Statistical Software Package Heats Up Cool Research

Solutions
 Benefits
  • HVAC
  • Simulation modeling
When researchers at Johnson Controls of Milwaukee wanted to check their theories on more efficient ways to heat and cool large buildings, the Fortune 500 manufacturer of heating, ventilation and air condition (HVAC) systems had two choices. Johnson Controls could build a model equivalent of a small town, tear it down and rebuild it using trial and error -- or researchers could recreate those scenarios on computers and monitor the results. All they needed was the right software.

"We're a controls company, and we build automation systems for controlling large buildings' heating systems," explained Jim Braun, a Johnson senior research engineer.

The company chose the IMSL® Library, made by Visual Numerics. The software library consists of three major Fortran programs: Math/Library for mathematical applications, Stat/Library for statistical analysis and Sfun Library for special function evaluation.

"I use simulation models to model buildings and the control systems in them," explained Braun, who works for Johnson Controls' systems and controls group.

"Other people here are involved in modeling thermal systems and mechanical systems," he said. "So what we use the library for is to solve some of the mathematical problems that arise in particular solutions of nonlinear equations, regression analysis in which you do model development and curve fitting of the parameters of the model."

"At present," Braun said, "we are using IMSL to look for a more efficient approach to modeling heating and cooling systems."

Regression is the mathematical technique used to determine the parameters of the model that best match the data. Using regression analysis, a researcher will attempt to match a theoretical model to measured data. In doing so, the scientist tries to tune the model to match the real characteristics of the system.

"Let's say we're interested in monitoring the performance of [cooling] systems," Braun explained. "In order to monitor the system and evaluate [it] to see if it's performing properly, we might have a model of how the system should perform."

"By comparing the real measurements," he continued, "the monitored measurements versus the model, we can determine whether the system is working properly. The regression could come in initially building the model. You've got a set of measurements for normal operations of the equipment. You want to match your model to that normal performance to determine the model parameters."

Portable Programs a Priority

"The IMSL Fortran package contains an estimated 800 algorithms, one of the package's main strengths," Braun said. "Not only do the algorithms allow researchers to forgo the often arduous task of writing their own, but the IMSL algorithms are accepted as a standard by numerous major government agencies and departments," he added.

Braun said he chose IMSL because he needed software that would allow him to test his theoretical models against empirical evidence and do it easily and inexpensively enough to justify the investment. The research engineer had used just such a package at the University of Wisconsin while he worked toward his doctorate in mechanical engineering, he said. That experience -- plus a few other considerations -- made the decision an easy one.

"Though alternatives to the IMSL Library exist, none has proved as effective," he said.

"I have some of my own routines," he noted. "We also have Minitab [a similar software package that does not contain as many formulas as IMSL], from Argonne National Laboratories, although it's much less complete. It doesn't do all the things that IMSL can do for us, such as optimization and statistical analysis."

As any corporate financial planner will attest, it's a lot easier to choose a software package when the program in question is already an institution. And according to Braun, what LotusTM is to microcomputer spreadsheets, Visual Numeric's trio of FORTRAN subroutine libraries apparently is to statistical analysis on minicomputers.

"I was aware that it is the standard mathematical library, which means the programs that I write are pretty portable. I can send them off to somebody else, who will have access to that same library. Practically everyone working in the scientific engineering field who does simulation modeling is familiar with that package, or is at least aware that it exists or has used it.

"Programs that I write I could send to them," Braun said. "I could send them a disk and there's a good chance that they would have access to that same package to use the routines that I wrote. So there's that portability. That standard nature of the program is very important."