Reducing Ambiguity by Learning Assembly Specific Behaviour

In this paper we present a technique for automatically generating constraints on parameter derivatives that reduce ambiguity in the behaviour prediction. Starting with a behaviour prediction using an initial library containing general domain knowledge the technique employs feedback about valid and spurious states of behaviour and knowledge about the causal dependencies between the parameters in the model in order to determine the constraints that remove the undesired states of behaviour that result from spurious ambiguity. In addition , the technique points out the assembly of physical objects to which the generated constraints apply. 1 Introduction A recurring issue in qualitative prediction of behaviour is the construction of non-ambiguous models that only predict valid states of behaviour. In particular, when using a library of model fragments that represent general domain knowledge (such as processes [Forbus, 1984] and device behaviours [de Kleer and Brown, 1984]) the ambiguity introduced by the qualitative calculus, together with the requirement of modelling device behaviour independent from the context in which it operates (the 'no function in structure' principle [de Kleer and Brown, 1984]), makes it difficult to define adequate prediction models for a specific system. In order to remove unde-sired ambiguity additional constraints must be specified that represent assembly specific behaviour, such as (1) order of magnitudes [Raiman, 1986], and (2) conservation of quantities for the system as a whole. In this paper we present a technique that automatically derives these constraints by analysing valid and spurious states of behaviour and by using a model of the underlying causality. In addition, the technique identifies the physical structure , with its specific mode of behaviour, to which the "The research described in this paper was partly supported by the Foundation for Computer Science in the Netherlands (SION) with financial support from the Nether-lands Organisation for Scientific Research (NWO) (project number: 612-322-307). constraints apply. Our approach can be thought of as supporting a knowledge engineer who, on the basis of a library containing general domain knowledge, has to develop a specific model that can be used for a behaviour prediction task. Given such a library the knowledge engineer is confronted with two problems: relating the elements from the real-world system that has to be modelled to the canonical entities present in the library, and modelling additional constraints to reduce the ambiguity in the behaviour prediction. In other words the construction of a qualitative model amounts to finding the …