TABLE OF CONTENTS Difficulties in obtaining satisfactory results from system identification are often caused by errors in setting up the problem statement. Typical errors are associated with incorrect assumptions about the units and signs of the flight data and incorrect derivation or implementation of the equations of motion in the identification software. These problems are manifested by errors between the flight-data frequency responses and those of the initial identification model. If left uncorrected, the identification algorithm will either not converge, or it will absorb these setup errors (e.g., units and signs) into incorrect values of the converged identified parameters.
The frequency-response format allows the initial model structure and parameter values to be readily evaluated and debugged. Large, consistent mismatches in magnitude and phase comparisons between the initial model and the flight data are easily seen in the individual response pairs and traced back to errors in sign convention, units, and model structure implementation. This constitutes a "sanity check" that should be completed before the optimization process is begun. Listed next are some typical problems and suggested debugging methods:
The rate responses for a flight-mechanics model should typically exhibit a -20 dB/dec roll off at higher frequencies and an additional constant phase lag of -90 deg. If they do not, check to see that the equations of motion are correct and properly implemented.
A consistent mismatch of 20 dB/dec magnitude and 90 deg phase between the flight-test response data and the initial model for the entire frequency range...
