Recent Research: Greg Miller
Chances are this is not as up-to-date as it should be, but it provides a brief overview of some of the research I have been involved with in recent years.last modified 10/2007
Live Modeling in Structures and Mechanics
Live modeling can be defined as a computational approach in which one interacts with an analytical model directly via controls or mouse gestures and observes the results of these actions immediately in an animated, quasi-real time fashion. In effect, the system being modeled appears "live" in the sense that it responds immediately and realistically to input from its environment. Some of the motivations for pursuing live modeling include the following:
- The usefulness of analysis results depends on the speed and convenience with which they can be obtained, interpreted, and understood.
- Reasonable, rapidly-obtained results have always played a central role in engineering practice, and this can be expected to continue being the case.
- Human brains tend to work well with temporal and spatial proximity for purposes of making comparisons and considering alternatives.
- People generally tend to learn and understand better when they can interact with and observe phenomena directly.
Building such modeling environments requires attention to both under-the-hood numerical performance and interface design and development issues, and the overall approach encourages reconsideration of many aspects of broader computational modeling tasks in design and research contexts. We have been working in each of these areas for many years; the following summarizes a representative sample of these activities.
Numerical Algorithms
The feasibility of live modeling for a given class of problems clearly depends on the availability of numerical methods and implementations that enable sub-second solution times. We have investigated numerous direct and iterative solver technologies for use in iteractive contexts, and have developed a family of tensor-based methods that are competitive with existing high-performance solvers, but which also support coordinate-free modeling abstractions. Our recent work characterizing the live modeling boundaries for single CPU machines has indicated that depending on the context (element type, problem topology, etc.), one can generate sub-second solutions for problems on the order of 30,000-60,000 DOFs.
Example Applications
We have developed a broad set of live modeling applications in various contexts. Examples that have evolved beyond the prototype stage include the following:
| Program | Description | Availability |
| Dr. Layer (w/ P. Arduino) | Models wave propagation in layered media. Includes linear and nonlinear material response, and time and frequency domain visualization. | Free for Mac and Windows: link |
| Dr. Shakes (w/ S.Cooper) | Interactive modal analysis visualization for simple structural systems. | Free for Mac and Windows: link |
| Dr. Quack (w/ S.Cooper) | Moment and shear diagram quizzer. | Free for Mac and Windows: link |
| Dr. Beam Pro (w/ S. Cooper) | Live modeling of beams. | Commercial: link |
| Dr. Frame2D/Dr.Frame 3D (w/ M. Rucki) | Live modeling of 2D and 3D structural systems. | Commercial: link |
| BeamVisualizer (w/ S. Cooper) | Real-time, interactive beam analysis | Distributed by Prentice-Hall (One-key) |
| StressVisualizer (w/ S. Cooper) | 3D Stress state visualization. | Distributed by Prentice-Hall (One-key) |
| KinematicsVisualizer | 3D visualization of kinematicsinvolving points and general reference frames. | Distributed by Prentice-Hall (One-key) |
| StaticsVisualizer | Visualization and interaction of statics systems. | Distributed by Prentice-Hall (One-key) |
Integrated Modeling, Analysis, and Authoring Environments for Structural/Mechanical Engineering Education
Historically, there have been many examples of tools that support faculty instructional authorship and communication being adopted readily once the barrier to entry has fallen low enough; word processors, course web sites, and presentation software being three ready examples. However, the blessing and the curse of such generic tools is that they do not allow one to work within an environment that is discipline-specific or discipline-cognizant. Materials must be captured (frozen, in a sense) and imported from other discipline-specific sources, and therefore dynamic interaction with the resulting materials is limited, if available at all. Tools that are designed to support interactive content also typically work by requiring external content creation and capture. In this work we have taken an opposite tack---rather than bringing discipline-specific material into generic presentation environments, we have developed a model in which generic authoring capabilities are brought into a discipline-specific, rich simulation environment, thereby allowing instructors to bring their personal authorship abilities to a whole new class of activities. In particular, we have used an advanced modeling tool with unique visualization capabilities coupled with presentation/scripting technology to enable the integration of guided explanation and conceptual development with unguided, student-centered exploration and experimentation.
Development of a Solid/Fluid Two-Field Material Point Method for Modeling Saturated Granular Materials
w/ P. Arduino and P. Mackenzie
Catastrophic infrastructure failures frequently are caused by the dynamic interaction of soil and water. Earthquake induced liquefaction, high-water induced levee failures, bridge scour, and rainstorm-induced slides are all examples of such phenomena involving various time scales. In many cases, engineering techniques for predicting onset of failures are available and widely used in practice, but there is a dearth of tools available for more general consideration of post-failure behavior. Modeling behavior up to and beyond the point of failure is complicated by the fact that the overall material response involves complex interactions between solid and liquid phases and transition from solid behavior to sediment transport. Frameworks designed for use in either solid-only or fluid-only contexts are not capable of tracking the significant phenomena throughout a given event history. This proposal focuses on the development and evaluation of a computational approach specifically designed to address this class of problems. In particular, the objective of the work is to develop a proof-of-concept formulation and implementation of the Material Point Method that explicitly includes solid and liquid modeling, and which can track liquid/solid transitions in behavior.
Rapidly-Deployable Secondary Support Systems for Collapse Prevention
with J. Berman, P. Mackenzie, I. Jensson, and G. Jankhah
The vast majority of structural earthquake engineering research and development has focused on understanding and improving the behavior and performance of structures during seismic events. However, in many parts of the world the majority of the structures where people work, live, and travel are not state of the art, nor are they being upgraded at a particularly rapid rate. Also, any structure is subject to collapse given a suitably damaging event, and this is especially so given the inherent uncertainty associated with earthquake hazards. The inevitable occurrence of structural collapse during significant seismic events is thus unavoidable in practical terms. This motivates the work presented in this paper, which is based on the same principle leading to the inclusion of lifeboats on ships, seat belts in cars, and sprinkler systems in buildings: even well-engineered systems should have back-up safety systems to handle catastrophic events. The goal is to provide alternatives to the default practice of dealing with collapse by trying to rescue people from the rubble after the fact, and then determining how to design a better structure the next time.
The overall strategy is to develop secondary structural shoring systems that can be rapidly deployed to localize damage to prevent progressive collapse, or to protect occupants directly or indirectly, or both. The particular approach being considered presently is based on the use of large-scale airbag technology, leveraging the fact that relatively low pressures can carry large loads given large contact areas, and exploiting the inherent efficiency of self-reacting tension structures.
Model Management
w/ L. Lowes, A. Lindblad
Current methods of analyzing multiple design configurations can be labor intensive, and typically require engineers to make relatively indirect comparisons between structural models. We have developed a prototype modeling system designed to support convenient management and analysis of multiple design configurations, and to provide a high degree of spatial and temporal congruence for model visualization, thereby allowing more direct comparisons to be made. This ultimately can allow for a more thorough evaluation of design alternatives, while decreasing the workload needed to perform these analyses.
Similarly, in the case of nonlinear analysis, the inherent complexity of these models and the lack of efficient model-building and model-management tools impede the use of state-of-the-art simulation to support research and design. As a result, nonlinear simulation is employed only occasionally by practicing engineers to improve their designs. We are interested in developing model-building frameworks to manage nonlinear analyses in such a way as to enhance the efficiency of the model building and validation processes, thereby expanding the range of projects for which advanced analysis is viable.
Next Generation Capture and Communication of Structural Experimentation Data
with D. Currit and G. Farrar
Experimental data collected at structural testing facilities are growing increasingly complex. This is due both to increases in the quantity of collected data and new developments in the types of instrumentation available. This project has been developing mechanisms for managing complex experimental sets of data to enhance the ways in which the information can be visualized, interepreted, and communicated. Our approach is built around the idea of using an analytical structural modeling environment as the starting point for viewing experimental set-ups and data.