The development of ADAPTx is the result of a very fortunate series of events over the past 30 years. In 1974, I (Dr. Wallace E. Larimore) had the good fortune to attend a series of lectures on some new methods in time series analysis and model order selection given by Dr. Hirotugu Akaike during his visit to Harvard University. Having recently completed a dissertation on time series analysis, I had the background to appreciate some of the deep insight in the concepts put forth by Dr. Akaike (see Akaike, 1973, 1974, 1975, 1976). To a large extent, my efforts over the years have been to develop more precisely these methods and to develop refined software for practical application. At the core of this technology is the Canonical Variate Analysis (CVA) method applied to system identification and time series analysis. The other crucial procedure is the Akaike Information Criterion for model selection. These developments are discussed in papers listed in the References section.
Early on, I was extensively involved in system identification using the maximum likelihood (ML) method. While the large sample theory is optimal and the computational results can be very impressive, the ML method often is computationally unstable for more than a few parameters. Even when convergence can be guaranteed, the computation time may not have a known upper bound. In contrast, the CVA method involves primarily a singular value decomposition (SVD) that is always computationally stable with a predictable amount of computation. This has, over the years, constantly pointed me toward the CVA procedure as a very intriguing idea if only it could be made to work well.
In the mid 1980's, the considerable potential of the method became apparent to others as well. As a result, I was involved in a wind tunnel demonstration of on-line adaptive control of aircraft wing flutter, an unstable interaction between the aerodynamics and structural bending of a wing (Peloubet, Haller, and Bolding, 1990). Since then, I have been fortunate enough to be a Principal Investigator on 10 Small Business Innovation Research (SBIR) contracts involving the application of the CVA technology in various fields. Some of these diverse applications are discussed elsewhere in this manual. With these various demonstrations of the achievement of automated system identification, Adaptics, Inc, was founded in 1990.
The CVA technology is a radical departure from traditional methods of system identification and modeling. I have attempted to give these ideas wide exposure at technical conferences by giving workshops on ``Automated System Identification with Control Applications'' at conferences sponsored by the IEEE Control Systems Society and by organizing invited sessions on automated system identification, the CVA method, and related topics (see References).
I am currently writing a book on Automated System Identification that will include a student edition of the software for use in undergraduate and graduate teaching. Not only are the methods fundamentally different from traditional procedures of system identification, they are conceptually much simpler and more geometric. This along with the automated aspect of the procedure will make the identification of dynamical models far more accessible to the practicing engineer, scientist, or business forecaster.
Return to Adaptics Home Page