Integration of casting simulation and finite element
calculation
Dr. ir. M.E. Heerschap - MG technical solutions, Noordkade 64, Waddinxveen, The Netherlands (www.mgts.nl) Email: mart.heerschap@mgts.nl Dr. Y. Ling, Dr.ir. F. Mampaey, ir. R.E.F. Kastelein - WTCM Foundry Centre, Technologiepark 915, B-9052 Zwijnaarde, Belgium (www.wtcm.be) Email: roy.kastelein@wtcm.be
ABSTRACT: Both casting simulation and finite element simulations exist for quite some time. Up to now these were separate fields of expertise. In general a finite element analysis of a casting is performed to make sure that all mechanical requirements for the product are met. For the finite element analysis the usual assumption is that the product is initially stress free in unloaded condition. The material properties, most often taken from literature, are assumed to be homogeneous throughout the complete product. Both assumptions are not correct. After casting important residual stresses remain, caused by solidification and cooling to room temperature. The material properties are distributed in an inhomogeneous way over the complete product, depending on the wall thickness in the part. By integration of casting and finite element analysis these drawbacks has been overcome. The material properties are modelled from experiments and the magnitude and the distribution are predicted by the casting simulation. The thermal or residual stresses in the casting are calculated too and transferred to the finite element analysis for a more accurate prediction of the actual behaviour. In this paper the integration of ViewCast and MSC.MARC is described.
2
Traditional development of products by casting Products manufactured by casting are designed with a focus on the functional requirements for the product. The limitations of the production process require design adjustments usually. These adjustments are in general small details like clearance angles, radii and moderation of thickness changes. These adjustments are made simultaneously or in a later stage of the design process. When the product drawings are finished, a model is made. This model also includes the pouring system and risers. The foundry experts indicate how the pouring system must look like. The design of the pouring system is based on years of experience completed with gut feeling. After fabrication of the production tools, the prototypes are cast. The prototype is examined by X-ray or cut in pieces to reveal voids in the projected area. Voids elsewhere in the casting remain undetected. If voids, or other casting defects, are found, the pouring system and risers are modified in an attempt to solve the problems. Besides the pouring system and risers adjustments also can be made to the design of the product or production process. When the prototype passes the visual inspection for casting defects, the prototype is subjected to an experimental mechanical test. If the product fails the test, obviously the design must change. After the design change the complete process is repeated to arrive at an acceptable product. This process is shown schematically in figure 1. It will be quite obvious that this way of producing cast parts is a slow and expensive one. To reduce the time for the design process it is quite common to over dimension the casting. In that case the specifications for both the functional behaviour of the product as well as the requirements originating from the production process are more easily met. In cases where the cast product is cheap to manufacture and the weight is of less importance this approach is a valid one. However, quite often the weight of the product matters. For example an expensive alloy is used for the casting. When the design process is also focussed on creating an efficient design, a number of design iterations are inevitable. These design iterations are time consuming and expensive. Improvement of the design process can be achieved by replacing the experimental mechanical tests by finite element calculations. A major leap in process improvement can be achieved by introducing casting simulation. All trial and error steps are the virtual, implying that the first prototype has no casting flaws anymore. Description of casting simulation The simulation of the casting process is carried out using the finite volume (FVM) or finite element method (FEM), figure 2. Both methods have their specific advantages and disadvantages, which can be summarized as follows. The FEM mesh is very suitable for thin-walled complex castings; the number of elements is usually quite low so the calculation can be quite fast. However the creation of the mesh requires a lot of skills and time of the operator, especially for the complex castings. The FVM mesh generator is fully automatic and takes also for a complex casting but a few seconds. The physical interpretation of a fluid medium passing though volumetric cells is more straightforward to understand and to solve. The representation of the geometry is like a LEGO-model like figure 3 and as a consequence some energy-loss can occur on curved surface boundaries. A fine mesh is needed to minimize this effect. Without going into too much detail, it is clear that hard work is being done on both methodes to overcome the disadvantages. Anyway the casting simulation program used in this paper is ViewCast, which is based on the finite volume method. ViewCast is used to predict voids in a casting. Voids can be generated in a casting by three different phenomena: - Gas formation - Air entrapment. - Shrinkage or porosity Gas formation is a problem that occurs when the melt is not enough degassed. Hydrogen or Nitrogen atoms are dissolved in the liquid metal, which recombine to H2 or N2 during solidification. Thousands of small spherical pine-holes are visible inside the casting. By a good degassing procedure this problem is solved. In the casting simulation a perfectly degassed melt is assumed.
3 Air entrapment is caused by a turbulent mould filling, which is indicated by carefully interpreting a mould filling simulation of ViewCast, c.f. figure 4. In reality the entrapped air will attempt to escape through the mould, but it is very difficult to simulate this effect accurately, because the permeability of the mould is not equally distributed. In ViewCast an empirical model, based on casting experiments, has been build-in to impose a velocity in the cavity below 50 cm/s at which no turbulence will occur in a liquid metal flow. Most of the problems in a foundry however are caused by shrinkage, due to a volume reduction of an alloy during the solidification process. The volumetric shrinkage depends strongly on the type of alloy. Aluminium alloys for example have a volume reduction of ca. 5-7%, but certain ferrous alloys with enough carbon can have the shrinkage fully compensated by carbon-expansion. Shrinkage defects appear like big craters in the casting and are very harmful for mechanical failure and leakage. Obviously foundries must prevent shrinkage defects or porosities in their castings thus they add so-called risers (reservoirs) on top of their castings to refill the casting during solidification. For this casting simulation software like ViewCast is developed. With ViewCast the mould filling and solidification can be simulated and the shrinkage defects are predicted accurately both in location and in volume, cf. figure 5. Empirical models in ViewCast prescribe the volume needed to compensate the shrinkage; a casting simulation is carried out to check if the risers can feed the complete product, what depends on the geometry of the casting. By adapting the riser location and the pouring system the shrinkage defects can be solved before tooling equipment is produced and a prototype is cast. Recent developments in ViewCast can solve deformation problems caused by thermal stresses during cooling after casting or after heat treatment, figure 6. High thermal stresses on certain spots can even generate cracks in the casting. It will be clear that thermal stresses in a casting are very important on the failure behaviour during mechanical loading in its application. All these phenomena can be predicted by casting analysis instead of working on a trial and error basis. Integration and interface Usually when products are designed finite element analysis is used to check that the (thermo-) mechanical requirements for the product are met. In a more advanced approach the finite element analysis is combined with an optimization, to create an optimum design. The results of a finite element analysis, or a finite element driven design cycle, however, are as good as its input. In other words, when the starting point of a finite element analysis contains errors, the final result will be wrong. Even when the structure has been optimized, the product is sub- optimal at best. It might even get worse; the requirements for the product are not met anymore. This situation is not purely hypothetical, illustrated by the following example. A part intended for the use in off shore applications was designed as casting of a high strength steel. After casting the part was subjected to several heat treatments. The structure was optimized using a stress-free initial state and using the strength data from literature. The prototypes were subjected to mechanical testing. The large specimens failed at substantially lower loads than was predicted by the finite element computations. Research on the remains of the failed specimens showed that the heat treatments had the intended results for the smaller specimens but failed for the large specimens. Modification of the heat treatment later on resulted in shrinkage cracks. All these problems could have been prevented when the integration of casting simulation and finite element simulation had been used. This example shows the clear need for information transfer from the casting simulation to the finite element simulation. In order to achieve an information exchange between the casting simulation and the finite element simulation an interface has been made. The general idea behind the interface is to use some of the data computed for the casting simulation in the finite element calculation. The interface is initially set up as a stand- alone program. The data required by the finite element simulation is in some cases already present and can in other cases be computed relatively simple. The next step was to define a standard for the information exchange. A dedicated file with a pre-defined format was used. This file is read by the interface. The interface needs also to know about the MSC.MARC finite element model. Based on this information the mapping of the results of the casting simulation to the MSC.MARC model can take place. This is shown in figure 7 schematically. The major difficulty in the mapping of the ViewCast data to the MSC.MARC mesh is the difference in the two meshes. The ViewCast mesh is a fairly dense finite volume mesh and the MSC.MARC mesh a coarser finite element mesh. In the example discussed later, the ViewCast mesh consists of a few hundred thousand elements, while the MSC.MARC mesh contains less than twenty thousand elements only. This means that per MSC.MARC element there are many more ViewCast elements. In spite of the fact that an averaging technique is implemented, it is likely that the resulting pre-stresses do not form an equilibrium. Therefore it is common practice to make the initial
4 load case in MSC.MARC a case where no loads, except the pre-stresses are applied. The structure can set and the initial stress state can be visually compared to the ViewCast data. Another issue is the CPU time required by the interface. As stated before, there are many ViewCast elements and quite some MSC.MARC elements. In order to search for the correct combinations of elements the most straightforward search algorithm requires a CPU time proportional to the number of elements squared. It is needless to mention that CPU times using this amount of elements will be excessive. To limit the amount of CPU time required a voxel based search technique is implemented. This search technique is far more efficient and limits the CPU time to less than a minute. The output of the interface is a part of a MSC.MARC input or a MSC.MARC Mentat procedure file. The part of the MSC.MARC input file is the part of he file that contains the initial stresses of the FE model. When the original MSC.MARC input file and this extra file are merged, the analysis can be started. A more solid way is to write out a MSC.MARC Mentat procedure file, read it using Mentat and have Mentat write out a new MSC.MARC input file. In the latter case the data of the initial stresses are also stored in the Mentat mud file.
Example A manufacturer of lighting poles and accessories produces base plates for lightening poles in large quantities. An example of such base plate is shown in figure 8. Primary objective of the manufacturer was to create a lighter, more efficient base plate. Secondary objectives were to create a smoother exterior and reduce the assembly time. An initial design of the new base plate was made, analysed using MSC.MARC and adapted where required. These analyses took place in the usual way. Meaning that a homogeneous strength and an initially stress free state were assumed. When this converged to an acceptable situation, the current configuration was used as start point for the casting analysis. The casting analysis started with the initial design of the fillers and the risers. The first analysis was completed. Design changes were incorporated and re-analysed. In a certain stage design modifications were made also in the product. By smart redistribution of material this was achieved without a weight penalty. When the casting analysis had converged a new finite element analysis was performed, now introducing the thermal stresses in the finite element model. A modified resulting stress distribution was found. Remarkable to mention is that the maximum Von Mises pre-stress was around 80 MPa. The yield stress of the used aluminium alloy is 160 MPa. This implies that at some locations of the model approximately 50% of the yield strength is added or disappeared, depending on the sign of the stresses. These pre-stresses are apparently influencing the final stress distribution significantly. Incorporating the effect of the pre-stressing some additional material redistribution and weight savings were performed. Resulting in a truly optimum structure. The finite element model consists of two base plates four pre-stressed bolts, a part of a lighting pole, a rigid surface representing the concrete base and four rigid surfaces representing the anchor bolts. This is shown in figure 9. The lighting pole is loaded subsequently by bending moments in two directions, shear forces in two directions and a simultaneously acting twisting moment. The analysis started with a load case where the two parts of the base plate were bolted together, by pre-stressing the bolts. The pole is squeezed in between the base plates. Some local plasticity occurs in the pole. The subsequent load case tightens the anchor bolts and finally all load combinations are applied. This procedure resulted in a new range of base plates, of which an example is shown in figure 10.
Conclusions and recommendations The procedure as demonstrated here shows that the accuracy of finite element of cast products can be increased significantly by integration with a casting simulation. The magnitude of frozen stresses after casting of the product, can reach 50% of the yield stress. Ignoring these effects can lead to expensive redesigns in a later stage in the design process. The technique of casting simulation is valuable as a trouble shooting technique, but even more as predictive engineering tool. Future improvement of the integration software is aimed to make the initial stresses available to other finite element codes than MSC.MARC only. At the moment it is investigated what effect the frozen stresses have on vibrating structures. It is believed that for example the cast parts of an internal combustion engine may be effected significantly by the frozen stresses and possible casting defects.
5
figure 1: traditional design and production of cast products Initial design Design adjustments Model with fillers and risers Prototype casting Visual inspection by saw cuts
Requirements met?
Mechanical tests Requirements met?
Functional prototype Y Y N N
6
figure 2: typical difference in mesh between the finite element method (FEM) and finite volume method (FVM)
figure 3: example of a FVM-mesh FEM FVM
7
figure 4: mould-filling example of a part (which is calculated half for symmetrical reasons)
figure 5: shrinkage prediction in a casting (which is calculated half for symmetrical reasons)
8
figure 6: thermal stress (up to 100 MPa) and deformation (50x) due to different cooling rates
figure 7: flow diagram interface between ViewCast and MSC.MARC
Interface ViewCast interface file MSC.MARC input file MSC.MARC procedure file Addition to MSC.MARC MSC.MARC Mentat New MSC.MARC Manual merge