Under a governmental grant, the LAAS is involved with an industrial partner (“Société Vieille Montagne”) in the analysis, modelization and optimization of a zinc production plant, essentially the residue hydrometallurgical treatment part of the global plant.This part of the process is in itself a complex system with different interconnected subsystems, a cascade of 5 reactors and strong non linear phenomena.The first part of the paper discusses the analysis of the interaction in this kind of process, dealing with measurements currently used for the management of the plant.This analysis is based on a comparison between classical statistical tools (requiring often the hypothesis of linearity for the relationships between the variables) and informational techniques using entropy concepts (which do not require the hypothesis of linearity) .The simultaneous use of these two approaches allows detection, in one hand, of wrong measurements and fault sensors, and in the other hand, recognition of non linearities in the process.In the second part of the paper, the non linear relationship detected between the variables are modelized with the aid of chemical considerations. The corresponding models are validated using available data on the process.After the definition of a performance index, an optimization problem is written in order to find the best steady state point with respect to the chosen criterion.The problem is solved with dynamic programming technique.In conclusion, some extensions toward a dynamic analysis of the process and control of its different parts are addressed.