The supervision and monitoring systems have a major role to the security of an industrial plant and the availability of its equipments. Forewarn the operator earliest about the deviations of the process nominal behaviour is fundamental to carry out preventive and corrective actions. Some kind of industrial plants like the chemical process and batch systems set significant complexities to the monitoring and supervision systems because of their hybrid nature (discrete and continuous interactions), the number of active variables and the complexity of theirs behaviour relations. This complexity becomes more pronounced in systems characterized by numerous operating mode changes leading to a numerous transitory phases.The monitoring of these transitory phases is a delicate issue. The large amount of variables to be taken into account leads to a difficult reasoning and interpretation of the process behaviour. In case of fault, the diagnosis becomes then a complex task. The marginal deviations of process behaviour may indicate a dysfunction that degenerates slowly or be caused by operator misbehaviour or piloting system fault.To cope with this complexity we propose a hierarchical monitoring and control system completed by a procedural decomposition proposed by the ISA88 norm, defining from the upper level the recipes, procedures, operations, phases and tasks. Our monitoring and diagnosis approach is located at two high levels of this procedural hierarchy: i) at the operation level, particularly on the transition period of operating modes where the influence relations between the variables are weekly or not known, ii) at the phase level where the influence relations are known in a period of time belonging the transition operating mode horizon. Concerning the complexity of the behaviour relations between the variables during the transitory phases, it is neither always possible nor necessary to have a precise mathematical representation of the process dynamics though algebraic-differential equations. Thus, the qualitative approaches become an interesting solution. They allow the representation of the process behaviour with an abstraction degree that offers more robustness and a better suitability with the considered monitoring high level.
