The identification of processes is a huge field grouping very different approaches. This diversity is linked, on the one hand, on the model diversity: parametric knowledge models (transfer functions, state models) or behavior model (neuron networks, vague logic, transfer functions, state models, etc.), non-parametric models (unit responses, frequency responses, etc.), deterministic or stochastic models, and on the other hand, at the various operating contexts: online or offline, in open or closed-loop, with or without the control of input signals, etc. The objective of this chapter is to report on the major approaches adapted for the real-time parametric identification of dynamic processes, and especially, induction motors.
But, before considering the solutions, we must present the outline of the problem. The implementation of real-time identification techniques is a complex problem because of the large number of issues that need to be addressed simultaneously and coherently. First, there are problems linked to the definition of the model structure (characterization phase) requiring good expertise of the process, notably in the case of a knowledge model:
– is the model appropriate in relation to the process and the identification technique?
– would a simpler model not be as satisfactory?
– are the inevitable simplifying hypotheses that it presumes realistic and relevant for the application involved? ...