This paper is concerned with the problem of optimizing surface following control in automated systems. Tracking a surface is an integral task for many autonomous system. It can be used for navigation, surface preparation or object recognition. There are two types of control for surface following, continuous and discontinuous. The robot may maintain contact and continuously track the surface or touch the surface at discontinuous points. A balance is sought between each surface tracking method in the path planning phase, in order that the whole process be optimized in terms of time to complete the task and the amount of data collected. The tracking method is computed by the tracking algorithm using the partial data sets provided by sensors. It is common practice to outfit automated systems with the ability to gather data from many sensors. As the environmental conditions change, sensor reliability changes. Thus, the system's reliance on sensor data must also change. This work focuses on the addition of the supervisory learning module for choosing the method of surface tracking.

Design and manufacture of orthopedic devices using rapid prototyping technologies has been until recently a highly iterative process that involves multiple users, including doctors, design engineers and rapid prototyping experts. Existing systems for creation of orthopedic parts through rapid prototyping do not follow the principles of concurrent engineering and design for manufacture. This leads to excessive communication between parties and delays in product realization time. In this paper, we lay out the framework for a unified expert system that will enable a doctor to create quickly and easily fully functional prosthetics and orthopedic implants. Necessary components of the model acquisition process should include volumetric segmentation of objects from a CT or MRI dataset and NURBS surface fitting to the boundary points. Finite element analysis and surface model modification modules are also needed, but should be provided in an intuitive fashion for doctors who are not experienced in computer aided design. Preprocessing for rapid prototype building should be automatic, and should include optimal orientation, support structure generation and build simulation modules. Finally, the model should be passed to the rapid prototyping machine in a presliced format for speed and accuracy.

This paper discusses automated scheduling as it applies to complex domains such as factories, transportation, and communications systems. The window-constrained-packing problem is introduced as an ideal model of the scheduling trade offs. Specific algorithms are compared in terms of simplicity, speed, and accuracy. In particular, dispatch, look-ahead, and genetic algorithms are statistically compared on randomly generated job sets. The conclusion is that dispatch methods are fast and fairly accurate; while modern algorithms, such as genetic and simulate annealing, have excessive run times, and are too complex to be practical.

Oil- and gas-pipelines must be examined in regular intervals for defects like metal loss. For this reason the Pipetronix company has developed different probes which collect a high number of ultrasonic readings of the wall condition. Based on this measurement the research center for computer science has implemented the automatic inspection system NeuroPipe. The kernel of this inspection tool is a hybrid neural classifier which was trained using manually collected defect examples. The following paper focuses on the aspects of the successful use neural network learning technology for this industrial application. Furthermore the difficulties, when applying these techniques, are discussed.

Thermoplastic tow placement is an merging manufacturing technology that gives more flexibility in the design of parts and in cost reduction through on-line consolidation, compared with traditional labor- and time-intensive autoclave processing. This research is focused on developing an on-line control technique that is fast and reliable and can be used to maximize the process throughput. Recently developed process models are integrated into a neural- network-based optimization package, which is capable of locating optimum setpoints in minimal time. A control algorithm with feedback from an IR thermal camera has been developed and is used to achieve the desired process temperature. A feedforward neural network with a cascadien architecture is used to simulate the process. Simulation computations are now possible in less than a second, compared to around 20 minutes for the original FORTRAN simulation. This allows running of the process models on- line for control purposes. An optimization algorithm that suits the rough topology of the network has been written and tested. The algorithm is based on weighting the quality outputs of the neural network and finding the highest process speed for a given minimum quality. The output of the optimization is used as an input to the robot controller and to a temperature controller for the process.

A patented technique for remotely measuring layer thickness of transparent fluids and solids on surfaces has been incorporated into Canpolar East's Smart Camera and been validated for measurement of ice layer thickness. This application addresses the need for reliable detection of icing on aircraft. The thickness measurement technique is carried out by directing a laser beam at the surface with the ice layer. A pattern of illumination appears on the surface as a result of internal reflection of the laser light within the ice layer. The layer thickness is calculated by a simple formula involving the size of the image pattern and the refractive index of layer material. The system was tested using a wedge-shaped ice sample with a maximum thickness of about 10 mm. The data indicated a system measurement accuracy of +/- 0.1 mm.

During the past few years, the need for large scale and complex systems has become obvious. They are necessary to intelligently carry out tasks in the area of transportation, manufacturing, and maintenance. Up to now, the control structures were usually designed as a hierarchical and centralized structure with a top-down process for planning and decision making. The number and complexity of the hierarchical layers determine the time that is required for a reaction and also for the quality of a chosen action. In most cases, additional components, actuators or sensors have to be added during the development cycle to improve the capability of the overall system. In this case and if the integration of the components' capabilities is required, it is easy to see the disadvantages of the hierarchical and centralized approach in comparison with the advantages existing at the initial system design process. In contrast to that, distributed or decentralized approach in comparison with the advantages existing at the initial system design process. In contrast to that, distributed or decentralized control architectures reveal their main advantages when it is necessary to enhance the system, to integrate components, and to maintain the system. The main disadvantage of not centralized architectures is having to make sure that the system will fulfill an overall or global goal. To investigate these problems, a distributed control architecture for intelligent systems was developed at the University of Karlsruhe. In this paper, the methods for dead-lock-free coordination and cooperation are explained, further it is described how the architecture can be used in the manufacturing area.

At present, the machine tool technology in the US is not in the state-of-the-art of leading international competitors. Conventional machine tools under use are being pushed into their machining accuracy limits. There is a pressing need calling for revitalizing the machine tool industry. This paper presents, a mechatronic system developed for reducing tool vibration during machining. It consists of electrical and mechanical components, and is realized by placing electrically driven electrostrictive actuators in a specially designed tool post mechanical structure. The linear neural network controller, namely, digital filters, are implemented using a signal processing board. The experimental investigation is conducted in two stages. In the first stage, a test bed is established to use an electro-magnetic shaker to resemble the excitation of cutting force acting on the tool. In the second stage, experiments were conducted using a lathe on the shop floor. In-process vibration cancellation was observed. In the laboratory experiment, a percent reduction in the 90 percent was possible using a feedforward scheme. The improvement in surface roughness during the turning operation was confirmed from measurements of surface roughness profiles.

Fault detection or condition monitoring is critical to the safe and reliable operation of today's complex automatic production systems. This paper describes an approach which performs condition monitoring of production lines based on a max-plus model of the event-time dynamics. In particular, this work pertains to discrete part manufacturing systems such as automotive assembly lines or electronic circuit board lines where the dynamics are governed by the action of certain events, e.g., the processing of a part by a machine or the departure of a part from a buffer. A state model of the event-time dynamics, based in the max-plus algebra, is used to emulate the nominal operation of a production line. The machine completion times are taken as the states of the model. Since only output measurements of the production line are available, an observer is needed to provide estimates of the machine completion times. Because of the nature of the max-plus algebra, a recursive state observer is not obtainable; instead, a block-form state observer is used to update the model state on a periodic basis. A timing residual is formed using the outputs of the model and the output measurements of the plant. The residual is then analyzed using standard statistical process control techniques to detect failures. Simulation results for a simple production line illustrate the approach. The novelties of this approach are the introduction of an event- time observer and the application of the observer to the problem of condition monitoring.

A great deal of research has been done in fuzzy logic control (FLC) and its applications since Mamdani's pioneering papers in 1974 and 1977. FLC has also been applied to manipulator control which is a very challenging nonlinear control problem. Both classical and advanced robot controllers have problems because of high nonlinearity or uncertainties in robot dynamics. FLC, as an alternate, suffer from lack of analytical methods for design, tuning and stability analysis. A nonlinear controller which is robust in the presence of modeling errors and disturbances is presented in this paper. A computed torque controller can be designed based on an approximate model and FLC can be used to minimize the tracking error due to modeling errors and disturbance. Since the approximate model of the system reduces the overall nonlinearity, FLC works with very simple rules and it is easy to tune.

The major problems in the successful implementation of industrial robots are related to the difficulties in programming, resulting in long set-up times, and the lack of integration with other factory automation equipment. Several experts have suggested the need for a standardized robot programming language, that will enable robots of different makes to seamlessly converse and reduce high training costs. Researchers worldwide have tried to develop a standardized robot programming language, but have been only partially successful because the level of variability in the robot programming environment is very high. This paper discusses the need for a high-level, generic icon-based interface to address the issue of standardization. The interface would allow the user to specify the task using a user-friendly interface, which would then be stored as a neutral format. The neutral format file would then be able to call drivers pertaining to different automation equipment within the cell, which in turn, would be able to generate the program for a specific controller to execute the user-defined tasks.

In the paper, a fibre optical torquemeter which is suitable for on-line monitoring the turning moment of a drill turntable is researched. The meter consists of a fibre optical Mach-Zehnder Interferometer, a signal processor and a monitor. The fibre optical Mach-Zehnder Interferometer is formed with a signal and a reference optical fibre arms, two 3dB couplers, a GaAs LD and two photoelectric probers. Both the signal optical fibre arm and reference optical fibre arm are made with polarization maintaining optical fibres. The signal optical fibre arm is made with a shearing elastic tubeshaped object, and a polarization maintaining optical fibre is wound upon it. Both ends ofthe shearing elastic tube-shaped object is fixed upon an axis being measured. The shearing strain of the axis will make the size of the signal optical fibre change, and in this way the phase of the light wave in the optical fibre is modulated between the signal optical fibre arm and the reference optical fibre arm. By measuring the phase change with probers, the shearing strain of the axis is known, and the turning moment is obtained. The signal is processed by MCS-5 1 . The result is shown by the monitor. Key words: Mach-Zehnder Interferometer, Fibre Optical Torquemeter, polarization maintaining optical fibre

This research develops a robot plan for a system that automatically debrides burned tissue on burn victims using a high energy laser for the ablation of the burned tissue. The automated robotic system consists of: a robot whose end effector is equipped with a laser head whence the laser beam emanates and a vision system that is used to acquire the 3D coordinates of some points on the body surface; 3D surface modelling routines for generating the surface model of the treatment area; and control and interface hardware and software for control and integration of all the system components. The entire process of automated burn debridement is achieved in two phases: an initial survey phase during which a model of the treatment area on the skin is built and used to plan an appropriate trajectory for the robot in the subsequent phase--the treatment phase during which the laser surgery is performed. During the survey phase, the vision system acquires points on the surface of the patient's body by using a camera to capture the contour traced by a plane of low power laser light generated by the laser source gut distinct from the high power laser beam. During this phase, a robot plan is generated for moving the end effector along the correct direction so that the camera can capture enough contours needed to build an accurate surface model of the treatment are. During the treatment phase, the surface model developed during the survey phase is used to generate the robot plan for ablating the dead skin tissue. To achieve this, the burned area is first defined on the model. Then based on the shape of the patterns, the trajectory to be followed by the laser head to accomplish complete debridement of the dead tissue is generated. Fourthly, with the point interpolated trajectory necessary for effective treatment obtained, the robot plans the motions necessary for laser ablation of the dead tissue without an over- or under-cut. Accomplishing these steps leads to the generation of a plan for effective dead tissue ablation.

Sequential function charts have been sued to specify control programs for programmable logic controllers. We show how these charts can be used for robotic control using a hierarchical control structure. A process is defined as a set of tasks and for each task a set of actions. Tasks and actions can be executed in serial and parallel with one or more programmable devices. Sequential function charts can provide a device independent method for specifying control programs for robotics systems.

How to reliably deduce life distribution of sensors is one of the practical problems in engineering. The methods currently used either lack accuracy or can not be applied until extensive test data are available. In this paper, a method combining with graphical estimation and parameter estimation has been proposed to solve this problem. Based upon the conventional methods and the theories described in this study, we have developed a computer software to deduce life distribution for engineering applications. A practical example was also given to show how the program performs statistical inference of life distribution model of sensors. By comparing the distributions, an optimum life distribution can be selected and a fit straight line and distribution parameters can be acquired. Key words: sensor, reliability, life disthbution, statistical inference.