Diagnostic reasoning based on means-end models: experiences and future prospects

Multilevel Flow Models (MFM) are graphical models of goals and functions of technical systems. MFM was invented by Morten Lind at the Technical University of Denmark and several new algorithms and implementations have been contributed by the group headed by Jan Eric Larsson at Lund Institute of Technology. MFM provides a good basis for computerbased supervision and diagnosis, especially in real -time applications, were fast execution and guaranteed worstcase response times are essential. The expressive power of MFM is similar to that of rule-based expert systems, while the explicit representation of means-end knowledge and the graphical nature of the models make the knowledge engineering effort less and the execution efficiency higher than that of standard expert systems. The resulting models can be used for different diagnostic tasks, such as fault diagn osis, causal explanations, and qualitative predictions. MFM has several properties which makes for a relatively easy knowledge engineering task, compared to mathematical models as used in classical control theory and compared to the rule bases used in standard expert systems. In addition, MFM allows for diagnostic algorithms with excellent real-time properties. The paper gives an overview of existing MFM algorithms, and different MFM projects which have been performed or are currently in progress. Introduction Multilevel Flow Models (MFM) are graphical models of goals and functions of technical systems. The goals describe the purposes of a system or subsystem, and the functions describe the capabilities of the system in terms of flows of mass, energy and information. MFM also describes the relations between goals and the functions that achieve those goals, and between functions and the subgoals which provide conditions for these functions. MFM was invented by Morten Lind at the Technical University of Denmark, see Lind (1990 a). Several new algorithms and implementations were contributed by Jan Eric Larsson at Lund Institute of Technology, see Larsson (1992, 1994 a, 1996). MFM provides a good basis for diagnostic algorithms. The work of Larsson (1996) describes four algorithms based on MFM. Measurement validation checks consistency between redundant sensor values, and can discover flow leaks, sensor failures, and other measurement errors. The alarm analysis algorithm analyses any (multiple) fault situation and can tell which faults are primary and which faults that may be consequences of the primary ones. The fault diagnosis uses sensor values and queries to the operator to discover the faults of the target system. The explanation generation algorithm uses the states discovered by the fault diagnosis to produce explanations and remedies in pseudo-natural language. Other algorithms have been developed later. The failure mode analysis uses MFM with added timing information to predict the consequences of failures. It can be used both during the design phase of a plant and in real-time during actual operation, Öhman (2000 a, b). The fuzzy alarm analysis works in a way similar to the discrete alarm analysis, but is based on fuzzy logic, which makes it more robust when faced with noisy signals close to decision boundaries, see Dahlstrand (1998), Larsson and Dahlstrand (1998). MFM research is currently in a phase of maturing and further development, and this paper aims at giving an overview of both what has been done and what it is currently ongoing. It is focused mainly on the efforts of the author’s research group at the Department of Information Technology, Lund Institute of Technology, Sweden. This paper also treats knowledge engineering for MFM . While knowledge engineering in general is a difficult and time-consuming task, MFM has some properties, which makes for a relatively easier task than usual. The MFM semantics use concepts that are very abstract, high-level, and also, we believe, close to those used by human designers and operators. Our experiences are based on three projects, reported in Chapter IV, but first, a short description of MFM is given. Multilevel Flow Models MFM is a graphically represented, formal modeling language, in which the intentional properties of a technical system are described. The purposes of the system and its subsystems are modeled with goals, which can be either production goals, safety goals, or economy (or optimization) goals. The abilities of the systems are modeled with flow functions, connected into flow paths or functional networks. The main functions are sources, transports, storages, balances, barriers, and sinks, and they describe either mass or energy flows. Observers, decision makers, and actors describe information flows. The manager function describes control systems. Each flow network can be connected to one or several goals via achieve relations, which means that the functions in the network achieve the goal. A goal may be connected to one or several functions via condition relations, meaning that the goal is a condition for the function. For a full description of MFM, see Larsson (1992 c, 1994 a, 1996) or Lind (1990 a). The symbols of the MFM graphical language are shown in Figure 1.