JOURNAL OF SCHOOL OF ENGINEERING
Year:2005 | Volume:17 | Issue:2|Papers Count:12

This paper presents a design and the associated solution of a cellular manufacturing (CM) problem modeled in dynamic and stochastic conditions. Actually, cellular manufacturing systems (CMSs) are the implementation of group technology (GT) in manufacturing and Production systems aiming at classifying parts based on the physical or operational similarities of the parts in different aspects of their design and production. A number of previous researches in CMSs have been taken under the condition of static production and deterministic demand. However, the production is dynamic and the demand for parts is stochastic in a real-world situation. Due to the adjustment of a CMS in real conditions, it needs a grate number of variables and constraints in the model to be increased. Besides, the model consequently needs a lot of time by using classic optimization methods. However, because of model complexity, it cannot obtain global optimum in most of time. That is why modern techniques and methods such as simulated annealing (SA) are of great fame and attention nowadays. SA is a part of common techniques by searching at random in a controlled manner in order to solve such a CM problem known as a NP complete. In this paper, a nonlinear integer model of CM is first presented and then solved by SA. Finally, the computational results are compared with the optimal solution obtained by Lingo. 

In this paper the morphological umbra sampling scheme is introduced. In addition, a new morphological pyramid is proposed that provides a higher performance than other pyramidal schemes. The high performance of the proposed pyramid is due to: 1- it is applied not only to spatial dimensions, x and y, but also to the gray levels, 2- in the new scheme, the error signal (image) is non-negative. Results of experiments done on a variety of images have validated the proposed method.

To develop a Text-To-Speech (TTS) system in Iranian Farsi language, in this paper three subsystems are designed and implemented. The first subsystem, called NLP (Natural Language Processor), consists of a Part-Of Speech (POS) tagging unit, a text to phoneme converter, and a module for disambiguation of homographs. The second subsystem is a prosody generator that uses a Recurrent Neural Network (RNN) with 289 nodes implemented in four layers. A concatenative speech synthesizer using Harmonic plus Noise Model (HNM) with new approaches in prosody modification is implemented as the third subsystem. To provide training data for prosody generator more efficiently, a couple of novel and hybrid algorithms are used for automated segmentation and labeling of speech at phoneme level with 97.2% accuracy. To evaluate the performance of system, rating scales recommended in ITU-T P.85 are used and average MOS (over six scales) of 3.59 was reached. This MOS shows that the performance of this Farsi ITS is comparable with modern English ITS systems.

In this paper, we consider robust regulation of a class of nonlinear systems, via H¥, controller using slow-fast decomposition. First, using normal form equations, we eliminate the nonlinear part of the system matrix of equations of system and transform it to a linear diagonal form. Separating new equations to slow and fast subsystems, due to the two-time scale property of system, and with the assumption of norm-boundedness of the fast dynamics, we have treated them as disturbance and design H¥, controller for nominal system with a lower order than the original one that stabilizes the overall closed loop system. The proposed method is applied to a single link, flexible joint robot manipulator.

In this paper, a method for online recognition of isolated Farsi characters is presented. These characters are divided into 12 groups based on their dots and signs. First, the dots or other signs of the input character are recognized and the related group to that character is chosen. If there is only one member in that group, its class is assigned to the input character, otherwise, the body of input character is compared to those of the group members and a member with minimum distance to it is found. The class of this member is assigned to the input character. The proposed method was tested on a database of 4144 characters from different writers. The correct recognition rate was 93.3%. If the characters are written in a slightly constraint style, the correct recognition rate will reach to 98%. For this purpose, the writer can adapt its writing style with the system requirements.

Using the Faraday's Law of electromagnet induction, the electromagnetic flowmeters are use to measure the industrial process flow rate of the fluids. In these devices, the windings around the pipe are designed to make the required magnetic field and electrodes which are mounted on two sides of the pipe wall, are used to measure the induced voltage in proportion with the flowing liquid rate. This paper presents the design and optimization of electromagnetic flowmeter for conductive liquids. In this respect, a two dimensional mathematical model by Finite Difference (FD) numerical solution approach is used for calculation of electric potential difference between the electrodes. The basic concepts of the electromagnetic flowmeters design and simulation have been presented using m-file programming in Matlab software. Then, with respect to the fact that fluid flow depends on two variables: liquid level and conductivity coefficient of liquid and pipe bed, a three layer neural network is used for an accurate calibration of electromagnetic flowmeter. In this new approach, for a circular cross-section pipe, the correction factor, used for the calibration, is accurately estimated finally, simulation results are provided to show the accuracy of the applied technique in this paper.