Evaluation and visualisation of surface defects on auto-body panels

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

journal homepage: www.elsevier.com/locate/jmatprotec

Evaluation and visualisation of surface defects on auto-body panels A. Andersson a,b,∗ a b

¨ Volvo Cars Body Components, Olofstrom, Sweden Division of Production and Materials Engineering, Lund University, Lund, Sweden

a r t i c l e

i n f o

a b s t r a c t

Article history:

The ability to predict surface defects in outer panels is of vital importance in the automo-

Received 21 October 2007

tive industry, especially for brands in the premium car segment. Today, measures to prevent

Received in revised form

these defects cannot be taken until a test part has been manufactured, which requires a

15 February 2008

great deal of time and expense. The decision as to whether a certain surface is of acceptable

Accepted 23 February 2008

quality or not is based on subjective evaluation. It is quite possible to detect a defect by measurement, but it is not possible to correlate measured defects and the subjective evaluation. If all results could be based on the same criteria, it would be possible to compare a surface

Keywords: Surface defects

by both FE simulations, experiments and subjective evaluation with the same result. In order to find a solution concerning the prediction of surface defects, a laboratory tool

Sheet-metal forming

was manufactured and analysed both experimentally and numerically. The tool represents

Simulation

the area around a fuel filler lid and the aim was to recreate surface defects, so-called “teddy

Finite element method

bear ears”. Several different sheet materials were analysed in order to evaluate their sensitivity to surface defects, and to investigate the possibility to predict defects in different materials. A major problem with the evaluation of such defects is that the panels are evaluated manually and to a great extent subjectivity is involved in the classification and judgement of the defects. In this study the same computer software was used for the evaluation of both the experimental and the numerical results. In this software the surface defects were indicated by a change in the curvature of the panel. The results showed good agreement between numerical and experimental results. Furthermore, the evaluation software gave a good indication of the appearance of the surface defects compared to an analysis done with the existing tools for surface quality measurements, e.g. D-Sight [D-Sight, 2007. http://www.lmint.com] results. Since the agreement between numerical and experimental results was good, this indicates that these tools can be used for an early verification of surface defects in outer panels. © 2008 Published by Elsevier B.V.

1.

Introduction

The ability to predict surface defects in outer panels is of vital importance in the automotive industry, particularly for cars in the upper segment of the market. Today, measures to prevent

∗

these defects cannot be taken until a test part has been manufactured, which involves the expenditure of much time and money. If these defects could be predicted at an early stage in the development process, the time and cost reduction would be significant. A means for achievingthis is to use sheet-

¨ Correspondence address: Volvo Cars Body Components, Olofstrom, Sweden. Tel.: +46 454 265280; fax: +46 454 265788. E-mail address: [email protected]. 0924-0136/$ – see front matter © 2008 Published by Elsevier B.V. doi:10.1016/j.jmatprotec.2008.02.078

822

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

metal-forming simulation. The use of this type of simulations has become more common in the automotive industry over the past decade and are efficient tools in many applications. Makinouchi (1996) and Makinouchi et al. (1998) have described various uses of sheet-metal-forming simulations in the automotive industry. Today it is possible to predict thinning, strain distribution and forces to a high accuracy, but there are still challenges to be overcome. One challenge is the prediction of surface defects. Surface defects are small deviations from the nominal surface of a panel, and can be of varying size and depth. The defects can appear as: • • • • •

Depressions Elevations Bimps Orange peel-like Local thinning

Defects with relatively large depths (wrinkles) are visible in an optical check, while small defects are detected by a method in which a specialist manually examines the panel. These defects do not become visible until the panel has been painted. However, they must be distinguished from the type of defects which appear on the micro-scale, i.e. depth variations in the ␮m-range, e.g. the Ra-value which represents the surface roughness. Surface defects often appear on relatively flat panels with some kind of embossment, e.g. on doors in the area of the door handle, and on rear fenders with a fuel filler lid. The areas around the corners of the embossments will be subjected to compressive stresses. Since the panels usually have low stiffness in these areas, and the plastic strains are insignificant, they are very sensitive to springback which results in surface defects. Today there are methods to detect small defects on autobody panels by using interference of light (Kinell, 2003). These methods are able to visualise the defects, but are limited in efficiency, and the interpretation of the results is difficult. This work shows a method where experimental and numerical assessment of surface defects is done with a common software. The advantage with this is that the assessment of the surface will be based on the same scale, independent of if it is assessed by experimental or numerical procedures.

2.

Fig. 1 – Description of methodology.

the “3DS - digital Die Design System” project, EU-contract G1RD-CT-2000-00104 (3DS). These tests were then followed by a test on an automotive part, the outer panel of a door.

3.

Method

The methodology used in this study is described in Fig. 1.

3.1.

Definition of surface defects

The human eye is very sensitive to discontinuities in the shape of a surface, especially changes in the curvature. These can be related to changes in the gradient of the curvature. As the gradient increases, the severity of the defects increases. Since the gradient of curvature is more sensitive to disturbances in

Objective

The objective of this study is to investigate whether simulation of forming can be used in the prediction of surface defects. It includes a proposal of a method for comparing simulation results and experimental results using the same software. By using the same analysing instrument for the evaluation of numerical and experimental results, the ability to compare experimental and simulation results will increase. This also simplifies the dialog between pre-production and production, since the same criteria are used. A simple forming tool, a double-curved panel, which corresponds to the area around the fuel filler lid on a side panel of a car, was used for the analysis. This tool was developed within

Fig. 2 – Procedure for establishing the accuracy of the evaluation system.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

823

Fig. 3 – The stylus instrument and measurement result.

the surface than height variation, it is easier to evaluate the surface quality with respect to curvature variations than to height variations. Any disturbances in the curvature indicate that either the shape is changing or that a defect has appeared. The curvature is defined by: Curvature =

1 R

(1)

where R is the radius of the curvature. Other important factors, which affect the observer’s apprehension of surface defects are the following: • Wavelength: The human eye is most sensitive to defects in the wavelength ranges 30–50 mm (3DS Report, 2002). • Distribution: The direction of the defect also affects the observer’s perception of the defects. If the main distribution of the defects is perpendicular to the direction of the incoming light, it is more visible than if it is positioned along the direction of the incoming light (3DS Report, 2002). • Relation distribution/depth: Denote the area of the defects as A and the depth as D. The relation between them, S, is then defined as: S=

D A

(2)

Of two defects (at a similar location), the one with the larger value of S will have the more severe defect. Dividing the analysed surface into different zones, and denoting the severity of the location of the defects by Z (e.g. by grade 1–10 were areas on the car which enhance the perception of defects have higher grade), then a variable R, relating to both location and appearance of the defects, can be defined

as: R=S∗Z

(3)

R denotes a relation which takes both the location and the geometry of the defect into consideration. The larger R the more severe the defect. With this concept it can be seen that many factors influence the perception of a defect. At Volvo Car Corporation (VCC), manual inspection is used in the assessment of surface defects. The evaluation is done either by hand or by optical inspection of the surface, coated with a film of oil, in a light-frame. The inspector uses a scale from 1 to 10, where 1 is worst and 10 is best. For acceptance, a level of 7 must be reached. A lower value than this must be treated with special care or adjustments must be made. The use of the scale is based on experience from production parts. Both the geometry and location of the defects are regarded in the assessment. As a reference a “master part” is used, which is taken from the start of production, and which is evaluated before and after the painting process. When the required surface quality is reached, the part is used as a master part and represents the minimum quality level the production parts must reach. All inspections are based on optical assessments. In the process of producing a master part, a team of least two independent inspectors judge the panel.

3.2.

Examination of shape and size of surface defects

The appearance of surface defects, which are defined in Sections 1 and 3.1, was evaluated by tests on a complete door

824

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

of a Volvo S60, which was taken from the production line. The method of determining the typical shape and size is described in Fig. 2. The surface was first assessed by visual inspection, and the defect areas were determined. A surface replica of the defect area was then made with Technovit 3040 from Heraeus. This replica is applied in soft condition on the surface and hardens on the surface. This hardened material can be regarded as the mirror-image of the defects on the panel. The replica of the surface was then analysed with a stylus instrument (ASAME) (Thomas, 1999) and a section of the surface, A, was plotted. The section was chosen to be on the deepest part of the evaluated depression (the area marked with a white circle in Fig. 3). In order to describe the deviation from the original curvature, a section B was created as a curve fitting to curve A with a 2nd degree polynomial. Section A and B were overlaid and the difference was measured. The assumption was made that the defect could be regarded as a local defect, and since the surface replica covered a larger area than the defect, a 2nd degree polynomial was used as the description of the original curvature. The purpose was to see the magnitude of the depth of the defect and not the absolute value. Therefore, this method was sufficiently accurate for the evaluation. The set-up for the stylus instrument measurements and the results of the section measurement can be seen in Fig. 3. This study provided a reference when the demands on accuracy for the measurement system were established. Based on experience, a typical visible surface defect would have a depth of >50 ␮m. The results of the measurements (area marked with a white circle in Fig. 3) show that a defect ranked as a 1 on the Volvo assessment scale (severe defect) has a depth of ∼40 ␮m. Another visible defect (shown in Fig. 3 with a black circle) has an indicated depth of <10 ␮m. This defect was ranked as 9. The results are not exact, but give at least an indication that the accuracy must be much better than the 50 ␮m indicated by assessment based on experience. Therefore the demand on the accuracy of systems to be used to detect surface defects was set to 10 ␮m.

3.3.

Visualisation and measurement of surface defects

In order to visualise the surface defects, several different techniques can be used. Kinell (2003) divides them into three different techniques: interferometry, triangulation and time-of-flight. In Table 1 the most common methods: interferometry and triangulation, are described. The above-mentioned techniques must of course be adjusted and tuned into the correct configuration in order to achieve the desired features, e.g. accuracy over the specified area, speed, etc. These techniques are used for a description of the surface. Another important issue is to interpret the results in order to classify defects which are not acceptable. In systems used for surface evaluation of automotive parts, systems based on triangulation are mostly used, since they are accurate enough, can handle areas of required size, and are not so sensitive for disturbances.

Fig. 4 – Methodology for surface evaluation.

3.4.

Numerical and experimental evaluation

In Section 3.1, the method of assessment of surface defects in current use is described. This method is based on subjective judgement and is difficult to use when comparison to numerical results is to be performed. It is possible to compare experimental and numerical results qualitatively but not quantitatively, since different evaluation scales are used. In order to have the same evaluation scale in numerical and experimental assessment of surface defects, it is necessary to have the same evaluation methodology or software. Fig. 4 describes how such a methodology can be reached. All the individual methods are available, but the difficulties lie in the possibilities to have the same evaluation software. Today, the decisions whether a surface is approved or not are based on subjective evaluation (see Section 3.1). It is quite possible to detect a defect by measurement, but there is no possibility to correlate measured defects and the subjective evaluation. If all results could be based on the same criteria, it would be possible to compare a surface in both finite element (FE) simulations, experiments and subjective evaluation with the same result. The transformation of the point cloud into a mesh was done in TEBIS (2007). In TEBIS the number of points was minimised without violating the surface quality and connected to a mesh. This mesh could then be used in the evaluation program. It is important that an accurate prediction of the springback is reached. In order to investigate if surface defects could be detected and evaluated, a test model, a double-curved panel, was evaluated. The test model was analysed both numerically and experimentally (see Sections 7 and 8).

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

825

Fig. 5 – Basic principle of the optical technique and set-up of the WMS-system. The surface is illuminated from two opposing lamps. The sloop angle ˛ affects the light intensity to the camera. This method is referred to as a triangulation method (mix between photometric stereo and projected fringes) in Section 3.3.

For confirmation of the applicability in a real production process an automotive part, a Volvo S40 door, was also analysed. The study was limited to experimental assessment, and the part was measured in the WMS-system/NXT post processor and in the D-Sight system. The results were then compared. Based on this information and on the previous results from the double-curved panel the usefulness of the methodology and of the evaluation tools could be validated.

4. Description of measurement system, WMS In this study a system called WMS was used for surface measurement of the experimental parts. The system is described by Max et al. (2003) and an outline of the set-up of the system can be seen in Fig. 5. The accuracy in height is 0.01 mm on a surface of 1 m2 .

5.

Evaluation program

As an evaluation program the NXT post processor (3DS Report, 2002) was used. Furthermore, an early version called 3DS Evaluation Processor is described by Kase et al. (1999). This

software was then modified to the NXT post processor. These programs were developed during the IMS-project 3DS (Col, 2002). In the NXT post processor it is possible to detect changes in curvature. These changes indicate a change in geometry. If the change in geometry is not a design feature, it is regarded as a defect.. The severity of the defects is evaluated by analysing the size and gradient of the curvature change. The severity of the defect increases with increasing value of the curvature change. The input to the program is a mesh, which can be generated from both numerical and experimental measurements. In the program the wave length and the evaluation direction are chosen. Furthermore, it is possible to do a section analysis in user defined planes, where the angle change is plotted. Since the angle change is plotted, defects appear as deviation in the plot. The position can easily be detected since the coordinate plot is shown next to the angle plot. These are linked and a point moves in both diagrams as the mouse is moved over them. In Fig. 6 a picture of the graphic window is displayed. Since the evaluated values are very small, an accurate measurement system is desired. Otherwise, the defects will not be captured in the measurements. In Section 3.2, it was discussed that an accuracy of 10 ␮m was needed to capture the defects. Furthermore, this also demands accu-

Table 1 – Common optical shape measuring techniques Method

Example of systems

Interferometry

Holography

˚ This technique is described in Gasvik (1995) and Kinell (2003). The light is divided into two parts, which takes separate ways to the hologram. One part acts as reference wave and the other as object wave. At the hologram they interact with each other and create a topology map of the surface

Triangulation Passive

Photometric stereo

This principle can be seen in Hansson (1999). The intensity from the different light sources can be evaluated and create a picture of the surface topology The principle is based on that every disturbance is causing a phase shift in the intensity of the light, and by analysing this a representation of the surface topology can be obtained. By superimposing several images from varying stripe widths in the grating, a good presentation of the surface can be created, see Huntley and Saldner (1997), Saldner and Huntley (1997), and Kinell (2003) ˚ This technique is described in Gasvik (1995). It is based on the principle that the grating and its shadow form a moire´ pattern. This pattern builds a topological description of the surface

Active

Projected fringes

Shadow Moire´

Comment

826

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Fig. 6 – View of NXT post processor.

rate FE-models. However, since we analyse the curvature change instead of the height of the defect, we can visualise defects without the demand of 10 ␮m in accuracy. Comparisons between experiments and simulations (see Section 9) indicates that used simulation models (see Section 8) are accurate enough. But it demands that the springback prediction is good. With this software both experimental and numerical results can be evaluated. Since the software evaluates both kinds of results with the same method, the results can be compared in the same scale. The surface quality from both experiments and FE simulations can therefore be analysed both quantitatively and qualitatively in this software. It is also possible to compare numerical and experimental results in the same scale.

6.

Test models

6.1.

Double-curved panel

As a test model a double-curved test panel with an embossment was used. This model resembles the area around the fuel filler lid on an automotive body side. The model can be seen in Fig. 7. Several different materials were tested and both geometry and surface was analysed. The material and its characteristics can be seen in Table 2.

6.2.

Fig. 7 – Geometry of the experimental tool.

Table 2 – Material characteristics the double-curved panel Material AA6016 DC06 HSLA DP600

T (mm) 1.0 0.7 0.7 0.7

Rp0.2 (MPa) 127.9 124.7 364 289.4

R0

R45

R90

n

0,7 2.63 0.67 0.79

0.57 1.90 1.33 0.88

0.69 3.02 0.86 0.85

0.22 0.27 0.14 0.19

Door

In order to have a real automotive part as a test model, a section of a door panel, which is sensitive for surface defects, was analysed. The door which was used is the Volvo S40 rear door. This part was analysed after the first forming step and can be

seen in Fig. 8. The area which was analysed is marked with a rectangle. The door was analysed with the WMS-system/NXT post processor, and comparisons were made with results from a system, which is used for detection of surface defects, D-Sight.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

827

2. Application of lubrication with a roller. Adjust the amount by weighing the blank before and after application of lubricant. 3. Application of required amount of lubrication. Distribute the lubrication uniformly over the surface with a sponge. Weigh the sponge before and after application of lubricant.

Fig. 8 – Volvo S40 rear door after the first forming step.

7.

Experimental evaluation

7.1.

Double-curved panel

In order to have a uniform and well-conditioned blank holder pressure, gas springs were mounted around the punch and were internally connected. This gives a well-distributed blank holder pressure and possibilities to control the process accurately. Furthermore, the depth of the embossment was adjustable by applying/removing distance plates under the “embossment punch”. The tool was designed to be used in a single action press. Since the gas-springs supply all the necessary blank holder force, the tool is to be considered as a closed unit regarding the blank holder pressure. It only needs to be supplied with external gas pressure for the gas-springs. The die movement, blank holder force and punch force were registered. The surfaces in the draw radii and in the blank holder were induction hardened to 54–58 HRc since High Strength Steel (HSS) was studied in the project. This prevents the tool to be damaged by wear during the tests. In order to see the variation in the experiments, eight panels were manufactured for each material. In the current report results for panel #4–8 are presented.

7.1.1.

Surface roughness

In order to verify that the surface roughness did not vary between the experiments, seven points distributed over the surface of the punch and blank holder were measured before and after the tests. The measurements showed small deviations between the measurements (max 0.2 ␮m). It can be concluded that the tool wear was negligible.

7.1.2.

Lubrication

During the 3DS-project, different lubrication strategies were tested (Andersson and Krantz, 2003). 1. Application of abundance of lubrication on the blank in upright position and allowing all the excess of the lubrication to drain off.

In order to establish the best method, method 1 and 2 were used on two different blanks of the same material (DP600) and the distribution of lubricant was analysed (method 2 and 3 are similar). The amount of oil was analysed with an IR instrument called SC4000 oil detector. In order to have enough accuracy in the weight-procedure for method 2 a balance, Precisa Analytical Balance 6200C SCS (Series 300 SCS) FR, with an accuracy of 0.01 g was used. The results showed that method 1 gave an increasing amount of lubrication from top to bottom of the sheet. The difference was significant (a factor 2) while method 2 only indicated a distributed scatter due to the application method The scatter in method 2 is distributed over the surface, and therefore it does not have any significant effect on the result. In verification purpose, the draw-in of a test panel was analysed. The draw-in was found to be uniform, which confirms a well-distributed lubrication.

7.1.2.1. Conclusion. Method 2 gives a more balanced oil amount over the sheet surface. Therefore this method is used in the project.

7.1.3.

Blank dimension

The forming process was tested experimentally and numerically. After several iterations with the process parameters (Blank holder pressure, lubricant, draw depth, blank size, see Table 3), a suitable blank dimension was found to be a rectangle of 750 mm × 550 mm.

7.1.4.

Process parameters

Evaluation of different set-ups of process parameters was performed experimentally in the tool. The object was to create panels, which had surface defects around the embossment. Furthermore, no cracks or wrinkles should appear. The best conditions are listed in Table 3. These parameters were used for further analysis.

7.1.5.

Blank holder force

The pressure in the gas springs was linearly increasing with increasing forming depth. This means that the blank holder force increases with increasing draw depth. The blank holder force can be calculated as: F2 = F1 + K ∗ draw depth

(4)

where F2 is the blank holder force at the current draw depth. F1 is the start value. K is the linear growth coefficient. In Table 4 the values for the parameters F1 and K are defined.

7.1.6. Geometric evaluation 7.1.6.1. Draw-in. The draw-in was measured as the flange width with an ordinary steel scale (accuracy 0.5 mm) in order to have a reference to compare the simulation results with. A

828

Material

F1 (kN)

AA6016 DC06 HSLA DP600

535 533 553 630

K 4.5 4.5 4.5 5.4

1.5 1.5 1.5 0.7

Finarol B5754 Quaker 6130 Fuchs Quaker 6130

Table 4 – Blank holder forces

3.0 3.0 3.0 1.4

Amount of lubricant (g/mm2 )

Amount of lubricant (g/mm2 & side)

Oil

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

150/540 150/540 150/540 180/650

schematic picture of the procedure and the positions of the measurements points can be seen in Fig. 9.

37 40 43 43

30 30 35 35

7.1.6.2. Deflection after springback. The panel was placed on 3 support pins, which were placed on the embossment radii (see Fig. 10). Then two sections were measured with a CMM-machine. The distance between the points was 5 mm. Closer distances did not improve the results (1 mm was tested), and therefore all panels were tested with this distance. The position of the sections can be seen in Fig. 9.

AA6016 DC06 HSLA DP600

Draw depth (mm) Material

Table 3 – Suitable process parameters

Draw depth before embossment starts (mm)

BHP (initial setting of gas spring system (bar/kN)

Fig. 9 – Measurement of draw-in and evaluated sections. Section A corresponds to y = −58 and section B to y = −130.

Fig. 10 – Support points for section measurements.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

829

Table 5 – Simulation parameters Material

Young’s modulus (GPa)

AA6016 DC06 HSLA DP600

70 210 210 210

Coefficient of friction,  0.1 0.12 0.1 0.1

Tool speed (m/s)

1 2 2 2

Element size (mm) 8–1 8–1 8–1 8–1

Material model

Hill/Barlat Hill/Barlat Hill/Barlat Hill/Barlat

Hardening law

Isotropic/mixed Isotropic/mixed Isotropic/mixed Isotropic/mixed

Fig. 11 – Set-up for the measurement with the WMS-system.

Fig. 13 – Flange width.

7.1.7.

Surface evaluation

The surface was measured with the WMS-system (see Section 4). The configuration used is schematically described in Fig. 11. In the WMS-system a point cloud which described the shape of the surface was generated. The distance between each point was about 2 mm. The point cloud was then exported to TEBIS, which created a mesh from the point cloud. Since the amount of points was very large, the number of points was minimised in TEBIS before the mesh was created (see Section 3.4). The mesh was then imported into the NXT post processor (see Section 3) for evaluation. As a reference,

Fig. 12 – Elements for which the stress histories were analysed.

the panel was also analysed in the D-Sight system. The whole surface was analysed for the double-curved panel, but the WMS-system could not handle the whole door panel with enough accuracy. Instead, the area around the door handle was chosen to be analysed. The results were then transferred and analysed in the NXT post processor. The same door panel was also measured in the D-Sight system.

Fig. 14 – Punch forces.

830

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

8.1.2.

Element formulation

LS-DYNA’s fully integrated elements ((Hallquist, 2003) with 5 integration points through the thickness were used for both forming and springback simulation. Previous studies (Andersson and Holmberg, 2002) showed that 5 integration points through the thickness was enough and therefore this number was chosen also in this study.

8.1.3.

Element size

The effect of adaptive mesh was compared to uniform mesh size. The difference was negligible, and therefore adaptive mesh was used. Furthermore, the effect of using smaller element size in the area around the embossment was studied, but it also showed negligible effect and was not used. Convergent results were achieved with the adaptive element size from 8 to 1 mm (Andersson and Krantz, 2003).

8.1.4. Fig. 15 – Springback in evaluated sections, material AA6016.

8.

Simulation of the double-curved panel

It is necessary to consider the springback when surface defects shall be evaluated. The model should represent the geometry after springback, and therefore the same demands as for achieving accurate springback result is applicable for surface defects. The simulations were performed in the dynamic explicit code LS-DYNA (Hallquist, 2003; LS-DYNA, 2007). A similar model have been analysed in a previous study (Andersson and Hertzman, 2001). In that work both parameter studies and tests of different measuring techniques were performed. However, it was concluded that both numerical and experimental results showed the same tendency, but the springback was not predicted accurately. Furthermore, it was not possible to detect any “teddy bear ears”. In this work these results were used and complemented with additional studies on effects of symmetry and locking of rigid body modes for springback analysis. Furthermore, the effects of different material and material models were analysed in this study.

8.1.

Simulation model

The simulation parameters are described in Table 5. As contact algoritm the penalty method, available in LS-DYNA, was used.

8.1.1.

Tool geometry

The tool was described with 12 elements over the radii and a maximum element size of 10 mm. In order to have a better tool description also VDA-surfaces were evaluated. The difference in results between a discretesised tool and a VDA-description were negligible and therefore discretesised tools were used for the study (Andersson and Krantz, 2003).

Coefficient of friction

Since all process parameters were well known except the coefficient of friction, this parameter was studied and the value was fitted to the punch force and the draw in. The draw-in was evaluated as the flange width after springback (see Section 7.1.6). In the optimisation process of the friction coefficient parametric design (Ford Design Institute, 2007) was used. A base model (described in Table 5) was used and the friction coefficient was varied until convergence with experimental punch force and draw in was reached. The quadratic yield function introduced by Hill (1948) with isotropic hardening was used for material modelling for all materials during these tests, and the obtained friction coefficients were used in the other simulations. In the 3DS-project, the coefficient of friction was measured by different methods, which corresponds to different regions of the tool (blank holder surface, draw radii, etc.). The measurements were done by 3 different partners (Cockerill, Pechiney and Arcelor (Wouters, 2002)). In the measurements the same amount and sort of lubricant was used as in the experiments. Furthermore, in the tests, the same tool material as in the experimental tool (with the same roughness) was used. The achieved coefficients of friction are pressure dependent, but since a constant friction was used in the simulations, an approximated value was used. In the experimental tool the pressure does vary over the surface, but this was simplified to an average value. However, different coefficients of friction were used for the different tool parts (die-holder, punch-die, draw radii, etc.). The achieved values were very high and when they were applied in the simulations, the fit to draw-in and punch forces in the experiments was very poor and the simulations indicated cracks in the corners. The achieved results can be seen in Table 6.

8.1.5.

Material and hardening model

Two material models were compared, Hill and Barlat’s (Barlat and Lian, 1989) three-parameter yield function (Barlat) models. For Barlat’s model stress exponent m = 6 was used for steel and m = 8 for aluminium. The isotropic hardening was described by Swift’s law for steel material and Voce’s law for aluminium.

831

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Table 6 – Coefficient of friction Material

Fitted coefficients of friction

Experimental coefficients of friction Draw radius

AA6016 DC06 HSLA DP600

0.1 0.12 0.1 0.1

Punch

0.16 0.25 0.22 0.19

Blank holder

0.21 0.3 0.27 0.24

0.175 0.18 0.2 0.16

Table 7 – Material parameters with isotropic hardening Material

Y0 (MPa)

AA6016 DC06 HSLA DP600

127.9 124.7 364.0 289.4

ε0

n

0.00385 0.00958 0.00099

0.257 0.139 0.19

n

(5)

Voce: Y = Y0 + RSAT [1 − exp(−CR ε¯ p )]

X

SAT

¯



p (  − X) − X ε¯˙ ,

CR 11.0

522.0 693.8 1082.0

Bouvier, 2002). The obtained material parameters can be found in Tables 7 and 8. According to Banabic et al. (2000) the r-value in the Hill model should be ≥1. Therefore r = 1 is used when the measured r-value is <1. If the r-value is <1 in all three directions we use von Mises’ (1928) model.

(6)

8.1.6.

and the kinematic hardening, i.e. evolution of the back stress (Lamitre and Chabouce, 1990; Teodosiu and Bouvier, 2002), by: ˙ = Cx X

RSAT (MPa) 200

Swift: Y = C(ε0 + ε¯ p )

C (MPa)

X(0) = 0

(7)

The kinematic hardening was only implemented and evaluated with the Hill-model. The detailed material characterisation of these materials, as well as the selection and identification of appropriate material models, was done within the 3DS-project (Teodosiu and

Stress analysis

The stress is the driving parameter for springback. The relaxation of the internal stresses in the material during unloading results in elastic recovery, which changes the geometry of the part. In order to understand the FE simulations, the stress history was analysed for a number of elements, who did slide over the draw radii (see Fig. 12). Their initial position is defined in Table 9.

9.

Results

9.1.

Double-curved panel

The results from the tests with the double-curved panel indicated big differences in the behaviour for the different materials. Even though the thicknesses were different, clear tendencies for larger springback and surface defects could be observed for the HSS-materials and the aluminium compared to mild steel. Furthermore, the predicted results were rather good, and the surface defects detected in the experiments could be foreseen in the simulations. A comparison of the numerical and experimental results can be seen below.

9.1.1.

Fig. 16 – Springback in evaluated sections, material HSLA.

Optimisation procedure for friction coefficient

The parameters, which were used for optimisation of the friction coefficient for the model, were punch force and drawin as a function of the coefficient of friction. The draw-in was measured by the flange width after forming (see Section 7.1.6). Comparison between numerical and experimental results showed that the draw-in agreed within 1 mm. These results were regarded as satisfactory, and the corresponding friction values were used. The comparison can be seen in Figs. 13 and 14. In the diagrams the variations in the experimental results are marked with bars in the top of the histograms and in the curves. As mentioned in Section 7.1, eight panels for each material were tested.

832

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Table 8 – Material parameters with kinematic hardening Material

Y0 (MPa)

AA6016 DC06 HSLA DP600

127.9 121.1 361.86 285.0

ε0

n

0.00224 0.00697 0.00128

0.221 0.076 0.14

C (MPa)

1 2 3 4 5

9.1.2.

X

Y

−175 0 175 315 315

−215 −215 −215 −130 0

Springback

The springback was measured in two different sections as described in Section 7.1.6. In the experiments the top surface was measured. This corresponds to the outer surface in Fig. 12. In the numerical evaluations, however, the middle surface is modelled. In order to compare the same location for experiments and simulations, an offset of t/2 is applied to the numerical mesh. This generates an error due to the thickness change, but since the strains in the upper surface are very low (see 3DS report (Andersson and Krantz, 2003)), this effect was insignificant. The results can be seen in Figs. 15 and 16. In the diagrams the variations in the experimental results are marked with bars in the curves. The different material models showed similar results which are in good agreement with the experimental results regarding the small amount of springback. For material AA6016, the reason for the shift in z-coordinate can be that the panel “opens” (springback) differently in the experiments and the simulation. This appears as a shift in the z-coordinate. For the HSLA material, the experimental results indicate a warping in the section analysis. This is due to an “oil canning effect”. This effect is pronounced by the choice of support points (see Fig. 20). Tests by exchanging panels between the partners in the 3DS project (Andersson and Krantz, 2003) showed that the influence of choice of reference points was significant. Symmetric results were obtained when the support points were located in the corners of the top surface, or when the panel was placed on a flat table without pins as support. The location in the centre was chosen in order to affect the panel surface as little as possible. The panel made of the HSLA-material is very instable and this is causing these results. There is also a big variance in experimental results for this material compared to the others. Since the part is symmetric, this “oil canning” behaviour cannot be captured by the simulation. However, the different material models showed similar results if the overall curvature is regarded.

9.1.3.

146.5 1.87 8.5 21.9

466.9 528.0 726.2

Table 9 – Initial location of elements (mm) Point no.

Cx

Surface defects

The surface defects are visualised in the NXT post processor and the results can be seen in Figs. 17 and 18. The results are taken in the x-direction with a wavelength of 50 mm. In these figures the scale was set to −0.1 to 0.1 (1/m). The original curvature of the punch is 0.067 (1/m). Dark indicate depression

XSAT (MPa) 34.9 58.1 102.4 198.7

RSAT (MPa) 200

CR 11.0

and bright indicate elevation. These results are complemented by a section cut, where the angle variation (see Section 9.1.4) is shown. The results were also compared to D-Sight results. In Figs. 17 and 18 it can be seen that the surface defects mainly appear as large waves on the surface (shift in curvature value). The results achieved with the WMS-measurement and NXT post processor shows good agreement with the results obtained with the D-Sight system for all materials. It can be concluded that the experimental results, obtained with the WMS/NXT post processor are reliable. It can also be seen that the results from the FE simulations also show good agreement with the experimental results. The appearance of “teddy bear ears” cannot directly be seen, but the area around the embossment shows a dent, which is more pronounced at the smaller corner radii than at the larger. For material HSLA, the warping cannot be seen in this type of evaluation since it is the curvature change which is plotted. This can be explained by that the surrounding area is the reference and not any fixed points in space.

9.1.4.

Section analysis of curvature change

In order to visualise how the curvature of the surface changes, the gradient of the changes is plotted for one section. In the diagram the changes in curvature will appear as deviations from 0. If the change in curvature is not a design feature, the change indicates a defect. Section B (y = −130, see Fig. 9) is chosen for comparing the overall angle change. The results are shown in Figs. 19 and 20. Regarding the overall curvature change (section B) the simulation is in good agreement with the experimental results. A small difference in magnitude can be seen. The deviation is smallest for AA6016 and largest for DP600. Furthermore, the results are not symmetrical. This can probably be explained by the unsymmetrical appearance of the embossment.

9.1.5.

Stress analysis

The stress histories of 5 elements, which passed the draw radii during the forming process, were recorded. The positions of the analysed elements are described in Section 8.1.6. The results can be seen in Figs. 21 and 22. It can be seen that the stresses for the different material models are very similar. This similarity can also be seen in the section plots of the springback. It can also be seen that the magnitude of stresses are much higher for HSS-material than in AA6016 and DC06 (Andersson and Krantz, 2003). Furthermore, it can be observed that the springback for HSS is much higher than for AA6016 and DC06. Since the level of stresses will cause springback, this is a confirmation of the statement in Section 8.1.6.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

833

Fig. 17 – Curvature evaluations with D-Sight and the WMS/NXT post processor of material AA6016.

9.2.

Door

In order to test the NXT post processor as evaluation system, a door outer was analysed. The door was also analysed in the D-Sight system for comparison. The results can be seen in Fig. 23. These defects are very small and by visual inspection the defect in the handle area was rated and got an 8 (see Section 2.1) in the VCBC rating system.

9.3.

Summary

The results from the simulations were in good agreement with the experimental results. The surface analyses indicated that the method proposed in this work is applicable for both sim-

ulation end experimental verification of the appearance of surface defects. It can also be concluded that it was very difficult to create pure “teddy bear ears” in this experimental tool. However, it was possible to create similar defects such as the dent around the embossment. It was also possible to create a surface which contained surface defects which appear as waves in the surface. The material parameters used in the simulations are based on experimental tensile and shear test. The yield surface was not compared between experiments and simulations. Therefore it is difficult to draw any general conclusions from these tests, but for this set of data following conclusions can be drawn:

834

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Fig. 18 – Curvature evaluations with D-Sight and the WMS/NXT post processor of material HSLA.

• Barlat’s model would be preferable to use. • Mixed hardening is favourable to use compared to isotropic hardening for the Hill model.

Fig. 19 – Curvature change for section cuts for AA6016.

10.

Discussion and conclusions

Based on the results it can be seen that the methodology used in this study can be used for predicting and verifying the appearance of surface defects. The strength in the methodology is that the same tool, the NXT post processor, is used for evaluation of both numerical and experimental results. Then there is no issue whether the evaluation procedure of the numerical or the experimental results is the better. It will also be easier to compare the results since the same scale and method is used. It was observed that it was impossible to use measured friction coefficients in the FE simulations of the double-curved panel (see Section 8.1.4), even though great care was taken in order to have the same conditions during the friction tests

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Fig. 20 – Curvature change for section cuts for HSLA.

as during the forming of the double-curved panel (same tool material, lubrication, etc.). However, the friction in the FE simulations was assumed to be constant over time, but varied over different area of the tool. Since the friction coefficient is

835

dependent on several parameters, e.g. pressure and speed, a better result could have been reached, if the friction had been modelled more accurately. However, this was not tested in this study. The different materials showed the same tendency, but different magnitude in the macro-geometric defects. The different materials also showed different scatter in the experimental trials. One explanation for these differences is the hardening behaviour of the different steel materials. The larger material hardening, the more significant small variations in the stress field will be. The exception is the HSLA-material, which showed a significant oil canning effect (see below). Therefore, it is difficult to draw any general conclusions from the results for this material. It can, furthermore, be observed that the springback for HSS (HSLA and DP600) is much higher than for AA6016 and DC06. It was also noted that the accuracy in simulations was best for AA6016 and

Fig. 21 – Stress analysis, material AA6016. In the left diagram point #2 is analysed and in the diagram to the right point #5 is analysed. The Hill result with isotropic hardening is represented by a bright grey line, Barlat-isotropic hardening with a medium grey and Hill with mixed hardening by a dark grey line.

Fig. 22 – Stress analysis, material HSLA. In the left diagram point #2 is analysed and in the diagram to the right point #5 is analysed. The Hill result with isotropic hardening is represented by a bright grey line, Barlat-isotropic hardening with a medium grey and Hill with mixed hardening by a dark grey line.

Fig. 23 – Application of two systems for detecting surface effects to a production part (door outer). Areas marked with circles indicate depressions. Dark indicate depression and bright indicate elevation.

836

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

DC06. The reason for this can be that the hardening behaviour is more difficult to describe and is more significant for the HSS-materials. The reason for this can be that materials with larger material hardening require a more accurate description of the hardening behaviour in the material model used in the FE simulations. The material models used in this study simplify the hardening by assuming isotropic hardening. However, mixed hardening for the Hill-model was tested which improved the accuracy in the results. The use of mixed hardening requires better test procedures for retrieving accurate material data in order to have a reliable material description. One example is given in the work by Teodosiu and Bouvier (2002). Regarding the double-curved panel and material HSLA, the results showed an oil canning effect in the springback measurements. Depending on the choice of reference points, the oil canning effect was very different. Within the 3DS-project, different partners did chose different reference points for the springback measurements. If the reference points were chosen close to the embossment (see Fig. 10) a large oil canning effect was found. On the other hand, if the reference points were selected close to the corners of the top surface or the panel was placed directly on a flat surface, the oil canning effect decreased significantly (Andersson and Krantz, 2003). In the simulations, the results are symmetric and the oil canning effect cannot be detected. Regarding the simulation results, a clear tendency of mixed hardening giving smaller springback than isotropic hardening was observed. It was also concluded from these tests that Barlat’s model is preferable to use. However, since the yield surface was not compared between experiments and simulations, it is difficult to draw any general conclusions from these tests. During the 3DS-project (3DS Report, 2002; Col, 2002; Andersson and Krantz, 2003; Wouters et al., 2002; Teodosiu and Bouvier, 2002) different measurement systems were tested. Since the D-Sight system and the WMS-system were available at Volvo, these systems were used in the project. Furthermore, the WMS-system gave the possibility to export measured data with enough accuracy and was therefore used for surface evaluation in the NXT post processor. The results from the Volvo S40-door showed that the different systems, the D-Sight and the WMS/NXT post processor, indicated the same regions with defects. However, the distribution was slightly different. This can depend on differences in point density for the measurement points, differences in accuracy between D-Sight and WMS-system or errors introduced in the transformation process from point cloud into mesh (in TEBIS) for the WMS-measurement. Still, it is very difficult to compare the results quantitatively, since different criteria are used. For this more comparisons need to be done.

11.

Further work

• Perform more careful material characterisations, where also the shape of the yield surface is determined.

• Implement the Barlat material law, or some other more advanced model, in combination with a mixed hardening model. • Further evaluations of the accuracy in the springback predictions when using improved material data and material models. • In order to have an evaluation system, which is in conjunction with the subjective scale 1–10 mentioned in Section 2.1, the scale in the NXT post processor should be correlated to the subjective scale. • Further numerical and experimental evaluations of the current methodology on more outer parts.

Acknowledgements This projected was funded by the IMS-project 3DS (contract G1RD-CT-2000-00104 (2000)). The author is grateful to Fredrik Krantz (ProEngCo) for valuable help with the experimental part within this project. Furthermore Erland Max, Leo Gurmark, Ola Claesson (Volvo Corp resp. Volvo Cars Body Components) have been very helpful with the surface measurements. Thanks are also due to the staff at IDC (Industrial ¨ development Center in Olofstrom) for their valuable help and support. Another big help has been Professor Kjell Mattiasson (Chalmers University of Technology) both with help of implementation of the hardening law and in the discussions during the project. Dept 81153, Professor (Volvo Cars Body Components), Claes Magnusson (Kristianstad Univer˚ (Division of Production and sity) and Professor Jan-Eric Stahl Materials Engineering, Lund University) have also contributed with valuable help and discussions. At last I wish to thank ˚ Anders Skogsgardh and Professor Nader Asnafi (Volvo Cars Body Components) for their support of this project. The author would also like to acknowledge the support from PROPER and ProViking.

references

Andersson, A., Hertzman, J., 2001. Evaluation and Comparison of Surface Defects on a Simplified Model for the Area Around the Fuel Filler Lid by Simulation and Experiments. NUMIFORM 2001, Besancon, France. Andersson, A., Holmberg, S., 2002. Simulation and Verification of Different Parameters Effect on Springback Results. NUMISHEET 2002, Jeju Island, Korea. Andersson, A., Krantz, F., 2003. Report tool #6, Numerical and experimental evaluation, “3DS, Digital Die Design System”, contract G1RD-CT-2000-00104, Sweden. ¨ Banabic, D., Bunge, H.J., Pohlandt, K., Tekkaya, A.E., 2000. Formability of Metallic Materials. Springer Verlag, Berlin, Heidelberg, New York. Barlat, F., Lian, J., 1989. Plastic behaviour and stretchability of sheet metals. Part I. A yield function for orthotropic sheets under plane stress conditions. Int. J. Plast. 5, 51–66. Col, A., 2002. Presentation of the “3DS Research Project”, NUMISHEET 2002. Jeju Island, Korea. D-Sight, 2007. http://www.lmint.com. Ford Design Institute, 2007. www.fdinext.ford.com. ˚ Gasvik, K.J., 1995. Optical Metrology, second ed. John Wiley Sons. Hallquist, J.O., 2003. LS-DYNA Keyword User’s Manual. Livermore Software Technology Corporation, Livermore.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 9 ( 2 0 0 9 ) 821–837

Hansson, P., 1999. Optical Characterization of Paper Structures and Print Defects. Fysik-Optik, Stockholm, KTH. Hill, R., 1948. A theory of the yielding and plastic flow of anisotropic metals. Proc. R. Soc. Lond. Ser. A 193, 281. Huntley, J.M., Saldner, H.O., 1997. Shape measurement by temporal phase unwrapping: comparison of unwrapping algorithms. Meas. Sci. Tech., 922–986. Kase, K., Makinouchi, A., Nakagawa, T., Suzuki, H., Kimura, F., 1999. Shape error evaluation method of free-form surfaces. Comput.-Aid. Des. 31 (8), 495–505. Kinell, L., 2003. Optical Shape Measurements Using Temporal Phase Unwrapping. Doctoral Thesis. Lulea˚ University of Technology, Sweden. Lamitre, J., Chabouce, J.L., 1990. Mechanics of Solid Materials. Cambridge University Press, Cambridge. LS-DYNA, 2007. http://www.lstc.com. Makinouchi, A., 1996. Sheet metal forming simulation in industry. J. Mater. Process. Technol. 60, 19–26. Makinouchi, A., Teodosiu, C., Nakagawa, T., 1998. Advances in FEM simulation and its related technologies in sheet metal forming. CIRP Ann. Manuf. Technol. 47 (2), 641– 649.

837

Max, E., Nilsson, P.H., Larsson, P., 2003. WMS, a new principle for measuring finish-related topography on sheet steel panels. In: Metrology and Properties of Engineering Surfaces 9th Int. Conference, Halmstad, Sweden. Saldner, H.O., Huntley, J.M., 1997. Temporal phase unwrapping: application to surface profiling of discontinuous objects. Appl. Opt. 36 (13), 2770–2775. TEBIS, 2007. http://www.tebis.se. Teodosiu, C., Bouvier, S., 2002. Selection and identification of Elastoplastic Models for the Materials used in the benchmarks, 18-months progress report, “3DS, Digital Die Design System”, contract G1RD-CT-2000-00104, France. Thomas, T.R., 1999. Rough Surfaces, second ed. Imperial College Press, UK, pp. 11–34. von Mises, R.V., 1928. Mechanics of plastic deformation of chrystals 592. Angew. Math. Mech. 8, 161–185. Wouters, P., Daniel, D., Magny, C., 2002. Selection and verification of friction models for the 3DS materials, determination of the friction behaviour (deliverable No 3b), “3DS, Digital Die Design System”, contract G1RD-CT-2000-00104, Netherlands. 3DS Report of Work Package 1, 2002. Development of Methods for Evaluating Forming Defects, Japan.