Magnitude sensitivity analysis for parameter identification applied to an autonomous underwater vehicle

A novel “identifiability” technique, named magnitude sensitivity analysis, is presented to determine identifiable parameters within a dataset. Parameters with high sensitivity can be successfully identified, whereas parameters with low sensitivity can be omitted, reducing system complexity. The technique is less computationally intensive compared to other published methods, preserves the physical nature of parameters, and is applicable in real-time analysis. Verification of the technique was tested for a simulated mass–spring–damper model with a step and sinusoidal forcing inputs to demonstrate the determination of identifiable parameters from different datasets. For the two forcing scenarios, magnitude sensitivity analysis predicted parameters with high and low sensitivity, which were then estimated using an extended Kalman filter. High sensitivity parameters yielded values with errors as low as 0.4%, whereas low sensitivity parameters yielded values with errors up to 533%. The technique was then applied to experimental data measured from an autonomous underwater vehicle (AUV) undergoing pitch maneuvers. The magnitude sensitivity analysis was used to reduce the nonlinear system model governing the AUV before proceeding with the estimation of the high sensitivity parameters. Results conclude that the estimation of the high sensitivity parameter deviated by 4% compared to the initial guess parameter from a numerical simulation.