Motor parameter identification using response surface simulation and analysis

Making models match measurements: Model optimization for morphogen patterning networks

“…1B–D. Some optimization techniques are more tolerant of “smoother” cost landscapes [48], and transformations may be used to reduce the magnitudes of the peaks in the landscape. Often, a logarithmic transform is applied to the objective function values; it acts to compress error values, which may be orders of magnitude apart before transformation, to the same magnitude.…”

Mathematical Analysis for Response Surface Parameter Identification of Motor Dynamics in Electric Vehicle Propulsion Control

“…The concept of a simulated cost surface is developed here [17] as an objective function to facilitate parameter extraction of the installed drive dynamics, during offline BLMD system identification, with MSE minimization. This methodology provides useful insight into the nature and formulation of the most suitable MSE objective function to be minimized, based on actual drive experimental test data available and BLMD model simulation, coupled with an effective system identification (SI) strategy for accurate motor parameter extraction [18].…”

“…Both simulated MSE response surfaces reveal on a macro-scale the presence of a 'line minimum' of possible feasible solutions in a stationary region, enveloping a global extremum within the central surface fold, principally in the B-parameter direction. A novel mathematical approximation [17], which provides verification of the cost surface shape in both cases, is given and is used to provide information on the existence of a unique global minimum with an accompanying optimal parameter set X opt = {J¯o pt , B opt } T instead of a multiplicity of candidate options, X opt j = {J opt , B opt j } T , along a 'B -line minimum', for j = 1,2 …. Details of BLMD model simulation at a finer parameter step size δX, which illuminates the problem of a noisy cost surface, are also provided for both objective functions.…”

“…In the absence of embedded drive application details from the BLMD manufacturer [17] no precise limits on the desired accuracy of the returned J and B parameter estimates could be affixed to the parameter identification strategy for velocity controller tuning purposes in the commissioning phase. However the use of a quantized metric δX L , as mentioned previously, in parameter space puts a limit on the parameter resolution accuracy possible during identification of the BLMD dynamics in electric vehicles.…”

Mathematical Modelling and Simulation of a PWM Inverter Controlled Brushless Motor Drive System from Physical Principles for Electric Vehicle Propulsion Applications

“…The existence of a stochastic or 'noisy' cost surface, which results in a proliferation of 'false' local minima about the global minimum, is unavoidable because of model complexity and depends on the accuracy with which inverter PWM switching instants with subsequent delay turnon are resolved during model simulation . Furthermore the number of genuine local minima, besides cost function noise, is governed by the choice of data training record used as the target function in the objective function formulation which in the case of step response testing with motor current feedback is similar to a sinc function profile (Guinee et al, 2001). The cost function is, however, reduced to one of its local minima during identification, preferably in the vicinity of its global minimizer, with respect to the BLMD model parameter set to be extracted.…”