Wojciech Cholewa

Identification of sensitivities in Bayesian networks

This paper presents a methodology for sensitivity analysis that can be applied to Bayesian belief networks, i.e. analysis of the influence of the quality of network parameters (such as conditional and a priori probabilities) on the values of the hypothesis variable(s). The presented methodology makes use of one-way sensitivity analysis and makes it possible to apply a particular mathematical model for relations between the considered parameter and distribution of values in the node of interest (hypothesis node). The sensitivity analysis has been applied to a network describing a Nuclear Power Plant during fault conditions.

Mechanical Analogy of Statement Networks

Mechanical Analogy of Statement Networks The paper demonstrates briefly the reasoning capabilities in condition monitoring offered by systems based on statement networks. The usefulness of the networks considered results among others from possibilities of their optimization related to the minimization of contradictions between rules acquired from different knowledge sources. A mechanical analogy of such networks introduces an interpretation of statements as material points that are able to move. Dependencies between statements are considered as approximate necessary and approximate sufficient conditions, which are represented by unilateral constraints imposed on the introduced material points. A model of a dynamic statement network can be obtained out of the network consisting of statements represented by material points with assigned masses, where the inertia of statements may be taken into account. The paper introduces a measure of conditional contradictions of statements, which can be used for monitoring knowledge bases in running expert systems.

Expert Systems in Technical Diagnostics

Modern measurement technology makes it possible to continuously observe and record signals connected with the courses of technological processes, and machinery or devices which take part in these processes. Most often, the signals are supplied to modules which analyse them in order to estimate a set of their features forming the symptoms of the present technical state of the observed object. A particular property of the problems of technical diagnostics is that they are usually related to objects (e.g., machines) of different constructions. It requires a distinction between the forms of databases, and specialisation of rule sets applied within an inference process dealing with the technical state of the object. Additional difficulty is that the history of changes occurring in the observed objects (e.g., modernization) and the history of their maintenance (e.g., repair and control) must be recorded and taken into account in the inference process. The need for applying monitoring and diagnosing devices to complex technical objects is the main reason for research whose goal is to find proper tools for aiding the processes of the design and maintenance of such devices. Interpreting the results of signal analysis is a difficult task. It always requires some kind of experience regardless of the fact that the diagnosing is based on an exhaustive model of the object or on diagnostic rules which are considered to be valid for a determined class of machinery.

Development Environment for Diagnostic Multimodal Statement Networks

Development of effective diagnostic systems for the recognition of technical conditions of complex objects or processes requires the use of knowledge from multiple sources. Gathering of diagnostic knowledge acquired from diagnostic experiments as well as independent experts in the form of an information system database is one of the most important stages in the process of designing diagnostic systems. The task can be supported through suitable modeling activities and diagnostic knowledge management. Briefly, this paper presents an example of an application of multimodal diagnostic statement networks for the purpose of knowledge representation. Multimodal statement networks allow for approximate diagnostic reasoning based on a knowledge that is imprecise or even contradictory in part. The authors also describe the software environment REx for the development and testing of multimodal statement networks. The environment is a system for integrating knowledge from various sources and from independent domain experts in particular.

Identification of loss-of-coolant accidents in LWRs by inverse models

Abstract This paper describes a novel diagnostic method based on inverse models that could be applied to identification of transients and accidents in nuclear power plants. In particular, it is shown that such models could be successfully applied to identification of loss-of-coolant accidents (LOCAs). This is demonstrated for LOCA scenarios for a boiling water reactor. Two classes of inverse models are discussed: local models valid only in a selected neighborhood of an unknown element in the data set, representing a state of a considered object, and global models, in the form of partially unilateral models, valid over the whole learning data set. An interesting and useful property of local inverse models is that they can be considered as example-based models, i.e., models that are spanned on particular sets of pattern data. It is concluded that the optimal diagnostic method should combine the advantages of both models, i.e., the high quality of results obtained from a local inverse model and the information about the confidence interval for the expected output provided by a partially unilateral model.

Dynamical Statement Networks

This paper addresses dynamics in diagnostic expert systems. The introduced dynamic statement network may be used not only in static diagnosing systems but also in dynamic monitoring systems. Within these networks, statements consist of contents and values. Statement values are based on a concept introduced in intuitionistic fuzzy sets, i.e. they contain independent belief about validity and invalidity of information presented by statement contents. The relationships between statements are modelled in the form of a set of necessary as well as sufficient conditions. The author compared requirements placed against static and dynamic statement networks and devised a manner in which a static network may be transformed into a dynamic one which, in turn, facilitates non-monotonic reasoning required for monitoring systems.

Gradual Forgetting Operator in Intuitionistic Statement Networks

The investigated intuitionistic statement networks facilitate designing and development of complex expert systems. Within these networks, statements consist of contents and values. Statement values are based on a concept introduced in intuitionistic fuzzy sets, i.e. they contain independent belief about validity and nonvalidity of information presented by statement contents. The relationships between statements are modeled in the form of a set of necessary as well as sufficient conditions, and these conditions are considered as corresponding inequalities between statement values. The paper introduces gradual forgetting operator of statement values. The networks with such an operator may be used as models of dynamic objects. They allow for non-monotonic reasoning required for monitoring and diagnostic systems.

Intuitionistic Notice Boards for Expert Systems

This paper presents a concept of an intuitionistic notice board. The board consists of notices including statements. Notices may be considered as variables of intuitionistic fuzzy logic. It was assumed that the notices in question are nodes of a network whose links represent necessary and sufficient conditions occurring between the nodes. These conditions are written down as a system of inequalities between the values of intuitionistic variables. Furthermore, a manner in which approximate solutions in a network of notices are determined was introduced. The main benefits to arise from the use of notice boards may include a possibility for a knowledge model compilation from independently developed individual submodels that, subsequently, can be easily consolidated to form a general model. In addition, another advantage of the proposed approach allows for both consistency verification of the designed model, and monitoring of potential contradictory cases that may occur in the model during its operation.

Local Pattern-Based Interval Models

This paper aims to inform about special diagnostic models designed for reasoning about causes of observed symptoms, e.g. a state of an object. The paper introduces the general idea of inverse models produced by numerical inversion of simulation results, where the simulation is being run by known cause-effect models transforming state features into diagnostic symptoms. Special attention has been drawn to a strict and interval modelling as well as to global and local models. Suggested methodology is focused on the identification of local models such as pattern based models, i.e. models spanned on particular sets of selected data. Although the presented approach is addressed to diagnostics it may be easily extended to other applications.

Methods of Signal Analysis

Information obtained as a result of observing objects that are subjects of a diagnostic process is a basis for inference in technical diagnostics. Depending on the kind of diagnostic observations considered, information can deal with physical quantities that are connected with the object operation (e.g., the flow intensity of a medium in a suction connector). Information can be also related to residual processes that are effects of each object operation (e.g., the level of acoustic emission during the operation of a cutting tool). To generalise the numerous kinds of observations, one can consider a signal (diagnostic signal) as a material carrier that makes it possible to transmit information about the observed object or process. In most cases this carrier is a set of any quantities. The application of information included in the signal requires its appropriate description. Signals may be effectively described by sets of values of features that are results of signal analysis. Examples of features of a stochastic signal can be the estimates of that process.