Mineral processing and extractive metallurgy plants consist of a large amount of processing units interconnected in a complex network of streams. Measurements of species mass flowrates and temperatures are routinely made for process monitoring, control and optimization. In such an environment, strategies for fault detection and isolation (FDI) could significantly improve the results of further steps of data processing.; The thesis first proposes a model-based method which is able to detect faults in the process model as well as in the measurements. The model consists of stationary mass conservation equations, allowing the calculations of the covariance matrix of the model uncertainties. The method is illustrated for an ore flotation circuit. Then, the thesis proposes three model-based FDI methods for processes having bilinear constraints. The models involved in both methods consist of stationary mass and energy conservation equations. The methods are illustrated for a simulated iron oxyde pellet sintering furnace. Then, a method for performing FDI during process transition is developed. The method is compared to the widely accepted Chow-Willsky scheme, and shows advantages on the robustness point of view. Finally, the last chapter discusses detectability and isolability problems in the multiple fault case. |