Deep Reasoning Research Articles

A second generation expert system for fault diagnosis

We propose a second generation expert system for fault diagnosis in a power system. This expert system combines shallow reasoning with deep reasoning that is based on a model reference approach. The model used for qualitative simulation of the power system in the deep reasoning part is based on a timed Petri net approach. The timed Petri net approach allows for accurate qualitative simulation of the protection system and power system components. The Petri net model permits one to model the function and behaviour of the power system components. It therefore allows for both static and dynamic simulation of the protection system and the associated power system. An application showing the modelling of part of substation configuration is given.

Some causal models are deeper than others

The effort within AI to improve the robustness of expert systems has led to increasing interest in ‘deep’ reasoning, which is representing and reasoning about the knowledge that underlies the ‘shallow’ knowledge of traditional expert systems. One view is that deep reasoning is equivalent to causal reasoning. By analyzing the causal reasoning of a particular medical AI system, we show that this view is naive. Specifically, we show that causal networks omit information relating structure and behavior, and that this information is needed for deeper reasoning. Our conclusion is that deepness is relative to the phenomena of interest, i.e. whether the representation describes the properties and relationships that mediate interactions among the phenomena and whether the reasoning method takes this information into account.

Integrating diagnostic knowledge

The use of artificial intelligence technique to access, analyze, and integrate different types of knowledge under a single diagnostic concept is described. Repair statistics and field experience are handled by an empirical knowledge (shallow reasoning) diagnostic system in order to retain the experience of expert test personnel. Computer-aided-design knowledge is handled by model-based (deep reasoning) diagnostic systems in order to extract diagnostics directly from design data. Combining these approaches overcomes limitations of the individual techniques and provides a more powerful diagnostic system. The Westinghouse expert diagnostic system is considered as an example.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

AUGMENTING EXPERIENCE-BASED DIAGNOSIS WITH CAUSAL REASONING

This paper describes an attempt to augment heuristic reasoning with deeper reasoning. The context of the work is the DIVA Expert System that has been designed to diagnose faults on rotating machinery. DIVA is essentially based on heuristic knowledge. We will describe how it should be possible to enhance its reasoning capabilities by enabling DIVA to reason on the causality of involved phenomena.

INTEGRATING CLASSIFICATION-BASED COMPILED LEVEL REASONING WITH FUNCTION-BASED DEEP LEVEL REASONING

Problem solving based on compiled associations between elements of the decision space and data is an efficient mode of reasoning for a large percentage of situations faced by an expert. But in some (usually small) percentage of cases, compiled associations are not enough by themselves to lead to correct results. Reasoning from “deeper” levels of understanding offers the advantage of producing correct results even in atypical cases, but at the cost of expanding more computational resources. Thus the trade-off between compiled level systems and deep level systems is between computational efficiency (at the compiled level) and problem-solving generality (at the deep level). We describe a hybrid system containing elements of both deep level reasoning and compiled level reasoning. More particularly, we propose a problem-solving architecture for category-based diagnostic problem solving which at the compiled level centers on classification problem solving and at the deep level uses a type of function-based reas...

Topology-based geometry representation to support geometric reasoning

The geometry of parts and components is traditionally represented in the computer using some form of coordinate data in three-dimensional Euclidean space. Such a representation is at too low a level of detail to enable deep reasoning based on an understanding of the geometric relationships that exist in the part. It is suggested that an application-specific hierarchic representation of part geometry, where geometric details at lower levels of the hierarchy are embedded within the higher level topological description of the object, provides a powerful framework for enabling such geometric reasoning capabilities. Such a representation can facilitate high-level topological queries, linguistic means for human-machine interaction, development of an intelligent interactive design aid, and an extension to more complex geometries. It can provide a language for expressing the appropriate domain heuristics, thereby leading to a smooth interface between the heuristics and the geometry representation. The authors analyze these issues and illustrate the concepts in the context of blocker design, which is encountered in the process of forging die design.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Architectural Issues for Expert Systems in Real-Time Control

There is an increasing interest in using expert system technology in the field of automatic control. We show that some particular requirements are imposed on the functionality and architecture of expert systems which shall support or extend conventional control systems. We discuss the necessity and the impacts of deep modeling, temporal reasoning and reason maintenance techniques from an architectural point of view. We argue for the utility of a separate mediating module, which tries to reduce the gap between the different knowledge levels of a conventional control system and an expert system. Furthermore, we show that it is highly valuable to have a distributed architecture such that the expert system and the mediating module consist of a set of parallel processes. The ideas developed in this paper are based on three years of experience in the German national venture TEX-I.

Architectural issues for expert systems in real-time control

Abstract There is an increasing interest in using expert system technology in the field of automatic control. We show that some particular requirements are imposed on the functionality and architecture of expert systems which shall support or extend conventional control systems. We discuss the necessity and the impacts of deep modeling, temporal reasoning and reason maintenance techniques from an architectural point of view. We argue for the utility of a separate mediating module, which tries to reduce the gap between the different knowledge levels of a conventional control system and an expert system. Furthermore, we show that it is highly valuable to have a distributed architecture such that the expert system and the mediating module consist of a set of parallel processes. The ideas developed in this paper are based on three years of experience in the German national venture TEX-I.

Deep-reasoning fault diagnosis: an aid and a model

The design and evaluation are presented for the knowledge-based assistance of a human operator who must diagnose a novel fault in a dynamic, physical system. A computer aid based on a qualitative model of the system was built to help the operators overcome some of their cognitive limitations. This aid differs from most expert systems in that it operates at several levels of interaction that are believed to be more suitable for deep reasoning. Four aiding approaches, each of which provided unique information to the operator, were evaluated. The aiding features were designed to help the human's casual reasoning about the system in predicting normal system behavior (N aiding), integrating observations into actual system behavior (O aiding), finding discrepancies between the two (O-N aiding), or finding discrepancies between observed behavior and hypothetical behavior (O-HN aiding). Human diagnostic performance was found to improve by almost a factor of two with O aiding and O-N aiding.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Second generation expert systems

Second generation expert systems are able to combine heuristic reasoning based on rules, with deep reasoning based on a model of the problem domain. This solves a number of major problems of current expert systems, most importantly the problem of knowledge acquisition: second generation expert systems can learn new rules by examining the results of deep reasoning. The paper outlines the components of second generation expert systems and gives an example.

Correctness of parallel programs: The Church-Rosser approach

For many purposes, asynchronous parallel programs may be viewed as sequential but nondeterministic programs. The direct translation to nondeterministic sequential form leads to a combinatorial explosion of program size before correctness proofs can even begin. The Church-Rosser approach to correctness of asynchronous parallel programs is a flexible way to divide a correctness proof into several lemmas, no one of which requires both deep reasoning and explicit enumeration of all the control states required in the nondeterministic sequential form of the program. The approach is stated and justified abstractly, demonstrated in detail for a simple example program, and compared with other approaches to the correctness of parallel programs. The abstract formulation is independent of the model of parallelism in the example and can also be applied to nondeterminism not derived from asynchronous parallelism. We conclude with a survey of prospects for computer assisted proofs structured by the Church-Rosser approach.