Search

Convergence rate of the normalized least mean square adaptive estimation algorithm is analyzed. An approximation of the adaptation coefficient value which yields the best convergence speed is derived in function of the input correlation. Results concerning the trade-off : convergence speed versus asymptotic mean square estimation error are explored to substract practical rules for the adjustment of the adaptation coefficient for best possible estimation performance within a limited time delay. These results are particularly interesting for the diagnosis and adaptive control schemes which are essentially based on real-time identification.

Fly-by-wire control systems (FBWCS) introduce particular correlations between the aircraft inputs (deflection angles) and outputs which lead to biased parameter estimates in the absence of de-correlating actions such as extra-signals. This problem has been investigated with the aim of finding quickly non biased parameter values from flight data on aircraft inputs, outputs and pilot inputs, in order to provide an iterative optimization scheme with good initial values. A methodology including input design, instrumental variable estimation and automatic estimation strategy, was developed, analysed on simulated cases and tested with real flight data.

A supervisory control scheme is proposed. It provides high adaptation capabilities to time-varying conditions such as processes changes, failures or noise variations. The supervision functions are of two types : perceptionfunctions continuously performing to provide information about non-stationarities and decisiondiscontinuous functions to identify events, to evaluate their effects and to decide how to react. A procedure for implementing the general supervision strategy in the case of a simple example is presented. Simulation results are given.

A predictive control technique (PFC) is introduced where the project of the future manipulated variable depends on the dynamic set point. No error lag is obtained. The strategy is CAD compatible. Application to Industrial Robot is presented

A predictive control technique using system inversion is presented in this paper. It is a natural extension to the existing predictive control techniques but while these techniques are essentially developed for linear models, the new technique can be applied to nonlinear models as well. In this last case, the proposed technique does not require any type of local or global linearizing transformations. The most essential feature of the new technique is that it advantageously profits of both the maturity of the predictive control approach and the recent progress of the theory of system inversion particularly in the nonlinear case. Simulation results are given to illustrate the proposed technique performances.

The predictive control by inversion (PCI) technique is presented in this paper. This technique permits extending, in a simple way, the predictive control principles to non-linear models without any linearizing transformation. Singularities and non-minimum phase phenomena which can arise from inversion are treated by introducing control discontinuities either directly on the control or some of its derivatives, or indirectly through reference trajectory discontinuities. Performance and robustness properties of the technique are illustrated by several simulation results.

A study is made regarding the precision of various types of samples in estimating the parameter of an exponential population. The following cases are considered : (i) a singly or doubly truncated sample of size Nrelative to a complete sample of the same size ; (ii) a singly or doubly censored sample of size Nrelative to a complete sample of size n, where nis the number of measured observations in the censored sample, It is observed that order case (i) right truncated and doubly truncated samples are inferior to the complete sample, whereas the left truncated sample is as good as the complete sample. Under case (ii) left censored and doubly censored samples are observed to be superior to the complete sample, while the right censored sample is as good as the complete sample.

It is shown that when estimating tbe mean μof a normal population, the standard deviation σ being known. the BAN estimate of μbased on a truncated sample of size nis less efficient than that based on a complete sample of the same size whatever be the point of truncation xa. On the contrary, when estimating σ alone μbeing known, the BAN estimate of σ based on the truncated sample is substantially more efficient than that based on the complete sample whenever xa>μin the case of left truncation, and xa<μin the case of right truncation.