Wimalin Laosiritaworn

Artificial Neural Network Modeling of Mean-Field Ising Hysteresis

In this study, the artificial neural network (ANN) was used to model ferromagnetic Ising hysteresis obtained from mean-field analysis as a case study. ANNs were trained to predict the effect of external perturbations, which are the temperature, the field amplitude and the field frequency, on the hysteresis properties, which are the hysteresis area, the remanence magnetization and the coercivity. The input data to the ANN were split into training data, testing data and validating data. Search were carried out to identify number of hidden layer and number of hidden nodes to find the best architecture with highest accuracy. After the networks had been trained, they were used to predict hysteresis properties of the unseen testing patterns of input. The predicted and the actual data were found to match very well over an extensive range. This therefore suggests a success in modeling ferromagnetic hysteresis properties using the ANN technique.

Artificial Neural Network Modeling of Ferroelectric Hysteresis: An Application to Soft Lead Zirconate Titanate Ceramics

In this work, the Artificial Neural Network (ANN) was used to model ferroelectric hysteresis using data measured from soft lead zirconate titanate [Pb (Zr1−xTix)O3 or PZT] ceramics as an application. Data from experiments were split into training, testing and validation dataset. Four ANN models were developed separately to predict output of the hysteresis area, remnant, coercivity and squareness. Each model has two neurons in the input layer, which represent field amplitude and field frequency. The ANNs were trained with varying number of hidden layer and number of neurons in each layer to find the best network architecture with highest accuracy. After the networks have been trained, they were used to predict hysteresis properties of the unseen testing patterns of input. The predicted and the testing data were found to match very well which suggests the ANN success in modeling ferroelectric hysteresis properties obtained from experiments.

Concurrent Artificial Neural Network Modeling of Single-Crystal and Bulk-Ceramics Ferroelectric-Hysteresis: An Application to Barium Titanate

This paper proposed an application of Artificial Neural Network (ANN) to concurrently model ferroelectric hysteresis properties of Barium Titanate in both single-crystal and bulk-ceramics forms. In the ANN modeling, there are 3 inputs, which are type of materials (single or bulk), field amplitude and frequency, and 1 output, which is hysteresis area. Appropriate number of hidden layer and hidden node were achieved through a search of up to 2 layers and 30 neurons in each layer. After ANN had been properly trained, a network with highest accuracy was selected. Query file of unseen input data was then input to the selected network to obtain the predicted hysteresis area. From the results, the target and predicted data were found to match very well. This therefore suggests that ANN can be successfully used to concurrently model ferroelectric hysteresis property even though the considered ferroelectrics are with different domains, grains and microscopic crystal structures.

Modeling of Ferroelectric Hysteresis Area of Hard Lead Zirconate Titanate Ceramics: Artificial Neural Network Approach

In this work, the relationship between hysteresis area of hard lead zirconate titanate and external perturbation was modeled using the Artificial Neural Network (ANN). The model developed has the applied electric field parameters and temperature as inputs, and the hysteresis area as an output. Then ANN was trained with experimental data and used to predict hysteresis area of the unseen testing patterns of input. The predicted and the actual data of the testing set were found to agree very well for all considered input parameters. Furthermore, unlike previous power-law investigation where the low-field data had to be discarded in avoiding non-convergence problem, this work can model the data for the whole range with fine accuracy. This therefore suggests the ANN success in modeling hard ferroelectric hysteresis properties and underlines its superior performance upon typical power-law scaling technique.

Application of artificial neural network to calculate Curie temperature of ferromagnetic materials

Ferromagnetic materials are now interested by researchers as they have applications in various industries. Due to the complexity of the materials, an important contribution to enhance the technological development has come from the theoretical and simulation studies especially from the Monte Carlo simulation. Nevertheless, the Monte Carlo is often limited in its performance because the computational limitations, such as the simulated system sizes and simulation times. These limitations also put a constraint on the simulation time which caps the numerical accuracy. The artificial neural network is used in this study in cooperating with the Monte Carlo simulation. The aim is to investigate the possibility in obtaining the Curie temperature of ferromagnetic Ising spin in a fine scale without an intense computational required. From the results, the extracted Curie temperature is found to agree well with those from the exact theoretical analysis which verifies the artificial neural network to be a very useful technique.

Improving Multi-Panel Lamination Process Optimization using Response Surface Methodology and Neural Network

2k factorial experiment is the most general method of experimental design unusually used for factor screening. It helps to reduce the number of experiment trials by investigating multiple factors at the same time. As each factor in 2k factorial experiment contains 2 levels, it can only use to construct first-degree polynomial model. If there is curvature in the system, other technique such as response surface has to be performed, which leads to additional cost in running more experiments. Instead of conducting actual experimentation, this paper proposes a new method of using neural network to construct a process model with factorial experiment data. This model is used to predict response for response surface experiments. A case study of multi-panel lamination process was used to demonstrate the proposed method. The result showed that optimization result achieved from response surface methodology via neural network model is better than the one from 2k factorial experiments.

The Knowledge-Based Modeling of Ferroelectric Hysteresis Area: An Application to Forming (1-x)PZT-(x)PZN Hysteresis Database

In this work, Artificial Neural Network was used to model the hysteresis behavior of lead zirconate titanate-lead zinc niobate (Pb(Zr1/2Ti1/2)O3−Pb(Zn1/3Nb2/3)O3 or (1-x)PZT-(x)PZN mixed ferroelectric systems. The hysteresis loops were measured with varying electric filed parameters and the composition x of the mixed ferroelectrics. A knowledge-based technique, i.e. the Artificial Neural Network (ANN), was employed in modeling the hysteresis to construct the database of how field parameters and the mixed composition affect dynamic hysteresis behavior. The input data to the ANN were composition x, field amplitude E0 and field frequency f, where the output data was the hysteresis area. The inputs-outputs were divided into training, validating and testing datasets for the ANN. Multilayer perceptron with back propagation training algorithm was applied in this work. Exhaustive search was used to obtain the best network algorithm that gives minimum error in the training process. With the best network, unseen input datasets were fed into the network to predict hysteresis area. From the results, the predicted and the actual data match very well over an extensive range of field parameters, where the scattering plot between the predicted and the actual area has R-squared greater than 0.99. This therefore indicates ANN capabilities in modeling dynamic-hysteresis phenomena across (1-x)PZT-(x)PZN systems even they have different ratios of structural phases at microscopic level.

Coil baking process modeling with neural network

Coil baking process is one of the critical processes in a case study company which is a manufacturer of hard disk drive actuator in Thailand. At the moment, the company is dealing with quality issue arisen from the distortion of the baked coil. Parameters affect the baking process are for example baking temperature, baking time, air ventilation pressure and position in the oven. Neural network technique was used to model the relationship between the mentioned factors and the distortion of the coil. The trained network can be used to predict coil distortion before the production occurs. Therefore, helps to reduce the defect rate.

Die storage improvement with k-means clustering algorithm: A case of paper packaging business

This paper presents die storage improvement for a case study company, who is a manufacturer of made-to-order paper packaging product. One of the critical equipment used to produce paper packaging is the dies used in die cutting machine. These dies are stored in the separate storage room and they are placed on any available shelf slot. Due to the wide variety of product design, number of die stored in die storage room is large and continues to grows every year due to the increasing number of customers. Die storage room has become untidy and packed, which make the die retrieve process become more difficult. K-means clustering, one of the data mining algorithms, was applied to cluster dies into groups based on their size, price and frequency of use. Then the layout of storage room was re-designed based on the new cluster to improve space utilization. After improvement, the time used for die retrieval was significantly reduced.

Optimizing Interface Area of Percolated Domains in Two Dimensional Binary Compound: Artificial Neural Network Modeling on Monte Carlo Experiments

This work investigated the competition effect between the cohesive and adhesive interaction on the domain (of the same atom) profiles with an emphasis on the percolation and interface in two-dimensional binary compound materials. The Ising model was used to represent the two types of the binary compound arranging on two-dimensional square lattice. Monte Carlo simulation and Kawasaki algorithm was used to thermally update the configuration of the system. With varying the system sizes, the environmental temperature, and adhesive to cohesive interaction ratio, the system microstructure were investigated to extract the average number of percolated domains and their interface via pairs of neighboring different-atoms. The results, displayed via the main effect plot, show the average number of percolated domain reduces with increasing the temperature and adhesive interaction strength. However, the normalized interface area decreases in larger system or in stronger adhesive interaction system. Nevertheless, with increasing the temperature, the normalized interface area results in peak where its maximum is due to the thermal fluctuation effect. Since the relationship among these considered dependent and independent parameter is very complex, to establish formalism that can represent this relationship is not trivial. Therefore, the artificial neural network (ANN) was used to create database of relationship among the considered parameters such that the optimized condition in retrieving desired results can be comprehensively set. Good agreement between the real targeted outputs and the predicted outputs from the ANN was found, which confirms the functionality of the artificial neural network on modeling the complex phenomena in this study. This work therefore presents another step in the understanding of how mixed interaction plays its role in binary compound and how a data mining technique assists development of enhanced understanding in materials science and engineering topics.