João Sentieiro

Production planning and scheduling using a fuzzy decision system

In this paper short-range planning and scheduling problems are addressed using a nonclassical approach supported by fuzzy theory. The proposed methodology uses a hierarchical structure which includes three decision levels (higher, middle, lower), each responsible for a different production problem with a different time scale. The methodology approaches the tasks associated with each level using a heuristic formulation and solves the short-range planning and scheduling problems with a nonstationary policy. The higher decision level determines safety stock levels used to compensate for future resource failures. At the middle level, loading rates are computed. This is accomplished through a fuzzy controller that tends to minimize the error between the cumulative production and the cumulative demand while keeping the work in process below acceptable values. Finally, the lower level controls the flow of parts among the resources, using a modified version of the Yagers fuzzy decision method. This method has the ability to use several criteria to generate a decision. Simulation results reveal that the proposed system exhibits good performance, in terms of a high production percentage and a low WIP, under resource failures and demand variations. >

Selection of controller parameters using genetic algorithms

In this paper the problem of setting the initial values for tho parameters of a class of controllers is approached using genetic algorithms. The controller parameters, computed off-line via the optimization of a quadratic error performance index, can be used to initialize the controller incorporated in the real plant. During the on-line control stage and to increase the overall performance, other tuning mechanisms can be considered [5].

The role of vision for underwater vehicles

Different sensing techniques have been used for a long time in the development of remote operated vehicles (ROVs) and, in the recent years, autonomous underwater vehicles, (AUVs). Sonar has been the most popular choice in what concerns depth information or obstacle detection in most of the existing systems. Vision, however is a powerful sensing modality and could be used in many tasks where accurate measurements at a short range, are needed. This paper discusses the possible use of vision in the context of autonomous underwater vehicles. Tasks, such as object avoidance or recognition, grasping, docking, sea bed reconstruction, underwater surveillance, inspection, cable maintenance, are among the set of those where computer vision may have an important role. As an example, the authors present a vision system designed for recursive depth estimation based on a underwater vehicle equipped with a camera. Results obtained with underwater images are presented.

Adaptive scheduling for high-volume shops

A conceptual algorithm for NP-hard scheduling problems that gives accurate reactive scheduling for high-volume shops is presented. The heuristic algorithm provides classical dispatch scheduling with an embedded optimization procedure. It delivers successive updates of the initially generated schedule by extending in an unconventional way the horizon of applicability. An adaptive mechanism is implemented. The successive schedule updates are generated using beam search. The conceptual algorithm is applied to problems where a periodic production pattern is assumed. Analysis of the algorithms behavior reveals that the performance measure converges to the calculated lower bounds and that the computational time can be polynomial, for each problem, in the number of parts to process. >

A computer vision system for the characterization and classification of flames in glass furnaces

It is pointed out the characterization and classification of certain phenomena that occur inside glass furnaces is of the utmost importance for the design of efficient and energy-saving control strategies. The authors describe a computer vision system dealing with both flame analysis and classification problems. The system was first tested in a laboratory environment where different operating conditions were created through the variation of the burner geometry, the number of active burners, and the rate of fuel used. Two classifiers, one based on a Bayesian formalism and the other on neural networks, were designed and tuned in a laboratory environment. In a second phase, images from an industrial furnace were acquired, several flame classes were defined, and the classifiers were tested. The results obtained by both the Bayesian and the neural network classifiers are quite encouraging and the success rates are similar for both of them.<<ETX>>

Generation of 3D Dense Depth Maps by Dynamic Vision

This paper presents a dynamic 3D Vision system that is able to estimate dense depth maps from an image sequence. The depth maps computed at each time instant are used in an Extended Kalman filtering structure, that integrates all depth measurements over time, reducing uncertainty. Results with images acquired by an underwater camera, are presented.

A 3D Vision System for Underwater Vehicles: An Extended Kalman-Bucy Filtering Approach

This paper presents a dynamic 3D Vision system that is able to estimate dense depth maps from an image sequence. The camera motion between two successive time instants is assumed to be known. The depth maps computed at each time instant are used in an Extended Kalman-Bucy filtering structure, that integrates all depth measurements over time, reducing uncertainty. Results on images acquired by an underwater camera are presented.

An architecture for the supervision of fuzzy controllers

A new approach to the supervision of fuzzy controllers is presented. The supervision loop is designed to overcome some of the problems that remain unsolved, such as time-varying plants, high nonlinear plants or fine tuning of the linguistic terms given by the expert in a fuzzy controller. The supervision concept is based on the continuous adjustment of the mathematical functions used for the definition of the linguistic terms which describe the actions of the rules. The amount of adjustment is the result of a weighted combination of the results using two features observed in the control system output-the rise time and the overshoot.<<ETX>>

Planning and Scheduling: Non-Classic Heuristic Approaches

Abstract Present day production systems are very complex involving many and different activities, which are interconnected aiming towards a common objective: the production of finished goods. Examples of some of such activities are production planning and scheduling. Due to the NP-hard nature of Scheduling problems, the use of classical approaches or search techniques in the generation of a complete schedule is a heavy task. This paper will focus on non-conventional heuristic approaches which, in recent years, have been the concern of the author and his research group at ISR. Several methods are briefly described and commented

Properties of a feasible direction algorithm which finds search directions by approximately solving a subproblem

This paper is concerned mostly with feasible direction methods for which the search direction is obtained by solving approximately a subproblem defined at iteration j by: (SP)j min{fs(zj, y): y?Q} where fs(zj, y) is a lower approximation to the objective function f(y5) and Q is not describable in the conventional form, Q = {x?Rn : g(x) ? 0, h(x) = 0}, so most of the existing optimization algorithms are not applicable. An existing algorithm obtains search directions by solving exactly (SP)j. However exact solution of the subproblem is not possible in many cases regardless of their simple nature. The effects on convergence rate of approximate minimization of fs are investigated and new results derived. Application to a classical optimal control problem is considered.