Journal of Advanced Materials in Engineering (Esteghlal)

---

Routing is one of the basic parts of a message passing multiprocessor system. The routing procedure has a great impact on the efficiency of a system. Neural algorithms that are currently in use for computer networks require a large number of neurons. If a specific topology of a multiprocessor network is considered, the number of neurons can be reduced. In this paper a new recurrent neural network and its energy function is introduced. This network requires a significantly smaller number of neurons compared to its counterparts. Also presented is the performance of this neural network. Keywords: Routing, Multicomputer systems, Recurrent neural networks, Mesh interconnection networks.

---

Finding the collapse susceptible portion of a power system is one of the purposes of voltage stability analysis. This part which is a voltage control area is called the voltage weak area. Determining the weak area and adjecent voltage control areas has special importance in the improvement of voltage stability. Designing an on-line corrective control requires the voltage weak area to be determined by a sufficiently rapid and precise method. In this paper, a new algorithm based on assigning a vector to each power system bus is presented. These vectors indicate buses conditions from the viewpoint of voltage stability. In this new method, using the clustering methods such as kohonen neural network, fuzzy C-Means algorithm and fuzzy kohonen algorithm, voltage control areas are determined The proposed method has advantages such as determining PV and PQ buses which belong to the weak area simultanously, under all operating conditions and without a need to system model. Also by comparing the results of applying clustering methods, it has been observed that, due to simplicity of implementation and precision of the results, the two dimensional kohonen neural network is a more suitable tool for clustering power system to voltage control areas than the fuzzy C-Means and fuzzy kohonen methods. Keywords: Voltage stability, Voltage weak area, Voltage control area, Corrective control, Pattern recognition, Kohonen neural network, Fuzzy C-Means algorithm, Fuzzy Kohonen algorithm.

---

In this research, behaviour of clayey soils under triaxial loading is studied using Neural Network. The models have been prepared to predict the stress-strain behaviour of remolded clays under undrained condition. The advantage of the model developed is that simple parameters such as physical characteristics of soils like water content, fine content, Atterberg limits and so on, are used to model the stress-strain behaviour of clays under triaxial loading, without performing exact and time-consuming tests on samples. Results from the network show that neural network is a good tool for prediction of stress-strain behaviour of clayey soils using simple physical characteristics of such soils

---

In this study, colorimetric calibration of scanner has been done via perceptron neural network with three or four layers by back propagation algorithm for colored polyester fabrics. The results obtained for random training samples are not satisfactory but application of selective training samples for L*a*b* or RGB leads to good results, with better results obtained for the L*a*b* method. On the other hand, the color differences between calculation XYZ and real XYZ for unknown samples, are not only in agreement with the results of polynomials and regression methods, but are also better than the results obtained in previous studies where neural networkhad been used for colorimetric calibration of scanner.

---

Artificial Neural Networks are information processing systems. Over the past several years, these algorithms have received much attention for their applications in pattern completing, pattern matching and classification and also for their use as a tool in various areas of problem solving. In this work, an Artificial Neural Network model is presented for predicting the tensile properties of cotton-covered nylon core yarns. Multilayer Feedforward network with Back Propagation learning algorithm was used to study the relationship and mapping among the process parameters, i.e. count of sheath part, count of core part, applying pretension to the core part, inserted twist to the core spun-yarn as well as tensile properties, i.e. breaking strength and breaking elongation. The results show that ANN is an effective method for the prediction of the tensile properties of these yarns. This is due to the fact that in each case, standard deviation of prediction error for test and train data was less than that obtained from the expreiments.

---

This paper presents a method based on neural networks to detect broken rotor bars and end rings in squirrel cage induction motors. In the first part, detection methods are reviewed and traditional methods of fault detection as well as dynamic model of induction motors are introduced using the winding function method. In this method, all stator and rotor bars are considered independently in order to check the performance of the motor for any faults in the parts. Then the frequency spectrum of current signals is derived using the Fourier transform and analyzed under various conditions. In the second part of the paper, an analytical discussion of the theoretical principles is presented to arrive at a simple algorithm for fault detection based on neural networks. The neural network has been trained using the information from a 1.1 KW induction motor. Finally, the system is tested with different values of load torque and is found capable of working on-line to detect all normal and ill-performing conditions.

---

In welded tubular joints, when the fatigue crack depth is less than 20% of chord wall thickness, the crack growing process is highly affected by weld geometry. Hence, T-butt solution and weld magnification factor (Mk) are applicable tools for evaluating the crack growth rate in this domain. In this research, the capability of Artificial Neural Network (ANN) for estimating the Mk of weld toe cracks in T-butt joints is investigated. Four Multi-Layer Perceptron (MLP) networks are designed and trained to predict the Mk in deepest point and ends of weld toe cracks under membrane and bending stresses. Training and testing data of networks are extracted from a reputable resource on finite element modeling. Comparison of the results obtained and those from the most recently published equations shows that using ANN seems to be very beneficial in this field

---

Prediction of SCC risk of austenitic stainless steels in aqueous chloride solution and estimation of the time to failure as a result of SCC form important and complicated topics for study. Despite the many studies reported in the literature, a formulation or a reliable method for the prediction of time to failure as a result of SCC is yet to be developed. This paper is an effort to investigate the capability of artificial neural network in estimatiing the time to failure for SCC of 304 stainless steel in aqueous chloride solution and to provide a sensitivity analysis thereof. The input parameters considered are temperature, chloride ion concentration, and applied stress. The time to failure is defined as the output parameter and the key criterion to evaluate the effective parameters. The statistical performance of the neural network is expressed as the average of three learning and testing results. The SCC database is divided into two sections designated as the learning set and the testing set. The output results show that artificial neural network can predict the time to failure for about 74% of the variance of SCC experimental data. Furthermore, the sensitivity analysis also exhibits the effects of input parameters on SCC of 304 stainless steel in aqueous chloride solutions.

---

In cast aluminum and its alloys, the microstructure varies under different solidification conditions, causing variations in their mechanical properties. These materials are basically produced in sand and metallic molds or through die casting, each of which is associated with a unique solidification regime with significantly different cooling rates so that the resulting microstructure strongly depends on the casting method used. In the present study, the effects of such important solidification parameters as cooling rate, solidification front velocity, and thermal gradient at the solid-liquid interface on secondary dendrite arm spacing were investigated. By a directional solidification system, the mathematical relation between cooling rate and dendrite spacing was extracted for several commercially important aluminum alloys. A neural network model was trained using the experimental values of cooling rates and secondary dendrite arm spacing. Reliable prediction of these values was made from the trained network and their corresponding diagrams were constructed. A good agreement was found between simulation and experimental values. It is concluded that the neural network constructed in this study can be employed to predict the relationship between cooling rate and dendrite arm spacing, which is difficult, if not iompossible, to accomplish experimentally.

---

Mechanical alloying technique is used for production of nanostructured soft magnetic alloys. In this work the back propagation (BP) artificial neural adopted to model the effect of various mechanical alloying parameters i.e. milling time and chemical composition, on the properties of Fe-Ni powders. Lattice parameter, grain size, lattice strain, coersivity and saturation intrinsic flux density are considered as the output of five BP neural networks. The results obtained show the efficiency of designed networks for the prediction of the properties of Fe-Ni powders.