Neural Network Logic Research Articles

NEURAL LOGIC NETWORK LEARNING USING GENETIC PROGRAMMING

Neural Logic Networks or Neulonets are hybrids of neural networks and expert systems capable of representing complex human logic in decision making. Each neulonet is composed of rudimentary net rules which themselves depict a wide variety of fundamental human logic rules. An early methodology employed in neulonet learning for pattern classification involved weight adjustments during back-propagation training which ultimately rendered the net rules incomprehensible. A new technique is now developed that allows the neulonet to learn by composing the net rules using genetic programming without the need to impose weight modifications, thereby maintaining the inherent logic of the net rules. Experimental results are presented to illustrate this new and exciting capability in capturing human decision logic from examples. The extraction and analysis of human logic net rules from an evolved neulonet will be discussed. These extracted net rules will be shown to provide an alternate perspective to the greater extent of knowledge that can be expressed and discovered. Comparisons will also be made to demonstrate the added advantage of using net rules, against the use of standard boolean logic of negation, disjunction and conjunction, in the realm of evolutionary computation.

Modeling Nonlinear Systems by a Fuzzy Logic Neural Network Using Genetic Algorithms

The main aim of this work is to optimize the parameters of the constrained membership function of the Fuzzy Logic Neural Network (FLNN). The constraints may be an indirect definition of the search ranges for every membership shape forming parameter based on 2nd order fuzzy set specifications. A particular method widely applicable in solving global optimization problems is introduced. This approach uses a Linear Adapted Genetic Algorithm (LAGA) to optimize the FLNN parameters. In this paper the derivation of a 2nd order fuzzy set is performed for a membership function of Gaussian shape, which is assumed for the neuro-fuzzy approach. The explanation of the optimization method is presented in detail on the basis of two examples.

Extraterrestrial and terrestrial applications of the Mössbauer spectroscopy

Mossbauer spectroscopy is a powerful technique used by physicists, chemists, geologists, biologists, metallurgists and scientists from other disciplines. A portable Mossbauer spectrometer, developed for extraterrestrial applications [1], opens up new industrial applications of MBS [1]. But for industrial applications it is also required to have a tool available for fast data analysis that should be easy to handle. The analysis of Mossbauer spectra and parameters is a barrier for a popularization of this wide-applicable spectroscopic technique in industry. Based on experience, it is time-consuming and requires the dedication of a specialist. But the analysis of Mossbauer spectra, from the fitting to the identification of the sample phases, can be fast using genetic algorithms, fuzzy logic and artificial neural networks [2]. Industrial applications are very specific ones and the data analysis can be performed using these algorithms. In combination with an automatic analysis, the Mossbauer spectrometer can be used as a probe instrument, which covers the main industrial applications. It is needed for on-line monitoring of products, processes and studies of case.

Fuzzy logic systems are equivalent to feedforward neural networks

Fuzzy logic systems and feedforward neural networks are equivalent in essence. First, interpolation representations of fuzzy logic systems are introduced and several important conclusions are given. Then three important kinds of neural networks are defined, i. e. linear neural networks, rectangle wave neural networks and nonlinear neural networks. Then it is proved that nonlinear neural networks can be represented by rectangle wave neural networks. Based on the results mentioned above, the equivalence between fuzzy logic systems and feedforward neural networks is proved, which will be very useful for theoretical research or applications on fuzzy logic systems or neural networks by means of combining fuzzy logic systems with neural networks.

Neural-Network-Based Autonomous Star Identification Algorithm

Most of the existing autonomous star identiŽ cation algorithms use direct-match algorithms that prestore the star feature vectors in a database. During recognition, the measurements are compared with the reference feature vectors in sequence or by using binary-tree search. The computation time for the star recognitionwith a traditional model-based system is high, and it increases as the number of the feature patterns in the database increase. We propose an autonomous star identiŽ cation algorithm using fuzzy neural logic networks. This is a parallel star identiŽ cation algorithm with fast training speed. The simulation results based on the SKY2000 star catalog (Myers, J. R., Sande, C. B., Miller, A. C., Warren, W. H., and Tracewell, D. A., “The SKY2000 Master Star Catalog,”AAS/AIAA SpaceMechanics Symposium, AAS, SanDiego, 1997,pp. 1–16) show that the proposed system can achieve both high recognition accuracy and fast recognition speed. Errors due to starmagnitudemeasurement imprecision can also be minimized.

A comparison of various forecasting techniques applied to mean hourly wind speed time series

This paper presents a comparison of various forecasting approaches, using time series analysis, on mean hourly wind speed data. In addition to the traditional linear (ARMA) models and the commonly used feed forward and recurrent neural networks, other approaches are also examined including the Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Neural Logic Networks. The developed models are evaluated for their ability to produce accurate and fast forecasts.

Variable sensitivity in unsupervised clustering tasks with an n-tuple-based self-organising neural network.

This article investigates the application of the SOLNN (Self-Organising Logic Neural Network) n-tuple-based network to character recognition and image segmentation clustering tasks, where the classes consist of a large number of distinct sub-classes. It is shown that the SOLNN clustering performance and node utilisation are both improved by virtue of the distribution constraint mechanism. The clustering results are supported by means of a detailed analysis of the characteristics of each pattern space. This analysis, coupled with comparative results obtained using other self-organising models, illustrates that the SOLNN clusters the patterns in accordance to the pattern space characteristics and thus is well-suited to clustering complex datasets.

Building an active material requirements planning system

A material requirements planning (MRP) system is usually implemented with several constraints, such as replenishment of inventories on a period-by-period bases, which obstruct its dynamic performance. This research proposes an active, bucketless, and real-time MRP system. The active MRP system utilizes hybrid architecture that includes an object-oriented database, fuzzy logic controllers, and neural networks. The object-oriented database, which maintains static data relationships, provides superior capabilities in reusability, complex structure operations, and potential integration. The complementary combination of fuzzy logic controllers and neural networks provides a model-free, human-like decision system. Adding triggers and assertions forms an active MRP model.

An integrated artificial intelligent computer-aided process planning system

This paper presents an integrated artificial intelligent (IAI) system for dynamic computer-aided process planning (CAPP). The system, IAI-CAPP, integrates fuzzy logic (FL) and artificial neural networks (ANN) to perform the dynamic recognition and adaptive-learning tasks of the workpieces and process plans. Also, it adopts the idea of important (critical) feature concept for evaluating the suitability of existing process plans for incoming product designs. In addition, the technique of expert system (ES) is utilized. The system combines variant and generative CAPP and is capable of generating plans that are suitable for workpieces that either are similar to existing workpieces or new. The proposed system described has been realized on a computer prototype program. An illustrative example is also provided.

A NEURO-FUZZY MODEL APPLIED TO FULL RANGE SIGNAL VALIDATION OF PWR NUCLEAR POWER PLANT DATA

The surveillance and control of any industrial plant is based on the readings of a set of sensors. Their reliable operation is essential since the output of the sensors provide the only objective information about the state of the process. The signal validation task is intended to confirm whether the sensors are functioning properly Real-time process signal validation is an application field where the use of fuzzy logic and artificial neural networks (ANNs) can improve the diagnosis of faulty sensors or drift in sensor readings in a robust and reliable way The present work describes the transient and steady state on-line validation method of plant process signals using ANN and fuzzy logic pattern recognition. This method has been developed at the OECD Halden Reactor Project and tested on simulated scenarios covering the whole range of Pressurized Water Reactors (PWR) operational conditions provided by Electricite' De France (EDF) and the Centre D'Etudes De Cadarache (CEA)in France.

Automatically designed fuzzy system for connection admission control in ATM networks

Connection admission control (CAC) is a vital function for asynchronous transfer mode networks. A CAC algorithm should be simple, i.e. economically implementable and fast, and it should be efficient, i.e. allow statistical multiplexing. A solution based on an analytical queueing model is too CPU-intensive and cannot be applied online. The paper proposes a new scheme based on fuzzy logic. The aim is to predict online the cell loss ratio that a connection will exhibit if it is accepted into the network. The CAC scheme is based on a consideration of fuzzy logic and artificial neural networks (ANNs). The ANN is used in the learning phase to tune the fuzzy system automatically. The structure of the neuro-fuzzy system is discussed. A training set obtained by an analytical algorithm, namely the convolution algorithm, is used to develop a learning algorithm and to check the applicability of the technique.

Reasoning with propositional knowledge based on fuzzy neural logic

In this article, a new kind of reasoning for propositional knowledge, which is based on the fuzzy neural logic initialed by Teh, is introduced. A fundamental theorem is presented showing that any fuzzy neural logic network can be represented by operations: bounded sum, complement, and scalar product. Propositional calculus of fuzzy neural logic is also investigated. Linear programming problems risen from the propositional calculus of fuzzy neural logic show a great advantage in applying fuzzy neural logic to answer imprecise questions in knowledge-based systems. An example is reconsidered here to illustrate the theory. © 1996 John Wiley & Sons, Inc.

Simple feedback logic, genetic algorithms and artificial neural networks for real-time control of a collection system

This paper describes a pilot-scale implementation of a simple, real-time control (RTC) algorithm based on feedback and also outlines the development and simulation testing of a new RTC methodology that combines genetic algorithms (GAs) and artificial neural networks (ANNs). Computer simulations indicated that the simple feedback logic could reduce pumping by 50 to 80 percent if used to replace the existing RTC system in the test area. Experience with the algorithm after its implementation has confirmed the potential of the algorithm to reduce pumping. Additional simulations with an emerging approach to control (based on GAs) indicated possibilities of reducing pumping still further. Although relatively simple flow routing was used in the GAs, these algorithms do not restrict flow routing to any particular method. If highly accurate flow routing is incorporated, GAs are likely to be rendered too slow for on-line applications. Nevertheless, GAs can still be used, because they can be combined with fast executing on-line algorithms, such as ANNs. This possibility was demonstrated by training a multi-layer ANN to approximate one of the GAs developed. In verification runs the trained ANN provided virtually the same control decisions as did the GA used as the source of the training data.

Artificial intelligence based techniques for processing segmented images of wood boards

Veneer boards are made by bonding together thin sheets of wood. Normally, grading of these sheets is carried out to ensure that only high-quality sheets are used to make high-quality boards. Frequently, this quality control task is performed by a human inspector. However, due to the speed and repetitive nature of the job, human graders cannot always grade the boards accurately. To improve the efficiency of grading, attempts are being made to automate it using automated visual inspection (AVI). Integral to the AVI process is segmentation, which is concerned with separating clear wood and defective areas in the image. The defective areas are labelled as segmented objects. However, after segmentation has been performed, two problems can occur. Firstly, clear wood areas may be falsely detected as defects and, secondly, a defect may be represented by more than one segmented object. This paper describes two techniques that have been used to overcome these problems. The techniques were inspired by the artificial intelligence (AI) techniques of fuzzy logic and self-organizing neural networks.

TMLNN: triple-valued or multiple-valued logic neural network

We discuss the problem of representing and processing triple-valued or multiple-valued logic knowledge using neural network in this paper. A novel neuron model, triple-valued or multiple-valued logic neuron (TMLN), is presented. Each triple-valued or multiple-valued logic neuron can represent a triple-valued or multiple-valued logic rule by itself. We will show that there are two triple-valued or multiple-valued logic neurons: TMLN-AND (triple-valued or multiple-valued "logic and" neuron) and TMLN-OR (triple-valued or multiple-valued "logic or" neuron). TMLN-AND can realize triple-valued or multiple-valued "logic and" while TMLN-OR can realize triple-valued or multiple-valued "logic or." Two simplified triple-valued or multiple-valued logic neuron models are also presented. We can show that a multiple-layer neural network (TMLNN) made up of triple-valued or multiple-valued logic neurons can implement a triple-valued or multiple-valued logic inference system. The training algorithm for TMLNN is presented and can be shown to converge. In our model, triple-valued or multiple-valued logic rules can be extracted from TMLNN with ease. TMLNN can thus form a base for representing logic knowledge using neural network.

Image segmentation with the SOLNN unsupervised logic neural network

In this article, an image segmentation method based on the SOLNN self-organising logic neural network is studied. The input image is initially processed using the TCS texture-highlighting technique and is then presented to the SOLNN network which segments it. The SOLNN is characterised by a variable sensitivity which enables it to be fine-tuned to detect different sub-textures within each texture to the desired degree of detail. The experimental results reported here illustrate the fact that the SOLNN indeed clusters accurately the textural information so that each cluster represents a single texture even for images which are objectively very difficult to segment. Thus, it is supported that the proposed approach leads to the design of an effective texture-based image-segmentation system.

Inductive neural logic network and the SCM algorithm

Neural Logic Network (NLN) is a class of neural network models that performs both pattern processing and logical inferencing. This article presents a procedure for NLN to learn multi-dimensional mapping of both binary and analog data. The procedure, known as the Supervised Clustering and Matching (SCM) algorithm, provides a means of inferring inductive knowledge from databases. In contrast to gradient descent error correction methods, pattern mapping is learned by an inductive NLN using fast and incremental clustering of input and output patterns. In addition, learning/encoding only takes place when both the input and output match criteria are satisfied in a template matching process. To handle sparse and/or noisy data sets, we also present a weighted voting scheme whereby distributed cluster activities combine to produce a final output. The performance of the SCM algorithm, compared with alternative systems, is illustrated on three benchmark problems: (1) mushroom classification, (2) sonar return signal recognition, and (3) sunspot time series prediction.

Hierarchical fuzzy controllers

The paper presents Fuzzy Logic Controllers (FLC) with hierarchically structured rule base as nonlinear blocks used in a control system for direct control or supervision of standard controllers. Unified approach based on Fuzzy Logic Neural Networks (FLNN) is used to reveal the essential functions of a FLC. The FLNN is then extended to the case where decomposition of a rule base leads to simplification of controller design and tuning. Two structures of hierarchically organized groups of rules are discussed. Singlestage decision making structure with groups of rules organized hierarchically according to their specificity and a multi-stage decision making architecture with chained rules. Data induced error backpropagation technique for fine tuning of parameters of the second structure is derived.

A novel system for coloured object recognition

Many industrial inspection tasks can be performed satisfactorily by conventional image processing systems which utilise feature extraction and related feature measuring algorithms. However, for the inspection of parameters such as texture or colour, these techniques may be inadequate. This paper outlines a new system which incorporates algorithms for coloured image processing and a logical neural network for subsequent pattern classification. Past research has indicated that logical neural nets can provide quite a good recognition performance for artefacts which are complex. However, in the area of coloured object inspection, the inspection tasks are more difficult. Furthermore, if the lighting source changes, the image of a coloured object will vary considerably and this will significantly affect the outputs of the networks. This paper presents a practical approach for colour invariant recognition and a method of adaptive sampling which utilises colour information more effectively. The system which is armed with these techniques provides a good recognition performance for artefacts which are both complex and coloured and, also, it is robust in terms of tolerance to noise and lighting.

Image morphological operations and edge detection using optical programmable logic neural networks

Based on the concept of 2-D programmable logic neural networks, an optical programmable image processor for neighborhood operations is proposed. The spatial polarization encoding and space-variant techniques are used in the design of an optical system which can perform multiple-instruction multiple-data image processing. The system is a cellular logic image processor with nine local neighbor inputs and three outputs. In particular, the applications of the optical scheme in binary image morphological operations and edge detection are described in detail. Simulation results are also presented.