By John Rumble, Jr.
Chief, Standard Reference Data Program
National Institute of Standards and Technology
jrumble@nist.gov
Presented at the 4th Symposium on Virtual Experiment Technology for Materials design, Tokyo, Japan, October, 1998
We face a paradox todayOn the one hand, science has developed an impressive suite of machines and technology that allow physical measurements to be made better than ever before. From space-based technology that leads to peta-bytes of observations to atomic force microscopes that observe and manipulate on the atomic level, today’s measurement capability is impressive.
At the same time, faster computers, advanced telecommunications networks, better algorithms and large data collections have elevated modeling and simulation into everyday tools. We can now create a virtual world of almost any object of interest, and modern materials are no different.
The physical testing methodology for engineering materials, which has been developed over the last 150 years, has provided materials scientists and engineers with important insight on how materials behave when put into service. Understanding the materials triad - structure, processing, and properties - as determined by physical testing has been an important factor in building the amazing functionality of modern products. The questions we must now ask are: How do we create a similar testing environment for our virtual world? How can we predict the performance of virtual materials in virtual products well enough to ensure that real materials in real products perform as predicted?
This paper will explore some of the issues related to virtual testing in the 21st century. And it will try to define the role of modern data technology in support of these virtual experiments. Examples will be drawn from many corners of the materials world.
We all know what materials are. We can touch them, feel them, pound them, and bang them. They are the stuff that makes our lives physical. We will not touch on the philosophical issues that so easily arise when we try to define the physical world, but here we can accept its reality. Modern computing allows us to make models of physical realities, an activity that has existed from the earliest attempts to use mathematics to express physical quantities, for example, Pythagoras’ Theorem. However, our models today are not reductionist - that is, not just reducing complex systems to simple and easily expressed physical laws or simplistic models. Instead, we are creating constructionist models; models that are intended to have as much complexity of a real system as possible. Ultimately the goal is to be able to model every pertinent aspect of a complex system that is of interest.
High speed computers and advanced applied mathematics gives us the tools to build these models, hopefully supplemented by our insight to the physical world.The result is a new ability to create virtual reality, a state in which we can model reality not only with mathematical formulae, but also with visualization tools and predictive calculation power that allow us to interact with our virtual world similar to how we interact with the real world
Because materials are the stuff from which our real world is made, virtual materials are the stuff from which objects in the virtual world are made. What is a virtual material? How can we examine it? Clearly we should be able to look at it. We should be able to determine its structure and composition. We should be able to specify how it is made (processing).We should be able to determine its response to an external stimulus. We should be able to state its intrinsic characteristics. And we should be able to understand its interactions with its environment. All of this is a long way of saying that we should be able to understand what our virtual material is and how it performs.
Today we have some experience with virtual materials. We design polymers and composites using a variety of tools, including finite element analysis. Pharmaceuticals are being designed using computational tools. We use phase equilibrium data to predict phases under specific circumstances. Crystal structure data is being used to identify possible new crystalline substances. However, these are small steps in a long journey.
Today virtual reality tools are being developed that will allow us to do much more. Materials at all levels of structure, from the atomic level to the macroscopic and beyond to joining, have a rich array of models that describe each of these levels. The day of virtual materials is fast arriving. This does not mean that physical materials no longer are of interest in a research sense. What it means is that the world of virtual materials opens new opportunities to explore and exploit the physical world. To span possible ranges of composition, structure, processing and reaction, virtual materials can be invented, extended and tested more easily, thereby complementing physical experiments. The truly important question is how easily, and when. These issues will be discussed next.
During the past few months alone, many articles have appeared on important new materials with the potential to make substantial improvements in performance. Superbatteries [1], crystalline nanofibers of polyethylene [2], cluster and crystal structure optimization [3] and nonlinear optical and electronic materials [4] have all received considerable publicity for recent advances. The first two have resulted from extensions of traditional physical experimentation. The last two are from modeling efforts. What is perhaps most interesting is to speculate when and how all four could result from modeling efforts. For example, in the case of the new Fe (VI) superbattery, the fact that iron (VI) does exist was reported over a century ago. However, the question to be asked is from existing models today, with support from crystal structure and thermodynamic databases, could the possibility of using Fe (VI) batteries be predicted? Perhaps so, and with the advent of larger and more comprehensive databases, as well as chemical computational tools, such predictions may become "routine."
An even more interesting question to ask is to suppose such a material, in this case MFe(VI)O4, where M=K2 or Ba) was predicted by modeling, a virtual material. What kind of tests could be run on these models to see that this virtual material was a possible battery material? Such testing would involve stability testing, compatibility with various anodic materials, toxicity, and fire performance. These complex tests have been suggested deliberately because when we talk about materials testing, we usually think of simple strength tests. In fact, material testing covers a very large range of types, including static and dynamic strength, corrosion, reactivity, tribological and many more.
Creating virtual tests for these complicated physical tests is a challenge, far more than the typical finite element analysis that are routine to determine stress concentrations for composites, metals and other materials. The present state-of-the-art is far from what is needed.Some interesting features in creating virtual testing can be identified.
Generic information models for virtual tests must be developed. Because such tests generate and use large amounts of data, the common features of tests must be defined and modeled. For example, a "standardized" physical test method usually contains procedures for specimen selection and preparation; standard equipment set up, deviations from the standard equipment and procedures, data recording, analysis and reporting, among other things. What corresponds to each of these features in the virtual materials testing world? What is a suitable size specimen for testing? How do you clean a virtual surface? What is a virtual failure? These are simple examples that are sufficient to demonstrate the complexity of the problem.
Relationships among processing, structure and properties must be modeled and quantified better. Much if not most of materials design is incremental design, that is, one feature of a material is singled out for change. The test related to such materials design involves considerable subtlety, so that other features are not inadvertently changed. Consider for a moment a change in a polymer involving the diameter of nanofibers. Many processing details contribute to this diameter, as well as various material properties. A robust model of the processing involved must take into account all these factors to produce credible results. It is likely that physical theory is not sufficiently advanced to provide accurate results. In this case, no amount of virtual testing will produce even empirical results as good as physical testing can do.
The natural statistical variability of real materials must be mimiced and sampled within the space of virtual materials. Every well-designed testing program in the physical world tries to account for the variability of materials themselves, as well as the inherent uncertainties of the test method. Virtual testing must develop protocols based on solid statistical principles.
An experience-based correlation between a "standard" test and in situ service performance must be developed. Very few materials tests actually sample the full range of service conditions to which materials are subjected. Years of experience have correlated the results of tests with actual service behavior, thereby allowing designers to use materials with appropriate and economical safety factors. No such experience exists for virtual testing, and it must be developed.
Data mining techniques offer wonderful opportunities to develop new insight on the meaning of test results. As new larger and more comprehensive databases of measurements - real and virtual - are assembled, new techniques in knowledge discovery are going to find important relationships among structure, properties and processing. Already drug companies are using such tools to find new drug leads. Combinatorial techniques have provided exciting methods to explore the space of possible parameters in the physical world; parallel techniques will become prevalent in the virtual world.
Real materials are not pure. They have defects, impurities, surfaces and are in contact with other materials. Virtual testing must take place on “realistic” virtual materials. To do this will often necessitate modeling large number of particles, grains, regions of space to create virtual defects, etc. Brute numerical techniques will not suffice, and more insightful approaches will be needed.
New techniques will be needed to capture, store, manage and provide access to results of virtual testing. Even today, the results of physical testing are often imperfectly kept. Virtual testing by its very definition is a computer-based technology, and intermediate and final results must be managed in a way that preserves the information content. For example, a physical specimen can be kept for years, and even analyzed long after it was test. A virtual material last only as long as the information is stored. Even mere archiving of the results of virtual tests is not enough, because to be useful the information must be found, retrieved and processed, archived to be useful.
Standard virtual test methods must be developed and ratified. Much of the advancement in materials testing in the past century has come about from standards that codify our testing experience to promote predictability, economy of scale and good engineering practice. Virtual materials offer the possibility of a unique material for every application. Standard virtual materials test methods will support this while building public confidence in the quality of the resulting real material. National and international standards development organizations must address this issue soon.
Sets of standard test data must be developed to facilitate checking of algorithms, conformance testing and deriving benchmark test results. Similar to standard reference materials and round robin comparisons in the physical world, these sets of standard test data will allow virtual materials developers, suppliers and users to develop confidence in virtual materials.
The challenges of virtual testing of virtual materials are real. We can not even see very far in the future in terms of answers to many of the questions and issues raised in the discussion above. However, the 21st century starts with an assembly of computing tools and software ideas that will open exciting opportunities to explore new dimensions in materials and their design and production that will result in new products, energy efficiency and economic progress that can barely be imagined. Today we are virtually starting; tomorrow it will become reality.
[1] S. Licht, B. Wang, and S. Ghosh, Science 285, 1039 (1999)
[2] K. Kageyama, J.-i. Tamazawa, and T. Aida, Science 285, 2113 (1999)
[3] D. J. Wales and H. A. Scheraga, Science 285, 1368 (1999)
[4] E.K. Wilson, Chemical and Engineering News, 29 (September 27, 1999)
Top of Talk | Back to SRD Presentations | Back to SRD Homepage