# A Resampling Based Approach to Optimal Experimental Design for Computer Analysis of a Complex System

## Abstract

The investigation of a complex system is often performed using computer generated response data supplemented by system and component test results where possible. Analysts rely on an efficient use of limited experimental resources to test the physical system, evaluate the models and to assure (to the extent possible) that the models accurately simulate the system order investigation. The general problem considered here is one where only a restricted number of system simulations (or physical tests) can be performed to provide additional data necessary to accomplish the project objectives. The levels of variables used for defining input scenarios, for setting system parameters and for initializing other experimental options must be selected in an efficient way. The use of computer algorithms to support experimental design in complex problems has been a topic of recent research in the areas of statistics and engineering. This paper describes a resampling based approach to form dating this design. An example is provided illustrating in two dimensions how the algorithm works and indicating its potential on larger problems. The results show that the proposed approach has characteristics desirable of an algorithmic approach on the simple examples. Further experimentation is needed to evaluate its performance on larger problems.

- Authors:

- Publication Date:

- Research Org.:
- Sandia National Labs., Albuquerque, NM (US); Sandia National Labs., Livermore, CA (US)

- Sponsoring Org.:
- US Department of Energy (US)

- OSTI Identifier:
- 9503

- Report Number(s):
- SAND99-0363C

TRN: AH200124%%297

- DOE Contract Number:
- AC04-94AL85000

- Resource Type:
- Conference

- Resource Relation:
- Conference: 1999 Joint Statistical Meetings, Baltimore, MD (US), 08/08/1999--08/12/1999; Other Information: PBD: 4 Aug 1999

- Country of Publication:
- United States

- Language:
- English

- Subject:
- 99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; ALGORITHMS; COMPUTER-AIDED DESIGN; PERFORMANCE; STATISTICS; MECHANICAL STRUCTURES; COMPUTERIZED SIMULATION; EXPERIMENTAL DESIGN; LARGE-SCALE SIMULATION; UNCERTAINTY ANALYSIS

### Citation Formats

```
Rutherford, Brian.
```*A Resampling Based Approach to Optimal Experimental Design for Computer Analysis of a Complex System*. United States: N. p., 1999.
Web.

```
Rutherford, Brian.
```*A Resampling Based Approach to Optimal Experimental Design for Computer Analysis of a Complex System*. United States.

```
Rutherford, Brian. Wed .
"A Resampling Based Approach to Optimal Experimental Design for Computer Analysis of a Complex System". United States. https://www.osti.gov/servlets/purl/9503.
```

```
@article{osti_9503,
```

title = {A Resampling Based Approach to Optimal Experimental Design for Computer Analysis of a Complex System},

author = {Rutherford, Brian},

abstractNote = {The investigation of a complex system is often performed using computer generated response data supplemented by system and component test results where possible. Analysts rely on an efficient use of limited experimental resources to test the physical system, evaluate the models and to assure (to the extent possible) that the models accurately simulate the system order investigation. The general problem considered here is one where only a restricted number of system simulations (or physical tests) can be performed to provide additional data necessary to accomplish the project objectives. The levels of variables used for defining input scenarios, for setting system parameters and for initializing other experimental options must be selected in an efficient way. The use of computer algorithms to support experimental design in complex problems has been a topic of recent research in the areas of statistics and engineering. This paper describes a resampling based approach to form dating this design. An example is provided illustrating in two dimensions how the algorithm works and indicating its potential on larger problems. The results show that the proposed approach has characteristics desirable of an algorithmic approach on the simple examples. Further experimentation is needed to evaluate its performance on larger problems.},

doi = {},

journal = {},

number = ,

volume = ,

place = {United States},

year = {1999},

month = {8}

}