We propose in this paper an adaptive robust fuzzy observer to estimate flux and to identify the load torque for Induction Machine's (IM) applications. Indeed, using a fuzzy IM model, we develop the structure of an observer where we consider the load torque as an unknown input. Sufficient conditions are presented to ensure the asymptotic convergence of the flux and the load torque errors estimation. In addition, a robustness criteria is considered, using the L2 technique, to take into consideration parameters uncertainties. Design conditions are founded on LMIs terms (Linear Matrix Inequalities) and an experimentation (using DSP 1104 real time controller) is carried out to show the effectiveness of the proposed results.