This paper proposes a causal modeling framework for the purpose of process supervision. It is based on a qualitative automata (or q-automata) concept that we have devised. A q-automaton captures both the dynamics of a process variable and the expert knowledge necessary for supervising the variable's behavior. We use a two-level representation scheme for the description of the relationships between the q-automata underlying a process: a local constraint level and a global constraint one. The local constraint level describes the qualitative causal relationships between the q-automata, and the global constraint level states the quantitative constraints among them. The formalism is shown to allow the modeling of deep knowledge as well as compiled knowledge. Furthermore, it is suitable for the modeling of partially-known, hybrid (numeric and symbolic), continuous and discrete processes. A causal engine (CA-EN) using the formalism is under intensive development. It is at the core of the process supervision system.