The main goal of this paper is twofold: To show a kind of robustness in a nonlinear multivariable (NL MIMO) system with feedback control, by employing some linearization and observation techniques well known in reference kind of model; and at the same time it is a proposal to immunize the measurements of a level capacitance-based sensor when there are changes in the measured variables properties, by employing an on-line compensation. The MIMO thermal-hydraulic nonlinear system is stabilized when some linearizing conditions are met and a design methodology for using a state feedback scheme is established. In order to assign poles to the control system, the Jacobian linearization is transformed to the controllable canonical form. This is accomplished by the Brunovsky similarity transformation, and by a static state feedback. A Luenberger observer is added to the control system in order to improve its stability. Illustrations of these techniques are shown via numerical and practical simulations carried out directly in the inexpensive physical process.