more effectively . . . . 8

2.1 A typical jet engine . . . . 10

2.2 Welding of a Turbine Rear Structure . . . . 10

3.1 Schematic view of creating simulation models . . . . 15

4.1 Geometry with Heat Affected Zones . . . . 18

4.2 Connection between cone, inner ring and strut . . . . 19

4.3 The full mesh of PW2000 . . . . 23

4.4 Meshed cone of the PW2000 Turbine Rear Structure . . . . 23

3

(7)

List of Tables

4.1 Description of the major subsections of a Turbine Rear Structure 18 5.1 Implementation-time on the PW2000 Turbine Rear Structure . . 25 5.2 Implementation-time on the Turbine Rear Structure of the second

development project . . . . 26 5.3 Time to execute the methods, both for the Knowledge Based

Engineering system and manually . . . . 26

4

(8)

Chapter 1 Introduction

1.1 Volvo Aero Corporation

Volvo Aero Corporation develops and manufactures high-technology components for both military and civilian aircraft engines. The company is also active in the rocket- and gas turbine market together with several other engine manufacturers. Furthermore, the company offers spare parts, sales and leasing of aircraft and aircraft engines. Overhauling and repair of aircraft engines is another business area.

The head office lies in Trollh¨attan, Sweden. Globally, Volvo Aero Corpo- ration employs approximately 3.300 individuals (Dec 2004) with about 2.270 situated in Trollh¨attan. The sales during the same year reached 6900 MSEK.

1.2 Weld Simulation

To assist in designing a component that meets requirements under extreme conditions, Volvo Aero Corporation uses extensive simulation methods throughout its whole organization.

During the last few years, intensive research has been performed in the field of weld simulation. The desire to fabricate components by welding of smaller structural elements rather than using the traditional method of casting is

Figure 1.1: Volvo Aero Corporation business areas

5

(9)

1.3 Knowledge Based Engineering 6

Figure 1.2: Weld simulation of a Turbine Rear Structure

justified by for example improved material properties of forged and rolled parts.

The complex geometry of the castings, together with advanced materials is costly and call for alternative manufacturing technology. One of the drawbacks with fabrication is that more manufacturing operations are needed to assemble the same component. Another drawback is an imminent risk of deformations and residual stresses caused by the introduced heat. By using simulation techniques, predictions about deformations and residual stresses in fabricated components has reached a breakthrough in deployment.

The method used for weld simulations is the Finite Element Method, see Lundb¨ack [10] for theory around the area. Simulation of the welding procedure, followed by heat treatment is validated against physical experiments. The simulation raises knowledge about what optimal welding parameters to use in order to minimise deformations and residual stresses.

The Pratt & Whitney 2000 (PW2000) Turbine Rear Structure can be seen in Figure 1.2. In the same figure, a weld simulation sequence can be seen and was performed at the department of Advanced Material & Manufacturing Technology at Volvo Aero Corporation.

1.3 Knowledge Based Engineering

For a company like Volvo Aero Corporation, who develops increasingly advanced derivatives of its products, lots of knowledge and engineering know-how is gath- ered by the employees. Isaksson [8] states that information about forthcoming products exists from earlier stages of design:

Rather than inventing new products every time, new products rely on knowledge gained from previous products, strategic technology development and market.

(10)

1.3 Knowledge Based Engineering 7

By capturing this knowledge, large lead-time reductions can be made and the methods of Knowledge Based Engineering (KBE) can be utilised. If representing and managing this knowledge in digital form, the quality during all stages can be increased. However, maintenance of the knowledge is important and it is crucial that the knowledge and information is easily shared and accessed by the staff. An investigation on how to visualize CAD information in the design process at Volvo Aero Corporation, has been performed by Ludvigson [9]. It involves questions such as how 2D and 3D CAD data can be shared more easily by the staff, thus leading to fewer misunderstandings.

During a products design cycle stages, it is important to generate and investigate as many different ideas and configurations as possible. Very often the time spent here falls short compared to subsequent stages like definition, production and aftermarket - the reason often being lack of time. To maximize the efficiency, Knowledge Based Engineering can shorten time consuming and repeatable tasks.

1.3.1 Support for automation

One of the major roles of Knowledge Based Engineering is in the form of automating tasks that are highly repetitive and hold routine patterns [11]. The contents of the tasks can normally be a set of rules depicting the scope of the task. These rules can be design specifications such as component dimensions, external forces acting on the component or material properties - variables that describe the objective of the simulation model.

Other types of rules can be conditional statements, the “what - if” scenario is available for the end-user. Whole set of equations can be implemented and drive the application. Imagine a fluid flowing inside a pipe of some shape. By embedding rules about the flow (density/velocity/temperature and so on), data for the pipe (inlet area/surface finish and more) equations related to if the flow is - or becomes - laminar or turbulent can be embedded in the system. If the goal is to avoid turbulent flow, the transition point when going from laminar to turbulent flow can be monitored. Not only can geometrical changes to the pipe solve the problem but it is possible that a change of the pipe’s surface finish can do the job. This combines engineering rules with output that drives the Design For Manufacturing (DFM) process.

Systematic and repetitive tasks can be automated, iterations are possible and errors can be minimised, thus speeding up the design cycle. Individual flavours and abbreviations normally embedded in a simulation model can be sorted out, hereby ensuring consistency and quality every time the simulation model is asked for. Design flaws, normally discovered further down in the product development ladder can be dealt with - maybe even eliminated.

1.3.2 Support for lead-time reduction

Companies are constantly forced to reduce lead times while at the same time not only maintaining product quality, but very often even increasing it. This affects all stages of the product development ladder, starting at the conceptual stage and leading on to delivery of the finished product.

The prime source for knowledge within a company is the employed staff.

Being too dependent on single individuals can be hazardous in the case of staff

(11)

1.4 Thesis assignment 8

Figure 1.3: Reducing lead-time by capturing and reusing knowledge more effectively

turnover and further strengthens the ideas of capturing knowledge in digital form.

Thus, particular strengths for a successfully implemented Knowledge Based Engineering strategy can lie in several fields, all with a common factor of increasing the quality of the required task. The design process itself becomes more robust due to the embedded knowledge. Additionally, support for faster design cycles and the possibility of optimising a design for a specific purpose is in the hands of the design team.

A large amount of time when creating a simulation model is spent prior to solving it (Section 3.4.3, page 16). By implementing systems that can shorten the time it takes to execute a give task, lots of resources can be released and used more effectively. An illustration of this can be seen in Figure 1.3 where the time it takes to solve and analyse the model stays the same, while the geometrical idealization and preprocessing stages are shortened. The benefit is that more knowledge can be found much faster.

1.4 Thesis assignment

This thesis focuses on supporting the task of creating a finite element simulation model to be used for weld simulations in virtual manufacture of an aircraft engine component. The component is a Turbine Rear Structure (TRS) with the purpose of withstanding mechanical and thermal loads during flight operation.

Strict manufacturing tolerances call for advanced manufacturing techniques supported by weld simulation methods.

Specific requirements and rules on the simulation model is to be captured and embedded into the 3D computer aided design software Unigraphics by UGS.

The simulation model should support alternative positioning and configuration of the weld paths. Topological changes such as applying the methods on different components with altering geometry should also be considered.

The full thesis assignment can be seen in Appendix A.1.

(12)

Chapter 2 The jet engine

2.1 Introduction

Most commercial aircraft today are powered by so-called turbofan jet engines.

These engines fall in the category of gas turbine engines.

Gas turbine engines

Gas turbine engines power not only aircraft but also helicopters, smaller power plants and even battle tanks. What they all have in common is that they are expensive. Extreme operational conditions with high spin rates at elevated tem- peratures stretch the limits of material sciences and fluid dynamics. Basically, the gas turbine engine is built around three parts:

• Compressor

• Combustor

• Turbine

The compressor compresses incoming air to high pressure - in some implemen- tations increasing the pressure by a factor of 30. The combustor burns fuel, generally kerosene, jet fuel, propane or natural gas and produces high-pressure, high-velocity gas. Finally, the turbine extracts energy from the high-pressure, high-velocity gas flowing from the combustor. The compressor and turbine are connected via a shaft, making them turn as one unit with the turbine driving the compressor.

The turbofan jet engine

The turbofan jet engine, Figure 2.1 combines the gas turbine engine with a large fan in the front of the engine. The fan increases the amount of air transporting through the engine, thus improving the thrust. The energy extracted from the last turbine stage drives the fan, as compared to the case of the gas turbine above. Since both the compressor and fan are driven by the turbine, the turbine often participates in two (or three) stages, a high pressure stage (driving the compressor) and a low pressure stage (driving the fan).

9

(13)

2.2 The Turbine Rear Structure 10

Figure 2.1: A typical jet engine

Figure 2.2: Welding of a Turbine Rear Structure

2.2 The Turbine Rear Structure

One component developed by Volvo Aero Corporation used by jet engines is the Turbine Rear Structure. It is located right behind the last turbine stage, see Figure 2.1 and make up the rear instalment to the wing. Its main function is to transfer load from the rear, low pressure shaft bearing up to the engine mounts.

It also feeds oil to the bearings and redirects the swirling gas flow from the low pressure turbine to an axial flow.

Manufacturing

Manufacturers such as General Electric, Rolls-Royce, Pratt & Whitney and Volvo Aero Corporation manufacture Turbine Rear Structures by using a combination of casted and fabricated components. The diameter of the castings can be in the magnitude of up to two meters and the materials used often require vacuum casting. This complexity is costly and result in a limited list of suppliers. Therefore extensive research has been carried out in the field of weld simulation where components are joined by fabrication. A typical welding operation on the cone of a Turbine Rear Structure can be seen in Figure 2.2.

(14)

Chapter 3 Theory

3.1 Thesis work approach

There are several systematic methods available when conducting research or writing a final thesis. The method used in this thesis fall in the category of so called participatory action research [16]. To summarize the method it is quite close to a way of “learning by doing”. By taking part in the process the actions involved are researched by the participants. The actions are often changed during the process why they are said to be re-researched in the ongoing development.

This thesis was performed at the department of Design methods & Systems at Volvo Aero Corporation. Since the goal was to support methods at the Advanced materials & manufacturing technology department, it was natural to have close cooperation and participation in their ongoing product development.

During the whole thesis new ideas appeared, leading to alternative, yet similar, areas of application when the methods had been developed.

A literature study was made, involving previous theses performed at Volvo Aero Corporation, research about what Knowledge Based Engineering (KBE) (Section 3.2) involves and study of source code written in Knowledge Fusion (KF) (Section 3.3.2, page 13) from earlier projects done at Volvo Aero Corpo- ration. Focus was on the used methods and not on rewriting the existing source code itself. However, parts of the source code could be used, confirming with reusing previous knowledge in new projects.

3.2 Knowledge Based Engineering

Knowledge Based Engineering is a methodology that provides the designer with a tool to capture knowledge and engineering know-how in a computer aided system. A definition of Knowledge Based Engineering by Stokes [13]:

The use of advanced software techniques to reduce lead time to capture and re-use product and process knowledge in an integrated way.

Knowledge Based Engineering can help eliminating design flaws, normally discovered further down the product development ladder and caused by bad decisions in early design. By automating lower level repeated tasks, throughout

11

(15)

3.3 Capturing the knowledge 12

the whole organization, more information becomes available at a higher level.

The repeated tasks can involve redesign of a product with the aim of finding its optimal configuration.

An advantage from using Knowledge Based Engineering is that the process of creating the product is captured and defined - thus increasing the quality and robustness of the design.

3.3 Capturing the knowledge

To be able to capture and embed knowledge about a design process, a fitting tool that serves the purpose must exits - in this case a programming language called Knowledge Fusion. One methodology for the capture of knowledge is by using so-called Object-Oriented Design (OOD).

3.3.1 Object-Oriented Design

Object-Oriented Design, as defined by Sun Microsystems, Incorporated: [6]

A software design method that models the characteristics of abstract or real objects using classes and objects.

In other words, by using design methods that are object-oriented it is possible to embed knowledge in packets or modules. These modules can be grouped together in the software in a flexible manner, supporting the need for faster design cycles. To quote Booch: [3]

An object has state, behavior, and identity; the structure and behavior of similar objects are defined in their common class; the terms instance and objects are interchangeable.

Objects

An object can be anything like a house, a car or a lamp; in object-oriented terminology, a particular object is called an instance that can have state and behaviour. In this case, the state of the lamp could be colour or material and its behaviour could be turned on or turned off. For a Turbine Rear Structure, possible states can be manufacturer and material while the behaviour could be being milled, being assembled or being shipped.

The software objects state is maintained in one or more instance variables.

A variable contains data such as dimensions or mass and is named by an identifier. The behaviour of the software object is implemented by instance methods.

These are functions (subroutines) associated with the object and perform certain tasks, for example dividing a surface into two new surfaces.

Classes

Instance variables and methods can be combined together to form a class. The class can be seen as a drawing that describes instances common to all objects of a certain kind. When a class has been created, it is possible to create any number of objects from this class.

(16)

3.3 Capturing the knowledge 13

Inheritance

When classes are grouped together a relationship between them is created, commonly known as sub- and super classes - alternatively parents and children. A derived child-class inherits the variables and methods of the parent-class. For example, a subclass of the lamp could feature multiple light bulbs or shifting colour. However, it would still be part of the superclass lamp and inherits basic information that is is extended with individual data.

3.3.2 Knowledge Fusion

Knowledge Fusion (KF) is one of several programming languages integrated with Unigraphics, it is an object-oriented language that enables the end-user to apply engineering rules and knowledge within a Knowledge Based Engineering environment. Knowledge Fusion is a declarative language, rather being proce- dural, which means that the rules (objects, classes and methods) can be written in the source code without regard for order. In Knowledge Based Engineering, only the data required to perform the specific task is evaluated, others are left alone - in Knowledge Fusion this is called demand-driven data.

A typical Knowledge Fusion object can be seen below:

# Intersects a body with projection curves (Child) Intersections: {

Class; ug_curve_intersection;

Set1_References; {Body_1:};

Set2_References; Lines_to_Project:;

};

Here, the identifier of the object is named Intersections:. In turn, this object calls the class ug curve intersection:. This is a routine that takes as argument a body (Body 1:) and a list with curves (Lines to Project:) - which in this case holds three curves. Here, one object will be created thus leading to projecting intersection-curves on the body.

By combining the above code with the code below they form a class:

# Collect the intersection curves in a list (List) Intersection_Curves: {

ref(nth(1,Intersections:), "Output_Curve_0:"), ref(nth(2,Intersections:), "Output_Curve_1:"), ref(nth(3,Intersections:), "Output_Curve_2:") };

# Subdivide the face of the body with the projected curves (Child) Subdivide_the_Body: {

Class; ug_subdivide_face;

Face_To_Divide; ug_body_askFace(Body_1:);

Curves_To_Divide_By; Intersection_Curves:;

DemandOrder; {Intersections:};

};

(17)

3.4 Model awareness 14

The resulting projected intersection-curves are collected in an instance variable-list named Intersection Curves:. The single face of the body is then subdi- vided with the curves, resulting in forming four smaller faces. DemandOrder;

means that the intersection-class has to be executed before the subdivision-class is executed. This needs to be specifically stated since the rules in the Knowledge Fusion code are executed regardless of pending geometrical changes.

Reflecting over the code shows that it can be improved to allow for a more flexible setup. The hard-coded collection of the Intersection Curves:-list can be made dynamic by using a loop that collects a number of curves determined by an instance variable. This makes it possible to vary the number of curves that should intersect and subdivide the face of the body.

Throughout the whole thesis this has been one of the areas where focus has been concentrated; thinking in a dynamic and modular manner and being aware of alternative input to the simulation model.

The general rule is that small and flexible modules are favourable compared to large and bloated chunks that are hard to maintain.

3.4 Model awareness

Understanding downstream activities is becoming more and more important during a products development process. Activities such as for example Fi- nite Element Analysis or Rapid Prototyping are becoming increasingly close- connected to the geometrical representation of the model. This necessitate an increasing need for communication between individuals involved in and around each downstream activity.

3.4.1 Model quality

In CAD systems, tolerances play a major role when creating and accessing CAD model data. Variable tolerances produce gaps and overlapping features within the models, however modelling kernels used by CAD systems share methods for evaluating validity of the model. Unigraphics for example uses the Parasolid [15] geometric modelling kernel which is commonly used by many CAD systems.

Automation in the product development phase requires a consistent and systematic procedure from start to finish. Even small changes to the geometrical model can have big impacts on previously made assumptions about these activities.

3.4.2 Model usage

It is crucial not only to have knowledge in how to define and create models, but also to be aware of how the models will be used in downstream activities:

Since multiple analysis types may be required for any design state there remains a need for defeaturing to various levels to support the range of analysis to be performed. [2]

By knowing what subsequent activities to perform, how they are carried out and what constraints they put on the model, cost reductions can be made.

(18)

3.4 Model awareness 15

Figure 3.1: Schematic view of creating simulation models

This requires planning ahead and communicating with the people involved in different activities in earlier stages.

A model used for one type of simulation, say a mechanical stiffness evaluation, hold a different mesh compared to a model used for weld simulations.

Earlier, the simulation model used for the mechanical stiffness evaluation has typically also been used for the weld simulation. This is not the best scenario, since models often contain flavours especially implemented to suit a specific simulation type - the mesh becomes restricted to a model that does not suit weld simulation.

A preferable way of working is to start with a clean geometry prior to implementing any geometrical adaptions to the geometry. This is illustrated to the right in Figure 3.1. To the left a fully parametric model of an Intermediate Case (Section 2.1, page 10) can be seen, developed by Andersson [1], Bylund [4] and Rajagopal [12] - this side represents a scenario where mechanical stiffness of the component is investigated. On the right side, three steps marked in orange, Ge- ometry Definition, Geometry Idealization and Mesh Generation involves using methods described in Section 4.3 - 4.6 (page 19 - 21) when creating the analysis model used for weld simulations. The second step, Geometry Idealization is clearly separated from the same step on the left side in the Figure to illustrates that they have different characteristics and should be treated individually.

(19)

3.5 Welding 16

3.4.3 Faulty geometry that hinders automation

The International TechneGroup Incorporated [7] states that up to 70% of the man-hours spent during Finite Element Analysis, involves correcting geometrical problems. Decreasing this time factor should be one of the major strives for any company performing simulation activities. This is further strengthened by Desaleux and Fouet [14] which states that costs for the creation of a FE mesh represents about 80% of the total analyst cost. Correcting CAD model geometry is considered to be the most timely activity [5].

3.5 Welding

By fabricating components using welding, rather than casting large structures, advantages lie in decreased monopoly sensitivity and better material properties of forged and rolled parts.

Welding is a method that bonds pieces together by heating the interface between the two. The welding process induces deformations and residual stresses that have to be minimized. The deformations originating from the welding process can affect the production where tolerances play an important role. Residual stresses inside a component can decrease its overall life, especially if the magni- tudes of the stress levels increase the risk of fatigue.

As computational power increases, it becomes possible to simulate welding and intermediate heat treatments within a reasonable time frame. Predictions regarding how a structure will be affected by welding and heat treatment can be done and the need for experiments can be reduced. Instead of only relying on experimentally obtained knowledge, validations between the simulated and the real world shifts the way of working.

The most commonly used computational method for welding simulations is the Finite Element Method, see for example Lundb¨ack [10] for a more thorough introduction in the field.

(20)

Chapter 4 Method

Note to the reader of this document

This chapter is a brief/introductory description of the methodology. More extensive details of the methodology is found in the Appendix of the internal report.

4.1 Object-Oriented methodology

Throughout the whole thesis work, the methodology of Object-Oriented Design has been used (Section 3.3.1, page 12). Components of the geometry described below have been treated as objects and rules have been designed to be as modular and flexible as possible. The relationship between the rules and geometry are hereby less dependent on one specific model and can be applied on other components.

One advantage with Object-Oriented Design is that it allows for reuse and extension of previous solutions. A modular approach can be carried out and in the case of the simulation model of the Turbine Rear Structure, subcomponents can be preprocessed independently of others. As an example, individual meshing of instance objects, for example the cone and inner ring can be executed by only changing a toggle in the rules (source code) of the simulation model.

4.2 The geometry of the models

A total of three Turbine Rear Structure models have been used in this thesis work. The first Turbine Rear Structure was used when developing the preprocessing routines described in this thesis. These were later successfully applied onto the two other projects; one existing in production today and one currently under development.

The following sections addresses the methods and geometry used when preparing the model for weld simulations.

For details about what steps to perform each time a new geometry needs to be supported, see the Appendix in the internal report.

17

(21)

4.2 The geometry of the models 18

Subsector Label Main function Manufacturing method

Engine mount - Connect to wing instalment Casting

Outer ring A Define gas flow channel Welded sheets

Strut B Redirect gas flow, oil Casting

Inner ring C Define gas flow channel Casting

Cone D Seal cavity, connect to bearing Sheet forming Table 4.1: Description of the major subsections of a Turbine Rear Structure

Figure 4.1: Geometry with Heat Affected Zones

The model and its subsectors

A similar model compared to the one used when developing the methods described in this thesis can be seen in Figure 4.1. The geometry illustrates the Pratt & Whitney 2000 Turbine Rear Structure, commonly known as PW2000.

The geometry is one circular-symmetric sector that consists of several subsectors listen in Table 4.1, together with a description of its main functions and typical manufacturing method. In total, 15 sectors form the Turbine Rear Structure. Three of the sectors have integrated engine mounts (not shown).

The different subsectors are each represented by a so called sheet body in Unigraphics. A sheet body is a surface that can contain multiple individual faces and edges. These faces can be used for Finite Element Method-meshing together with individual nodal- or element density settings on the edges (Section 4.7.1, page 22).

Weld zone

Four weld zones can be seen in Figure 4.1, two on the strut and two on the outer ring, all indicated by the green areas. Conventionally, the weld paths inside the weld zones have been created in the model by dividing the surface of the outer ring manually, either by the projection of a datum plane or a curve, hindering automation.

Welding the strut is more complex than welding the outer ring since it con-

(22)

4.3 Geometrical convention 19

Figure 4.2: Connection between cone, inner ring and strut

sists of four parts, one casted leading edge, two sheet metals and finally one casted trailing edge. Multiple welding operations are carried out on the strut to guarantee full weld penetration. This is performed with both robotic and manual welding.

For more details about the weld zones, see the Appendix in the internal report.

Heat Affected Zone

The Heat Affected Zone (HAZ) is the area of base material around the weld path which changes its microstructure and properties when welding. Heat from the welding process and subsequent cooling causes a microstructural change in the area surrounding the weld. The changed properties depends primarily on the amount and concentration of heat input, the base material and the weld filler metal.

In the following chapters, a Heat Affected Zone is referred to as the geometry created around a weld path - also called the weld zone.

For details about methods on how the Heat Affected Zones are treated, see the Appendix in the internal report.

4.3 Geometrical convention

To support the automation of a process, systematic and consistent sequences has to be identified, captured and implemented. Commonly, a Turbine Rear Structure consists of the outer and inner rings with interconnected struts, one cone and engine mounts.

Typical problem area

Going into detail and looking at Figure 4.2, it can be seen that in both pictures the strut is connected to the inner ring and webbing, creating a relationship between the two. Generally, the more existing connections and relationships the more complex it is to maintain the structure of the source code. Therefore, the scenario in the left picture is favourable - the strut shares the same unique edges of the inner ring and webbing. On the right picture however, the edge of the front lower part of the strut is shared between the front and middle faces of the inner ring. This causes problems since two edges of the strut connect to the webbing, making it harder to link the meshes together with the current functionality of Unigraphics.

(23)

4.4 Preprocessing the geometry 20

Geometrical suggestions

It is suggested that the strut is divided into six faces; two faces representing the pressure and suction sides, two faces in the front of the strut (as can be see in the Figure) and finally two faces in the rear. The advantage is that the geometry is quite simple - six faces, each with four edges - and allows for more control when meshing.

It is also recommended that the model should be as clean as possible without having any geometrical preparations used by other simulation tasks. This was addressed in Section 3.4.2 on page 14.

4.4 Preprocessing the geometry

By introducing an object-oriented method that permits automated division of the surfaces around a weld path, geometrical and manual dependency fades out.

When new faces and edges are created, they can be tracked by using a sorting routine (Section 4.5, page 21) for the purpose of applying a mesh with strict control (Section 4.6, page 21).

4.4.1 Flexible Dividing Method

Dividing an outer ring

The outer ring of the model depicted in Figure 4.1, page 18 contain several zones where welding takes place. Sectors are welded together by intermediate sheets on the outer ring to form a complete, 360 degree structure.

The methodology called the Flexible Dividing Method (FDM), used when creating the geometry representing the area around a Heat Affected Zone on the outer ring, is described in the Appendix in the internal report.

Dividing a strut

The strut of PW2000 is connected to the outer and inner ring by welding it against two hubs on either side. Several welding steps are performed in order to certify full weld penetration.

A comparable, but slightly different, method than that used for the outer ring is used when creating the area representing Heat Affected Zones on the strut. For details, see the Appendix in the internal report.

Dividing a cone

In total, six welding operations are done on the cone of the PW2000 Turbine Rear Structure. The exact same method used when creating Heat Affected Zones on the strut is reused on the cone.

For details about creating the geometry that represents the Heat Affected Zone on the cone, see the Appendix in the internal report.

(24)

4.5 Postprocessing the geometry 21

4.5 Postprocessing the geometry

When the Flexible Dividing Method has been applied on the geometry, it becomes necessary to keep track of all new surfaces and edges that have been created. By using a sorting method that expects disarranged input and delivers sorted output the benefits lie in consistency.

4.5.1 Consistent Sorting Method

All faces and edges of a subsector subjected to meshing, rely on a set of reoc- curring rules. They are all

• using a sorting convention based on a common global coordinate system

• sorted by their distance to one of more points

Additionally each subsector hold individual set of rules that handle the relationship between other subsectors, for example the relation between the strut and its connection to the outer and inner ring. The sorting method, called the Consistent Sorting Method (CSM) is executed in the Structures module in Unigraphics.

Sorting the geometry

General routines that can sort the output from any number of introduced Heat Affected Zones from the Flexible Dividing Method is a flexible way of approach- ing a geometry.

For details about sorting the geometry of the outer ring, strut and cone after the Flexible Dividing Method has been implemented, see the Appendix in the internal report.

4.6 Apply meshing rules

Automated mesh generation is especially powerful when different configurations are to be evaluated against each other, for example during a conceptual stage.

This requires strict control of the geometry, why large topological changes should be avoided. If the topology changes between each configuration, introducing unexpected additional faces and edges, the rules of the mesh generation model can be messed up. Keeping things as simple and consistent as possible is the key to success.

The big difference between manual and automatic meshing is that the latter require more control. Intuitively it is possible to allow for a varying meshing procedure that can be changed from time to time because of the human ability to adapt. This does not fully comply for automated meshing, but leads to additional rules that have to be created - often caused by small geometrical changes that prove to have little impact on the simulation model in the end.

4.6.1 Meshing with Knowledge Fusion

When the faces and edges of the sheet body have been sorted (Section 4.5.1) the subsequent step is to apply meshing rules. This is a three step process and is executed in sequential order:

(25)

4.7 Simulation model 22

1. Set element or nodal densities on edges 2. Set element sizes on faces

3. Call the meshing class

Setting nodal densities on all edges of one sheet body loosens the relationship with all the other sheet bodies in the geometry - an independence is created that allows for a more flexible combination of the subcomponents.

When completed, the mesh can be edited manually by the Graphical User Interface (GUI). This makes it possible to balance the automated meshing procedure with manual work.

4.7 Simulation model

To minimize the time it takes to create the simulation model it is crucial to find the simplest possible geometry that describes it. Often the geometry is fully or partly symmetric - as in the case of a Turbine Rear Structure - why this should be taken advantage of.

By working in an object-oriented way, where symmetry in the geometry is combined with symmetrical boundary conditions, the model creation time can be shortened. Note that the application of boundary conditions is not considered in this thesis work, it is done in the simulation software.

4.7.1 The mesh

The mesh of the simulation models has been generated on one sector of the geometry - taking usage of symmetry. Strict control of the mesh connecting to other sectors makes sure that elements on edges bond together. The mesh around and inside the Heat Affected Zone needs to comply with the theories and methods used by the simulation software. This means using the appropriate element type and weld path descriptions.

For more details about the requirements on the mesh, see the Appendix in the internal report.

4.7.2 Exporting the model

The mesh is exported to the simulation software, MSC Marc, where it is used for the weld simulation. By also exporting the geometry, it becomes possible to alter the generated mesh from inside the simulation software - in case it needs additional refinement.

Individually changed subcomponents, and their respective meshes, can be exported to the simulation software and joined with previously generated meshes.

This makes it possible to exchange an obsolete mesh, for example the mesh of a strut, with an updated one - all without having to remesh the whole model.

The elements and nodes are exported from Unigraphics using a format in- terpreted by ANSYS. In-house and standardized software at Volvo Aero Cor- poration is used to convert the data into a format MSC Marc can handle.

(26)

Figure 4.3: The full mesh of PW2000

Figure 4.4: Meshed cone of the PW2000 Turbine Rear Structure

Complete mesh of PW2000

Creating the mesh for the PW2000 Turbine Rear Structure was a joint effort between the two departments (both mentioned in the preface). The mesh can be seen in Figure 4.3 - note the coarse elements on the cone, leading to a much finer mesh close to the inner ring. Multiple simulation steps are performed where focus in this specific case is the welding procedure of the struts. The areas around the Heat Affected Zones were created with the methods explained in this thesis, both on the strut and on the cone.

Figure 4.4 shows implemented Heat Affected Zones on the cone. It can be seen that welding is done close to the inner ring. Since the interfaces between this and the other mesh match, the coarsely meshed cone in Figure 4.3 will be substituted in favour of this finely meshed cone. Next, weld simulation of the cone is performed.

(27)

4.7.3 Boundary conditions

The implementation of the boundary conditions used in the weld simulation are applied in a modular way, in the simulation software MSC Marc. The mesh and boundary conditions are copied and rotated around the axis of the Turbine Rear Structure. This generates the full, 360 degree simulation model ready for weld simulation and heat treatment.

Additional boundary conditions, that do not satisfy symmetrical conditions, are attached in a subsequent step. The simulation is then initiated, solved and postprocessed. Throughout the whole procedure, properties such as the quality of the mesh and the number of contained elements has to be balanced against the solution time.

(28)

Chapter 5 Results

5.1 Time to support the different projects

After creation of the methods described herein, the method implementation time was recorded for both the PW2000 project as well as for the second development project. Main features and differences between the two were identified and rules were adapted to support an automated mesh generation.

A listing of activities executed when adapting the rules for PW2000 can be seen in Table 5.1. The major difference in this case was a new type of cone that required expansion of the methods to support future configurations. A lead- time of five days included three days of methodology development. A change of creating Heat Affected Zones on the strut, and implementing the possibility of having any number of them, added to the functionality. By agreeing on a convention on how to create the geometry of the strut, implementation of the methods in the second development project was simplified. A combination of manual work and the usage of the automatic methods described herein proved to be successful.

Another listing of activities, corresponding to adapting the methods for the second development project, with measurements of time, can be seen in Table 5.2. Absence of topological changes in this project made it easier to support the creation of the weld simulation model without large alterations of the code.

Action Time to implement

Geometry

Identify geometry, configure Heat Affected Zones 8 hours Adapt rules for intersecting subcomponents 2 hours Adapt facial and edge sorting routines 2 hours

Mesh implementation 6 hours

Total time 18 hours

Table 5.1: Implementation-time on the PW2000 Turbine Rear Structure

25

(29)

5.2 Generating the mesh 26

Action Time to implement

Geometry

Identify geometry and subcomponents 1 hour Adapt rules for intersecting subcomponents 1 hour Configuring Heat Affected Zones 3 hours Adapt facial and edge sorting routines 2 hours Mesh implementation

Set edge nodal densities 1 hours

Set face element sizes 30 minutes

Total time 8 hours 30 minutes

Table 5.2: Implementation-time on the Turbine Rear Structure of the second development project

Action Execution time Execution time

Geometry Using the thesis methods Manually

Create Heat Affected Zones 20 secs 31 minutes Mesh setup and generation

Mesh generation 1 minute 40 secs 23 minutes

Total time 2 minutes 54 minutes

Table 5.3: Time to execute the methods, both for the Knowledge Based Engi- neering system and manually

5.2 Generating the mesh

When the actions described in Table 5.1 and Table 5.2 have been performed, it remains to execute the mesh generation, evaluate it and certify the result.

A similar table is constructed where the Knowledge Based Engineering system execution time is recorded, see Table 5.3. Table 5.3 also illustrates the time it takes for an experienced engineer to manually apply the methods on the model. Repetitive training on how to implement the methods resulted in a minimum validated time of implementation, shown in the right column.

It can be seen from Table 5.3 that embedding repetitive tasks within the system can decrease the time it takes to execute certain activities substantially.

The computer on which the implementation was made had a 3 GHz Xeon pro- cessor, running Microsoft Windows XP Professional Version 2002 SP2.

After 20 seconds, the positioning and geometrical preparations of the Heat Affected Zones, used by the Flexible Dividing Method, is complete. Doing this manually takes over 90 times longer. Less than two minutes after the meshing- methods are executed by the system, a fully meshed component is generated.

It takes more than 14 times as long to do the same procedure manually.

(30)

Chapter 6 Discussion

6.1 The present

By participating in model creation activities early in the design process it is easier to create rules and methods that can be used to support the creation of the simulation model.

Model simplifications

Automation, conventions and consistency go hand in hand. By agreeing on conventions between different disciplines on how to define, build and simplify geometry, it becomes easier to implement methods used when supporting the preprocessing of a simulation model.

Code reusage and knowledge capture

The strengths of Knowledge Based Engineering is starting to show its potential.

Methods and routines, originally intended for creating weld zones on the strut are reused under different conditions. In turn, these methods are derivatives from the ones developed and used on the outer ring. Meshing the strut and cone involves repositioning the rules and methods used for meshing the outer ring.

Several phases of the preprocessing chain are captured in digital form; how the geometry around Heat Affected Zones should be configured and created, their positions and the mesh characteristics of the whole model. Knowledge about the steps required to support the creation of a weld smulation model are implemented in a set of rules in the source code and can be reused in the future.

Semi-automatic procedure

A preferable way of working is by combining manual work with tools that supports repeatable and time-consuming tasks. These tasks are usually error-prone, tedious and hold a pattern. When identified, the tasks can be captured and methods can be developed and used in a system that supports automation. Au- tomatically generated Heat Affected Zones and meshing procedures, combined with manual meshing proved to be successful.

27

(31)

6.2 The near future 28

Flexibility

By using an object-oriented approach, individual model-dependency is proved to be of less importance - routines are adaptable and alternative application areas are possible. During the thesis work, a scenario appeared where the strut needed to be remeshed but where its geometry stayed the same. When finished, the mesh and geometry was exported to the simulation software and joined with the previously meshed cone, outer and inner ring. This shows that it is possible to merge different types of geometry with each other - as long as the interfaces between them match.

Lead time reduction

When tools and methods exist that covers the different configurations and components, it becomes possible to make a sound estimation of how long time it will take to deliver an automatically prepared and meshed model with a new geometry.

One engineer reported working for two days with meshing a typical geometry of a Turbine Rear Structure. In turn, the geometry was prepared by another engineer and multiple sessions between the two lead to a lead-time of several days for preparing the geometry alone. It is natural that this process will be shortened next time and that misunderstandings and errors will decrease - if the same two individuals work together again. Since no one can guarantee that the same people will communicate around a similar process in the future, there is always a risk for loss of knowledge caused by staff turnover.

A system however, captures this knowledge. Methods that describe the process can be used by the designers and aid them in the process while at the same time increasing the quality of their work.

The scenario described in the result-section (Section 5.2, page 26), depicts an optimal manual scenario where 100 % focus is on inplementing the methods without external disturbance. Here, the same people are used over and over again when preparing the simulation model - this is seldom the case in the real world. Normally, staff often work in parallel projects meaning they have to divide their time between these projects.

Concurrent engineering

Method functionality can be added in parallel with other activities. During the creation of a mid-shell model for the PW2000 Turbine Rear Structure, dis- cussions showed that some of the present methods needed update. The new methods were developed in parallel with the creation of the mid-shell model.

When the geometry became available, applying the expanded methods on it posed minimum hassle.

6.2 The near future

Each time new geometries with different configurations are presented, the rules and methods in the source code need to be updated and expanded. Knowl- edge about the behaviour betwen different configurations can increase faster,

(32)

6.3 Further suggestions 29

if similar methods like the ones described in this thesis are used earlier than traditionally in the design process.

Substitution

Since it is possible to mesh single parts of the geometry and export them individually to the simulation software new opportunities arise. Parts can be left intact while others can be changed, these can later on be combined into a new configuration.

The methods described herein can be used on other types of geometry. One example could be to support the fabrication of other components, for example an Intermediate Case (Figure 2.1, page 10).

Design studies

By working in an object-oriented way and setting demands on the geometry it becomes possible to make design studies. An example of a design study would be to investigate where the most suitable location for joining parts with welding would be. Looking in Figure 4.1, page 18, the position of the four weld zones are easily changed by just editing a few lines in the source code. Now, optimum weld path positioning can be investigated - this was not possible before to the same extent, due to lack of time and effective tools.

Sensitivity-studies

Investigating how the element sizes of the mesh influcences the result can be done. The element sizes in a mesh can be multiplied with a factor, resulting in creating a more or less densely mesh. Parts of the mesh - or all of it - can be subjected to several factors. Thus, data for sensitivity-studies can be generated within a short period of time for selected areas.

6.3 Further suggestions

Areas that need further attention in the future are based around geometrical conventions, further support to the creation of the analysis model and embedding the methods developed herein in modules that are easily accessed.

Agreeing on conventions

One very important issue regarding CAD-geometry is that it follows certain rules and conventions when created. Since a vast amount of time normally is spent correcting faulty CAD-geometry (Section 3.4.3, page 16) this is a field that needs special attention - especially if automated activities are to be successful.

By agreeing on conventions and features around how the geometry is created a lot of time can be saved.

Boundary conditions

It should be investigated if it is possible to simplify the implementation of the boundary conditions used in the simulation, either in the CAD system Unigraph- ics or the simulation software MSC Marc. Today, the boundary conditions are

Knowledge Driven Preprocessing for Weld Simulations

M A S T E R’S T H E S I S

Knowledge Driven Preprocessing for Weld Simulations

Contents

List of Figures

List of Tables

Chapter 1

Introduction

Chapter 2

The jet engine

Chapter 3

Theory

Chapter 4

Method

Chapter 5

Results

Chapter 6

Discussion