We focus on a shared augmented environment (SAE) as an almost ideal face-to-face collaborative virtual workspace. In an SAE, multiple users can observe both a virtual world and real partners through optical see-through head mounted displays. This paper describes two experiments for verifying the effectiveness of an SAE compared with a conventional shared virtual environment (SVE) and exploring improvement of them. Through the experiments, the effectiveness of an SAE compared with an SVE was confirmed. It was also confirmed that enhancement of the shared environment with computer graphics, i.e. displaying a partner's body in an SVE, drawing a linear as a partner's viewing directions and emphasizing virtual objects to which a partner pay attention, improves workers' feeling and collaboration efficiency.

This paper presents a novel approach for solving the challenging problem in intelligent control of manufacturing systems, i.e. the integration of low-level system sensing and simple control with high-level system behavior and perception. The proposed Max-Plus Algebra model combined with event-based planning and control provides a mechanism to efficiently integrate task scheduling, sensing, planning and real-time execution so that task scheduling, which usually deals with discrete types of events, as well as action planning, which usually deals with continuous events, can be treated systematically in a unified analytical model. More importantly, the unique feature of this approach is that interactions between discrete and continuous events can be considered in a unified framework. This feature allows the manufacturing system to intelligently cope with unexpected events and uncertainties so that the efficiency and reliability of the task schedule and action plan can increase significantly. A robotic manufacturing system is used to illustrate the proposed approach. The experimental results clearly demonstrate the advantages of the proposed approach.

It is a mission of this contribution to recognize and synthesize all the efforts in industry and in management science to strengthen our techniques and tools for successfully solving increasingly complex leadership problems in manufacturing industries. With the high standard of the work sharing method--the so called Taylorism principle--for cost efficient and mass production, invented at the beginning of the 20th century and the opening of the world market for global sales of goods and services a gigantic progress in living standards was reached. But at the beginning of the 21st century we are needing new ideas and methods for the guidance of overcoming increasing complexity. The holistic eigensolution presents a new operational framework for viewing and controlling the behavior of businesses. In contrast to the traditional process for viewing complex business systems through the intricate analysis of every part of that system, the authors have employed a technique used by physicists to understand the characteristic of `eigen' behaviors of complex physical systems. This method of systems analysis is achieved by observing interactions between the parts in a whole. This kind of analysis has a rigorous mathematical foundation in the physical world and it can be employed to understand most natural phenomena. Within a holistic framework, the observer is challenged to view the system form just the right perspective so that characteristic eigenmodes reveal themselves. The conclusion of the article describes why exactly the intelligent manufacturing science--especially in a broader sense--has the responsibility and chance to develop the holistic eigensolution framework as a Taylorism II-principle for the 21st century.

Multiple parts can be built simultaneously in one RP cycle, reducing significantly the build time for clients--an important issue for time compression technologies. Common to all the RP approaches is that they are based on sliced 3D, solid CAD-model, and they all add material layer by layer on the part being built. As in any manufacturing process, batch formation is important for increased production. Concurrently with rapid prototyping, a critical issue is the need for disassembly or unpacking of the job mix of multiple parts, after the rapid prototyping cycle. Ikonen et. al. have discussed a bin-packing algorithm based on a genetic algorithm, called GARP. In GARP, the CAD-files of the job mixes are used to evaluate the potential for being packed into the RP production volume. This paper addresses the generic unpacking issues for the SLS and other related RP technologies. The paper suggests an object-oriented framework for the development of heuristics and algorithms for evaluating the quality of packing obtained by GARP for the unpacking problem.

Genetic algorithms (GAs) are excellent approaches to solving complex problems in optimization with difficult constraints, and in high state space dimensionality problems. The classic bin-packing optimization problem has been shown to be a NP- complete problem. There are GA applications to variations of the bin-packing problem for stock cutting, vehicle loading, air container loading, scheduling, and the knapsack problem. Mostly, these are based on a 1D or 2D considerations. Ikonen et. al. have developed a GA for rapid prototyping called GARP, which utilizes a 3D chromosome structure for the bin- packing of the Sinterstation 2000's build cylinder. GARP allows the Sinterstation to be used more productively. The GARP application was developed for a single CPU machine. Anticipating greater use of time compression technologies, this paper examines the framework necessary to reduce GARP's execution time. This framework is necessary to speed-up the bin-packing evaluation, by the use of distributed or parallel GAs. In this paper, a framework for distribution techniques to improve the efficiency of GARP, and to improve the quality of GARPis solutions is proposed.

A unique 3D packing problem with non-convex parts and without a gravity constraint can be defined in a selective laser sintering rapid prototyping machine. The goal for the packing task is to pack the parts to be manufactured as tightly as possible to maximize volume and machine time utilization. A genetic algorithm is used as a search engine to find a good packing pattern for parts. Each individual in a population represents one packing solution. The chromosomal representation is a 3D ordered list of integers where each sublist has a different allele set. A fitness function simulates the packing of parts and also evaluates the quality of a solution. To calculate part intersections, the fitness function uses methods common in computational geometry. Due to the chromosome structure used, there is a lack of genetic material in the population. Methods to introduce new material into the population are defined and tested. Experiments with more difficult packing problems, where all parts are complex in shape, prove that the developed genetic algorithm is robust and able to find a good solution in most problem instances.

Time-to-market has become the most critical factor in the success or failure of any new product introduction, particularly in today's increasingly global economy. The quest for speed is nothing new, but the technologies available, and the manner in which they can be applied, are constantly evolving. In this paper a new quantitative process model for product development is presented, which focuses on the representation of design iterations and their effects on the total product development process. The relations between the individual process steps are represented in design structure matrices and separated into dependent, independent and interdependent processes. By introduction of activity duration and transition probabilities for each process step and by application of the z-transform a quantitative model for coupled processes is created. This model is called Dynamic Product Development because it allows dynamic coupling of all product development functions. The resulting model is examined by analysis methods of discrete dynamic systems for its timing behavior. Probability distribution, expected value and variance of the lead time as well as sensitivities concerning individual development steps are determined. On the basis of a state space representation further participation factors and dominant modes are determined. The results supply the possibility for analyzing complex dependencies and time reduction of development process steps and provide a decision support tool for the product development process.

A sensor system for inspecting the layer surface quality in stereolithography process is proposed in this paper. Since stereolithography process builds 3D shape by forming layers repeatedly, it is very important to process each layer of stereolithography process products in some favored conditions: Every layer should be cured uniformly and hardly enough so that the adjacent two layers can stick together to each other. However, in many applications, two kind of defects are frequently found, i.e. void and delamination. Void is cavity inside the built part and delamination is detachment of the bond between two adjacent layers. To inspect such defects, we propose a sensor system which consists of a laser source, a galvanometer scanner, a photo- detector, a few lenses, and a beam splitter. In this sensor system, the laser beam and the field of view of the detector are co-axially positioned and scanned over the product surface by the galvanometer. The reflected light is then detected by the photo-detector. And from the photo-detector signal, the surface condition and quality of the layer being inspected can be estimated. Since stereolithography products are very transparent, the system needs very fine tuning of the system parameters that include the power of laser beam and the sensitivity of the detector, and etc. The experimental results are obtained for products of a variety of shapes and several cases are presented and discussed in detail.

We propose in the paper a process, based on the use of a high velocity, electric arc metal spray thoroughly atomized on a master model in order to make a shell. The mold is realized by placing the sprayed alloy into a frame and reinforcing it with appropriate material. The first part of the paper presents briefly the main rapid prototyping and tooling techniques used in industry. We then describe the tool/moldmaking method which is a process to reproduce shapes very accurately. Given any model, a metal shell can be build up around it to desired thickness. We describe and comment the tasks leading to the shell making. The realization of the mold itself is also presented. In most cases, the spray alloy shell must be reinforced with backup. We expose the application fields of this new technique especially in term of product and process validation. We focus on the natural links with stereolithography. By comparison with conventional processes, we show how fabrication time and costs are dramatically reduced by integration of this new moldmaking method. Finally, we present the limits of the technique and the perspective of work and further research.

Ever-growing global competition forces manufacturers to deliver more competitive products with better quality, lower price and in short time. One of the most important and challenging tasks faced by value-added product manufacturing industry is substantial reduction of product development time. Rapid prototyping technologies have received significant interests from both research and industrial communities. Due to the model accuracy, integrity and strength problems, their applications are limited. The most common sources of errors in the rapid prototyping and manufacturing systems including Cubital machines can be categorized as mathematical, process-related or material- related errors. In this paper, we present an analysis of accuracy of a Cubital Solider 4600 machine and an application on pattern and mould design and manufacturing. The experimental study determines the relationships between the machine accuracy and dimensions and parameter setting, which can be used to control the accuracy of parts to be built.

In this paper, we propose a training algorithm for VLSI neural networks with digital weights and analog neurons using in-the-loop training strategy. The use of digital weights in a neural network implementation imposes new issues that are not present in simulation environments. One of the problems is that a neural network implementation will not work properly when using the digitized version of the continuous weight solution. This phenomenon is especially evident when the digital weight resolution is very low due to some fabrication constraints. In this paper the training strategies for dealing with digital weights are investigated. The proposed training algorithm is by measuring the sensitivity of each weight to its error function and then by perturbing the weights of higher sensitivity values to perform retraining process. Our experimental results indicate that the algorithm is feasible and particularly suitable for the digital weights with low number of bits.

In the complex environment of an automated manufacturing system, decision-making is one of the most important and difficult tasks. This responsibility lies with the manufacturing control system which must be capable of effectively monitoring the state of the manufacturing system and responding appropriately to a rapidly changing environment. The work reported here is motivated by the need to provide accurate information about the manufacturing system that can be used to assist the decision-making process for manufacturing control systems. In order to achieve this goal, an emerging analysis technique, infinitesimal perturbation analysis (IPA), for discrete- event system is used in combination with fuzzy set theory and manufacturing system information to determine relative service priorities for automatic stations in an assembly line. A high-speed assembly line, which is simulated in Arena, is used as a test bed. This model is integrated with C++ code, which is developed for IPA, and fuzzy set theory. The results are reported and compared for different scenarios.

Pulsed GTAW was used widely in butt welding of thin plate. Top surface depression occurred without filler wire in full penetration, while reinforcement height was assured with filler wire. Currently butt welding process control of thin plate welding during pulsed GTAW with filler wire was depended on manual experience and the consistency of seam shape was hardly attained. Based on self-developed vision sensor, double-side images of weld pool were captured simultaneously in a frame. By image processing the topside dimension and shape of weld pool, such as area, length, maximum width, the similarity of reinforcement, and the coefficients of multinomial regression of boundary, and the backside dimension such as area, length, maximum width and the similarity of height were calculated. A fractional factorial technique was used to design the experiment. Artificial neural network was applied to establish the steady model for predicting backside dimension of weld pool. The input of the model was the topside dimension, shape of weld pool and welding parameters, such as pulse current, base current, arc voltage, pulse duty ratio, welding speed, and wire feeding rate. The output of the model was the backside dimension of weld pool. Finally the variance method was used to test the validity of the model.

The increasing use of computer controlled mobile robots in various tasks has created a growing technological need for efficient low cost means and methods for mobile robot guidance. This work presents a robot navigation and guidance system that consists of a directional light sensor and a set of at least three uniquely coded light sources. The three light sources provide distinct light fields in the operating area of the robot. The light sensor, mounted on a mobile robot, is used to measure the angle of the direction of origin of each of the three light sources with respect to the robot reference frame. The local room coordinates of the robot can then be determined by an on-board microprocessor system. The planar position and orientation of the robot can be used to control the drive and steering mechanisms to reach a given target position or track a prescribed trajectory. Control methods for robot guidance are discussed.

Bar code labeling of products has become almost universal in most industries. However, in the steel industry, problems with high temperatures, harsh physical environments and the large sizes of the products and material handling equipment have slowed implementation of bar code based systems in the hot end of the mill. Typical laser-based bar code scanners have maximum scan distances of only 15 feet or so. Longer distance models have been developed which require the use of retro reflective paper labels, but the labels must be very large, are expensive, and cannot stand the heat and physical abuse of the steel mill environment. Furthermore, it is often difficult to accurately point a hand held scanner at targets in bright sunlight or at long distances. An automated product tag reading system based on CCD cameras and computer image processing has been developed by West Virginia University, and demonstrated at the Weirton Steel Corporation. The system performs both the pointing and reading functions. A video camera is mounted on a pan/tilt head, and connected to a personal computer through a frame grabber board. The computer analyzes the images, and can identify product ID tags in a wide-angle scene. It controls the camera to point at each tag and zoom for a closeup picture. The closeups are analyzed and the program need both a barcode and the corresponding alphanumeric code on the tag. This paper describes the camera pointing and bar-code reading functions of the algorithm. A companion paper describes the OCR functions.

This paper reports a new approach to error compensation for inaccuracies in position control for the end-effector of a Robot Arm. The goal is to overcome the problem of inaccuracy, due to the low precision in manufacturing of Robot Arms and the flexibility of their structure, by means of machine intelligence. Utilizing a mesh sensory system, a Real Time Monitoring System is designed. The position of the end-effector is monitored in real time and the positioning data for the end-effector is collected. A direction independent filtering system is designed to eliminate the noise from the collected data. After extracting the error map from the collected data, a novel Proportional Keen Approximation Method is implemented to generalize the error map. One of the main features of this method is the elimination of the training stage as in the Artificial Neural Networks. Using the knowledge obtained from the maps, the system compensates for the errors.

Intensive research in the development of Electro Discharge Machining (EDM) has led to highly sophisticated EDM machines suited for conventional precision engineering purposes as well as for microcomponent fabrication. Recently, increased attention has been paid to the exploration of Micro-EDM ((mu) -EDM). It could be shown that within EDM applications especially the electrodes quality, its handling and the adjustment procedures determine the accuracy of the machined product. This paper describes various electrode fabrication methods and further introduces the development of an actuator for (mu) -EDM. Driven by piezoelements it provides 3D vibrations of electrode or workpiece which enhances EDM- process performance significantly. Furthermore, the first steps integrating a vision control system into an EDM- machine are described. Problems, occurred hereby, are stated and analyzed. The demonstration of several (mu) -EDM applications accompanies the technical information and shows the extensive possibilities of the technology.

A major problem when using off-line programming system is today's robot-based workcells still is that for more complex tasks like e.g. arc welding, general coating, grinding or laser-based applications, the support by these programming tools is poor. Particularly, together with intricate workpieces designed from free-form surfaces (e.g. NURBS) the programming of the desired robot motion still requires a lot of manual, teach-in like work by the programmer. This paper presents two approaches which now support the programmer in a comprehensive manner to solve this problem efficiently.

Feature Based Cost Modeling is thought of as a relative new approach to cost modeling, but feature based cost modeling had considerable development in the 1950's. Considerable work was published in the 1950's by Boeing on cost for various casting processes--sand casting, die casting, investment casting and permanent mold casting--as a function of a single casting feature, casting volume. Additional approaches to feature based cost modeling have been made, and this work is a review of previous works and a proposed integrated model to feature based cost modeling.

This paper deals with the integration of reverse engineering with CAM for duplicating an existing part without a CAD model quickly and accurately. Two commercial packages, SURFACER and SmartCAM, were selected as the software tools for reverse engineering and NC code generation respectively. By utilizing the open architecture environment provided by SURFACER, a machining-oriented feature model can be constructed during the model creation phase. This model consists of not only the part's geometric entities but also all the machining information required for toolpath generation. Upon a knowledge-based process planning, this machining-oriented feature model is converted into a process model, which will then be used to create a job file and a macro file, both in the SmartCAM format. The execution of these two files inside SmartCAM can automatically generate the toolpaths and NC codes required for part machining.

This paper describes work being undertaken in the development of an intelligent distributed quality function deployment (IDQFD) system, which supports product design team to transfer and deployment the `Voice of Customer' through `House of Quality' into the various stages of product planning, engineering and manufacturing. The requirement modeling of products, the optimization in QFD are indicated. The framework of the system, including QFD tools and platform for distributed collaborative work in QFD, is described. The strategy and methods for the collaboration processing in QFD process are presented. It shows promise for application in practice.

The discrete cosine transform (DCT) is now well recognized as one of the most important technique in image data compression. Among the class of transforms possessing fast computational algorithms, the cosine transform has a superior energy compaction property. Due to its simple implementation scheme, the DCT is widely used as a substitute to the optimal Karhunen Loeve transform. In this paper, the discrete cosine transform is presented and an algorithm for its implementation is developed. The picture is firstly transform coded using 8 X 8 sub-blocks then a quantization and an entropy coding are used.

Virtual Computer Corporation has combined the latest reconfigurable component technology with a number of advanced software tools in one easy to use `system approach' to digital design. This Hardware Software Co-Design Development System contains all the components necessary for configurable computing implementation in one integrated package. This paper will describe the features and use of this system.

This paper describes an intuitive and efficient way to design not only 3D shapes but also their geometric constraints within an immersive modeler. The fundamental concept of the proposed method is based on our immersive modeler VLEGO II, which allows multiple users to create 3D virtual objects only by assembling simple 3D geometric shapes (shape primitives) like with real toy blocks. In our system, geometric constraints are presented as primitives (constraining primitives) like normal shape primitives, and the geometric constraints are imposed on virtual objects simply by assembling shape and constraining primitives. The system also provides flexible two-handed interaction to support intuitive and efficient manipulations of objects.

Maintaining stable, robust, and consistent software is difficult in face of the increasing rate of change of customers' preferences, materials, manufacturing techniques, computer equipment, and other characteristic features of manufacturing systems. It is argued that software is commonly difficult to keep up to date because many of the implications of these changing features on software details are obscure. A possible solution is to use a software generation system in which the transformation of system properties into system software is made explicit. The proposed generation system stores the system properties, such as machine properties, product properties and information on manufacturing techniques, in databases. As a result this information, on which system control is based, can also be made available to other programs. In particular, artificial intelligence programs such as fault diagnosis programs, can benefit from using the same information as the control system, rather than a separate database which must be developed and maintained separately to ensure consistency. Experience in developing a simplified model of such a system is presented.

Remediation of hazardous waste may require the application of new technologies. These technologies are usually designed by experts who know very little about safety and health issues. A user-friendly, easily accessible system that is devised to assist designers during the design phase by pointing out the different downfalls of their design with respect to safety and health issues and impact, can dramatically improve hazardous waste remediation safety, reduce costs, and most important, reduce the chance of injuries.

For many applications, which involve the processing and handling of highly variable natural products, conventional automation techniques are inadequate. Field applications involving the processing and handling of these products have the additional complication of dealing with a dynamically changing environment. Automated systems for these applications must be capable of sensing the variability of each product item and adjusting the way each product item is processed to accommodate that variability. For automation to be feasible, both fast processing of sensor information and fast determination of how product items are handled, is vital. The combination of sensor equipped mobile robotic systems with artificial intelligence techniques is a potential solution for the automation of many of these applications. The aim of this research is to develop a software architecture which incorporates robotic task planning and control for a variety of applications involving the processing of naturally varying products. In this paper we discuss the results from the initial laboratory trials for an asparagus harvesting application.

As the emphasis of forest stand management shifts towards implementing ecosystem management, managers are examining alternative methods to harvesting stands in order to accomplish multiple objectives by using techniques such as shelterwood harvests, thinnings, and group selection methods, thus leaving more residual trees to improve the visual quality of the harvested land. Contemporary harvesting practices require the creative use of existing cable and ground-based technology. Silvicultural operations such as group selection methods, shelterwood harvests, partial cuts and thinnings require substantial planning in order to realize a profitable logging endeavor. THIN-PC is a personal computer based version of the original THIN developed by Chris B. LeDoux and David A. Butler in 1981. THIN and THIN-PC are computer simulation models, which evaluate the single-stage pre-bunch and swing methods of cable yarding. THINEX, the expert system developed as a result of this research, sues the results from THIN-PC and allows planners, managers, loggers to evaluate, plan, and execute profitable harvesting practices and permit effective sensitivity analysis by linking the terrain data with the production system information. Using THINEX, the user will be able to analyze several kinds of scenarios pertaining to a stand of trees and their harvesting resulting in the acquisition of production rate and cost values.

The growing scientific knowledge and rapid progress in medical imaging techniques has led to an increasing demand for better and more efficient methods of remote access to high-performance computer facilities. This paper introduces a web-based telemedicine project that provides interactive tools for surgical simulation and planning. The presented approach makes use of client-server architecture based on new internet technology where clients use an ordinary web browser to view, send, receive and manipulate patients' medical records while the server uses the supercomputer facility to generate online semi-automatic segmentation, 3D visualization, surgical simulation/planning and neuroendoscopic procedures navigation. The supercomputer (SGI ONYX 1000) is located at the Computer Vision and Image Processing Lab, University of Louisville, Kentucky. This system is under development in cooperation with the Department of Neurological Surgery, Alliant Health Systems, Louisville, Kentucky. The server is connected via a network to the Picture Archiving and Communication System at Alliant Health Systems through a DICOM standard interface that enables authorized clients to access patients' images from different medical modalities.

This paper presents a fast printed circuit board (PCB) inspection methodology based on the segmentation of the PCB image into basic patterns and context sensitive filtering of the difference image for functional defect detection. The system consists of three parts: (1) segmentation of the artwork and test board images into basic patterns using the CAD data for the artwork, (2) comparison of the test board image windows with the corresponding artwork image windows, (3) context-sensitive filtering of the difference image obtained in step (2) for the determination of the defects and their locations. For image representation, runlength encoding (RLE) is used and processing is done directly on RLE data. This provides the needed efficiency in processing huge amounts of image data, which is typical in PCB inspection. Working on the difference image results in considerable reduction in the data to be processed during the filtering stage. Context-sensitive filtering enables classification of defects in such a way that only potentially functional defects are caught and reported in the final analysis.

An automated product tag reading system based on CCD cameras and computer image processing has been developed by West Virginia University and demonstrated at the Weirton Steel Corporation. The system reads both a 50-mil barcode and a string of six numbers on a four-inch by six-inch tag fastened to the end of a steel slab. Feedback from the image is used to point and zoom the camera, making the system effective at ranges up to thirty feet, and in bright sunlight-situations where handheld barcode scanners are ineffective. A video camera is mounted on a pan/tilt head and connected to a personal computer through a frame grabber board. The whole system is mounted on a slab hauler--a huge wheeled machine for carrying 100 tons of steel slabs at a time. The slab hauler backs into position and presses `start' on a touchscreen operator interface. A wide-angle image is grabbed, and the computer analyzes the images to find product ID tags in scene. The camera is then zoomed and pointed one-by-one at the tags for closeup images. Geometric warping is done on the closeup images to correct for viewing angle distortion, and both the barcode and the alphanumeric code are read by the software and reported to the inventory management system via radio modern. This paper discusses the neural network-based system for reading the characters on the tag. The camera pointing system and barcode reader are discussed in a companion paper.

Since February 1996 a large-scale European project called REMPRODUSE-Cu has been in progress. Its main objective is to provide a comprehensive approach to overcome the problems found when electromechanical systems reach the end of their useful life. How these problems could be overcome by a smarter recycling system and a smarter product design is in this project exemplified for electric motors. Today small electric motors when worn out are put in a shredder, due to problems with the disassembly. To be able to perform the disassembly in a proper way measurement and sensing as well as industrial robots will play an important part. In this paper a robotized work station for end-of-life treatment of industrial motors is presented. There are two main steps in the work. The first step is an inspection where the functionality of the motor is checked and the second step is robotized automatic disassembly for motors that can not be reused. This paper deals mainly with the second step. The robotized disassembly station consists of two industrial robots with appliances.

The increasing demand for on-time delivery and low price forces manufacturer to seek effective schedules to improve coordination of multiple resources and to reduce product internal costs associated with labor, setup and inventory. This study describes the design and implementation of a scheduling system for J. M. Product Inc. whose manufacturing is characterized by the need to simultaneously consider machines and operators while an operator may attend several operations at the same time, and the presence of machines requiring significant setup times. The scheduling problem with these characteristics are typical for many manufacturers, very difficult to be handled, and have not been adequately addressed in the literature. In this study, both machine and operators are modeled as resources with finite capacities to obtain efficient coordination between them, and an operator's time can be shared by several operations at the same time to make full use of the operator. Setups are explicitly modeled following our previous work, with additional penalties on excessive setups to reduce setup costs and avoid possible scraps. An integer formulation with a separable structure is developed to maximize on-time delivery of products, low inventory and small number of setups. Within the Lagrangian relaxation framework, the problem is decomposed into individual subproblems that are effectively solved by using dynamic programming with additional penalties embedded in state transitions. Heuristics is then developed to obtain a feasible schedule following on our previous work with new mechanism to satisfy operator capacity constraints. The method has been implemented using the object-oriented programming language C++ with a user-friendly interface, and numerical testing shows that the method generates high quality schedules in a timely fashion. Through simultaneous consideration of machines and operators, machines and operators are well coordinated to facilitate the smooth flow of parts through the system. The explicit modeling of setups and the associated penalties let parts with same setup requirements clustered together to avoid excessive setups.

In a previous paper we proposed a new kind of factory system that uses mobile robots to transport parts and materials between assembly stations. In this system, which we call an `assembly network,' the factory's assembly stations operate independently of a centralized production plan and negotiate with each other (via message passing) for access to robots. Using this assembly network as a testbed, we are developing distributed algorithms to allow the stations to organize themselves into small `communication clusters.' In this paper we will describe the current clustering algorithm that we are using in more detail than in the previous paper.

In the paper we develop a discrete dynamic model of concurrent cyclic processes and state two formal decision problems concerning testing the realizability of such system at its design phase and testing the safety (with respect to deadlocks) of control decisions undertaken on-line. We prove the NP-completeness of the problems and point out that such complexity clearly hinders a practical application of the optimal (i.e. the least-restrictive) approach proposed. Thus, we discuss a pragmatic solution for deadlock handling being a compromise between the restriction level of the algorithms required and their computational time-complexity. We introduce a satisfactory condition for realizability of the system and prove that in the cyclic systems which satisfy the condition we can employ deadlock avoidance algorithms developed for pipeline processes.

One of the most widely used applications of digital imaging in manufacturing is product inspection. Real-time control of robot motion is second emerging area. In both applications, detection and tracking of moving targets in the image are desirable and in some cases necessary abilities. Current hardware equal to this task is still extremely expensive, limiting its application in low-end manufacturing venues. Despite dramatic increases in processing speed for digital images, tracking moving targets in high-resolution images is still difficult for personal computer based systems. Various forms of image sampling have been used over the years to overcome this problem, allowing processing of a smaller number of pixels by localized searches, multiresolution processing, and other methods. Recently wavelet analysis has become popular as a powerful means of image compression and processing. In this work. Daubechies wavelets are used as basis functions for compressing sequences of images containing targets which include motion in three-dimensions (x and y translations plus rotation) against background clutter. The wavelet coefficients are used as an alternative representation of the image, and analysis is carried out to identify and track targets in the wavelet space. Inverse transformations to Cartesian space will be used to characterize the target motion once it has been determined in the wavelet domain. Preliminary results of the research are shown.

The process of the object selection by multispectral systems is analyzed. The example of such system is the aircraft control device, created for the selection of concrete surfaces, such as runways, against the natural backgrounds. The reliability of object recognition process depends on the probability of object selection. The optimal choice of bands for multispectral system, using the pattern recognition theory approaches, can provide the maximum of selection probability. The method of choice and substantiation the number of multispectral system channels and their bands is proposed. It is shown, that the two-spectral system is enough to solve the task of selection. The catalogue of channels optimal bands is composed. The solidity of results and recommendations is corroborated by Monte-Carlo simulation of object selection process. The results of the investigation can be used during the creation of aircraft control devices.