|
|
 |
Virtual Molding: A Challenge to the Analysis Industry
Geoffrey Engelstein, GR Technical Services, Mountainside, NJ
Introduction
Injection mold filling, cooling, and warpage analyses are very powerful diagnostic and troubleshooting tools. However interpretationof results can often be tricky, involving the synthesis of multiple analysis outputs to evaluate a single molding parameter. Advances in computer visualization and rendering techniques open the door to a new way of interpreting results. Instead of presenting results in complex contour charts, 'Virtual Molding' will simply show what the molded part will look like, incorporating weld lines, blush, color, texture, and warp.
The Problem
At its heart, Process Simulation is a simple concept. The purpose of analysis is to answer a few basic questions:
- What will the part look like?
- How will the part perform?
- What are the best machine settings to mold the part?
The last question is adequately covered by today's software (although the information could be presented in a simpler fashion). Part performance falls mainly into the area of structural and thermal analysis. These are typically outside the scope of process simulation, however new products are being developed which transfer process-related information (such as weld location and strength) to standard structural analyses.
The first question is what the bulk of process simulation, and VM is all about. Specifically, some of the problems we might be interested in are:
- Weld line location and visibility
- Gate Blush
- Flow Lines
- Sink Marks
- Burning
- Gas Traps
- Transparency
- Warp
Here is a partial list of results categories from an analysis package:
- Melt Fronts
- Pressure
- Temperature at Fill
- Layered Temperature
- Bulk Temperature
- Peak Temperature
- Flow Vectors
- Weld-Line Meeting Angle
- Weld-Line Underflow
- Shear Stress
- Layered Strain-Rate
- Mass Density
- Gap Thickness
- Volumetric Shrinkage
- Heat Flux
- Frozen Layer Thickness
Plus most of this information can be looked at at any point during the injection cycle, so there is a time element as well.
Let's say we want to know how bad the gate blush is going to be. Where do we look for this information? Is it temperature dependent? And which temperature data (bulk, layered, peak) is relevant ? Should we look at the beginning of the fill or the end of the fill or some point in between? What data do we look at if we want to know about flow lines?
(The answer, by the way, is that blush can be determined by looking at shear stress. Potential for flow lines can be evaluated by looking at pressure in conjunction with flow vectors.)
The correlation between output data and what we really want to know can be overcome by training -- the data is there -- but it presents an unnecessary hurdle to using the technology. Another problem is that even when we know where to look for the answers, the numbers that the simulation produces are not directly correlated with the actual appearance. For example, the analysis shows the shear stress as 50 psi maximum around the gate. Whether or not this results in visible blush will depend on many factors, including material specs, color, and surface texture. Polystyrene will blush at a lower shear stress than Delrin. Flat black surfaces will show more blush than textured white surfaces. Some software packages will help in this process, by giving recommended maximum shear levels for some resins, but they ignore the effects of color and texture.
Similar problems exist with evaluating the other problems listed above. Being able to determine 'weld-line meeting angle' is useful, but basically we just want to know what the part will look like.
Enter Virtual Molding
Solid modeling and rendering have made great strides over the last several years. Realistic, textured models can be viewed on the computer and dynamically rotated and scaled. This technology has enhanced the design process by allowing non-technical people as well as designers to view a true representation of what the part is going to look like and how an assembly will fit together before it is actually produced. These models, however, represent 'perfect' samples -- no surface defects or blemishes, dimensions as designed. The next step in this technology, and the basis of Virtual Molding, is the mapping of the analysis results onto the solid model itself. For example, an area of high shear stress would show up as a discoloration on the solid model. Rather than looking at a contour plot of shear stress it would be just like holding the part in your hand and looking at the blush.
Right away this allows the designer to make a informed decision about 'how bad it is' based not on the interpretation and filtering of numeric results but on a graphical and very intuitive basis. Simulation is immediately opened up to non-technical people and training time is dramatically cut down.
The quality we are striving for is 'intuitive interpretation' -- the ability to look at the computer and understand immediately what is going on. Just as solid modeling made that was once obscure to the non-technical -- part drawings -- completely understandable, so VM will open the gates of the analysis temple.
VM does not exist right now, but many of the technical pieces of the puzzle exist. We next look at what needs to be done to achieve this goal.
Integrating the Solid Model and Finite Element Model
Solid models are a true three-dimensional model of the part. Before finite-element analysis can be performed, the model must be broken up into discrete chunks, called 'elements'. In general elements can be one-dimensional 'beams', two-dimensional 'plates', or three-dimensional 'bricks'. Brick elements are not currently used for process simulation. Instead, a wall is modeled as a 2-D plate element at the center of the wall thickness. The element is then assigned a numeric value to represent its thickness. During the analysis the element is broken up into many layers, and the temperature, shear rate, etc are evaluated on each layer. Runner and other 'beam-type' elements are modeled by 1-D beams at the center, and have an associated diameter. The simulation makes certain assumptions about the behavior of the flow. The primary assumption is that the part is 'sheet-like', where the flow length is large compared with the wall thickness.
This assumption (Hele-Shaw approximation) is made to reduce the computational and analytic burden on the simulation. It creates, however, many problems. First, turbulent flow processes (eddies and backflows) are not properly modeled. This can lead to problems when trying to predict flow lines and jetting. Second, the finite-element model can be created in several different ways from the solid model, and information can be lost.
There are two ways to deal with this problem. First, a standard algorithm can be adopted to move between the solid and mid-plane models which would allow the computer to understand the relationship between the two. Currently there are systems that can automatically generate mid-plane models from a solid, however they do not deal well with ambiguous mesh situations and cannot recover the solid from the mesh.
A better way to do it is to create an intermediate solid mesh by filling the solid model with brick elements. This is a well-defined process with much algorithmic research behind it. In this case however, it is critical that at least two opposite walls of each brick lie on outside surfaces of the part. In other words, each brick must completely fill up a wall section. Once this model is created it is a relatively simple task to create a mid-plane model by collapsing each solid element and maintaining connectivity between touching bricks by subdividing elements if necessary. Each plate element would also know which brick was its parent. When the analysis is complete the results can be mapped from the plates to the bricks and from there to the solid.
A more comprehensive method is to change the basis of the analysis to work directly with solid elements. Much research needs to be done before this is approachable. It is even unclear how much processing time would increase. However it would be a big step forward, eliminating modeling ambiguities, increasing accuracy on bulky parts, and yielding new results (such as turbulent flow behavior), and so should be the eventual goal of process simulation.
Texture, Color, and Reflectivity
In order to fulfill the promise of VM it is necessary to render parts in their actual color and texture. Technology currently exists to do this. Two dimensional textures can be wrapped around complex shapes in a process called 'texture mapping'. In order to implement this it would be necessary for a library of textures to be created, based on texture standards (the SPI standards, for example). The overall texture for a model could then be defined, along with any different textures for other surfaces. It is also important to define the reflectivity of a surface. Blush, for example, often manifests as a reflectivity change rather than a discoloration.
Visualizing Weld LInes and Blush
Much work has been done on correlating these numeric results with visual results. It should not be too much of a leap to render these onto a solid model. There are a few interesting wrinkles, however. As discussed earlier, the severity of blush is based on the shear stress. However, blush has a 'streaky' behavior that can create a starburst look around a gate. The simulation will give uniform results (more of a halo). In the interest of realism it may be advisable to have a fractal blush texture which mimics this behavior.
Visualizing Sink Marks
The data for sink marks is currently produced by the analysis ('gap thickness'). The accuracy is sometime questionable, however. (This is true with all dimensional results, like warp and shrink. The reasons for this are outside the scope of this paper) For VM purposes, rendering of sink marks presents a special challenge. This requires a surface deformation of the solid model, which can be achieved by two methods. First, the surface definition itself can be modified to create the sink. This has the advantage of producing a smoother looking model, but in some relational modelers it will create a non-associative model. Another advantage of this method is that is can be applied uniformally over the whole model, modifying all surfaces, in conjunction with the warp and shrink data.
Alternatively a boolean operation can be used to cut a dimple in the surface. This is a simpler, more direct method, but it lacks generality and will create a less smooth look.
Visualization of Flow Lines
Flow lines are created by turbulent processes, which are not modeled by the current generation of simulation tools. However, the data supplied can be used to identify areas where the likelihood of flow lines is high, and special textures can be applied to those areas. The ultimate solution for the modeling of flow lines will have to wait until true three-dimensional analysis tools become available.
Gas Traps and Burning
Current simulations assume perfect venting; that is, it is assumed that gas can escape whever it needs to. So in parts where gas traps are formed the simulation will show the traps remaining at the point of formation (where the gas is vented). In actuality the gas trap will migrate in response to the pressure gradient, and will either reach a vent, create a void, or become super-heated and create burning in the material.
Much work has been done on the migration of gas in a plastic mold in the context of gas-assist molding. The effects modeled include bleed through and burning, which would be necessary to accurately and fully model gas trap behavior. It should be a simple task to adapt this technology to a conventional mold for VM purposes.
Warp
Warp analysis is prone to error, and from a VM perspective presents special problems. Distortion of the solid model is very difficult from a mathematical standpoint. In order to implement the display of warp it would be necessary to use a deformed brick finite element mesh rather than transfer the surface analysis data all the way back to the solid.
Transparency
Evaluation of transparency is basically an extension of all the discussions above. The complexity is enhanced, however, since while with an opaque model we can concentrate on material behavior on the surface boundary, we now have to consider sub-surface defects as well. This requires lots of other data (weld-line underflow, for example) to be evaluated. Also, the rendering of a transparent object accurately on-screeen requires much more computational power (from refraction effects) and so dynamic rotation might have to be sacrificed.
Witness Lines
An extension to VM technology would be the incorporation of mold-induced defects, as opposed to purely process-related defects. The most obvious of these would be parting lines and witness lines. In order to incorporate these defects on the part, a more complete model of the mold itself would have to be included as part of the finite-element model. Fortunately this already is done to model cooling systems and high-heat transfer inserts, so the basis is there.
Obviously some assumptions would have to be made about quality of fits, but these could be inputs into the simulation. Beyond witness lines, ejector pin marks could also be modeled (probably as texture changes rather than actual indentations).
A Phased Approach to VM
Based on the above discussion we can outline a two-tiered approach to implementing VM. Phase one is based primarily on existing technology. Phase two involves the extensions to existing knowledge and techniques.
Phase I:
- Algorithmic linkage between mid-plane model and solid
- Visualization for weld lines, blush, gas traps, and burning
- Visualization of surface textures, colors and reflectivity
- Transparency effects
- Basic sink marks
Phase II:
- Solid deformation based on warp, shrink, and sink results
- Full 3-D analysis
- Flow lines and other turbulent effects
- Mold effects - witness lines, ejector pins, etc.
Conclusion
It is obvious that Virtual Molding is a natural extension to current trends in simulation and visualization technology. The objective is not to implement another 'gee-whiz' feature, but to yield major enhancements in productivity and expand the marketplace for simulation software by enabling the non-technical user. Implementing the Phase I system described above would by itself be a tremendous improvement over existing process simulation systems. Now is the time for researchers and software developers to begin moving towards this important goal.
|
|
|
