Training Set For Neural Network Research Articles

Real-time capable first principle based modelling of tokamak turbulent transport

A real-time capable core turbulence tokamak transport model is developed. This model is constructed from the regularized nonlinear regression of quasilinear gyrokinetic transport code output. The regression is performed with a multilayer perceptron neural network. The transport code input for the neural network training set consists of five dimensions, and is limited to adiabatic electrons. The neural network model successfully reproduces transport fluxes predicted by the original quasilinear model, while gaining five orders of magnitude in computation time. The model is implemented in a real-time capable tokamak simulator, and simulates a 300 s ITER discharge in 10 s. This proof-of-principle for regression based transport models anticipates a significant widening of input space dimensionality and physics realism for future training sets. This aims to provide unprecedented computational speed coupled with first-principle based physics for real-time control and integrated modelling applications.

Study on Neural Networks Usage to Analyse Correlation between Spectrum of Vibration Acceleration Signal from Pin of Ball Mill and its Filling Level

The question about modern intelligent information processing methods usage for a ball mill filling level evaluation is considered. Vibration acceleration signal has been measured on a mill laboratory model for that purpose. It is made with accelerometer attached to a mill pin. The conclusion is made that mill filling level can not be measured with the help of such signal amplitude only. So this signal spectrum processed by a neural network is used. A training set for the neural network is formed with the help of spectral analysis methods. Trained neural network is able to find the correlation between mill pin vibration acceleration signal and mill filling level. Test set is formed from the data which is not included into the training set. This set is used in order to evaluate the network ability to evaluate the mill filling degree. The neural network guarantees no more than 7% error in the evaluation of mill filling level.

Алгоритм выбора структурных параметров искусственной нейронной сети и объема обучающей выборки при аппроксимации поведения динамического объекта

Алгоритм выбора структурных параметров искусственной нейронной сети и объема обучающей выборки при аппроксимации поведения динамического объекта

Improving of Procedures for Preparing of Training Set for Neural Networks

In the article procedure of rough-down of information is examined for teaching of neuron networks. Shown, that exists problem of normalization of ordinals of variables in part of their internal levels. The improved chart of normalization, allowing setting ponder ability both ordinal of variable on the whole and its separate levels, is offered to application. Reverse normalization formulas over are also brought for interpretation of gravimetric coefficients of neurons

Recognition of Similar Wooden Surfaces with a Hierarchical Neural Network Structure

The surface quality assurance check is an important task in industrial production of wooden parts. There are many automated systems applying different methods for preprocessing and recognition/classification of surface textures, but in the most cases these methods cannot produce very high recognition accuracy. This paper aims to propose a method for effective recognition of similar wooden surfaces applying simple preprocessing, recognition and classification stage. The method is based on simultaneously training two different neural networks with surface image histograms and their second derivatives. The combined outputs of these networks give an input training set for a third neural network to make the final decision. The proposed method is tested with image samples of seven similar wooden texture images and shows high recognition accuracy. The results are analyzed, discussed and further research tasks are proposed.

PREDETERMINED MOVEMENT OF MOBILE ROBOT USING NEURAL NETWORKS

This paper describes the experimental results of neural networks application for mobile robot control on predetermined trajectory of the road. There is considered the formation process of training sets for neural network, their structure and simulating features. Researches have showed robust mobile robot movement on different parts of the road.

NEURAL-NETWORK BASED METHOD OF CONTROL INFLUENCES FORMING IN COMPUTER SYSTEM CONTROLLING ENVIRONMENTAL PARAMETERS

A simulation model of a section of mine ventilation network is considered in this paper. The simulation modeling of transient aerogasdynamic processes of methane concentration changing is fulfilled at applying position and exponential control influences. There is proposed a neural-based method of control influences forming by neural network training on the set of optimal control influences. There are defined a criterion and developed an algorithm of optimal control influences forming as a training set of neural network. The simulation modeling of applying of control influences formed by neural network is fulfilled and decreasing of control parameter in the section of mine ventilation network is estimated.

SIMULATION MODELING OF NEURAL CONTROL SYSTEM FOR SECTION OF MINE VENTILATION NETWORK

Static and dynamic simulation models of a section of a mine ventilation network in order to research a sequential neural control scheme of mine airflow are developed in this paper. The techniques of neural network training set creation for both simulation models, a structure of neural network and its training algorithm are described. The simulation modeling results using static and dynamic models have showed good potential capabilities of neural control approach.

Application of an artificial neural network and morphing techniques in the redesign of dysplastic trochlea.

Segmentation and computer assisted design tools have the potential to test the validity of simulated surgical procedures, e.g., trochleoplasty. A repeatable measurement method for three dimensional femur models that enables quantification of knee parameters of the distal femur is presented. Fifteen healthy knees are analysed using the method to provide a training set for an artificial neural network. The aim is to use this artificial neural network for the prediction of parameter values that describe the shape of a normal trochlear groove geometry. This is achieved by feeding the artificial neural network with the unaffected parameters of a dysplastic knee. Four dysplastic knees (Type A through D) are virtually redesigned by way of morphing the groove geometries based on the suggested shape from the artificial neural network. Each of the four resulting shapes is analysed and compared to its initial dysplastic shape in terms of three anteroposterior dimensions: lateral, central and medial. For the four knees the trochlear depth is increased, the ventral trochlear prominence reduced and the sulcus angle corrected to within published normal ranges. The results show a lateral facet elevation inadequate, with a sulcus deepening or a depression trochleoplasty more beneficial to correct trochlear dysplasia.

Designing neural networks to improve timing performances of intelligent controllers

Intelligent techniques are applied to improve the control methods of physical quantities. Many researchers employ the fuzzy logic combined with genetic procedures to achieve good control results. Such intelligent techniques can be enhanced using Neural Networks. In this paper, an optimization algorithm to define the best training set for suitable Neural Networks is designed. The network trained with the optimal sample is introduced into genetic-fuzzy controllers to improve the timing performances. The results show that the genetic-neuro-fuzzy controllers are better than genetic-fuzzy controllers in terms of settling time and rise time.

Closed Form Solution for Deflection of Flexible Composite Bridges

Simply supported steel-concrete composite beams are widely used in bridge construction. Deflection is the significant parameter for serviceability limit state of bridges. A lot of computational effort is required for finite element analysis of bridges considering flexibility of shear connectors and shear leg effects. Neural network is presented for prediction of deflections, at service load, in simply supported steel-concrete composite bridges incorporating flexibility of shear connectors and shear lag effect. The training, testing and validation data sets for neural network are generated using finite element models. The finite element models have been developed using ABAQUS software. These models have been validated with available experimental results. Closed form solution is also proposed based on the developed neural network. The use of the neural network requires a computational effort almost equal to that required for the simple beam analysis (neglecting flexibility of shear connectors and shear lag effect). The neural network has been validated for number of bridges and the errors are found to be small. The network/closed form solution can be used for rapid prediction of deflection for everyday design.

DATA PROCESSING USING ARTIFICIAL NEURAL NETWORKS TO IMPROVE THE SIMULATION OF LUNG MOTION

To optimize the delivery in lung radiation therapy, a better understanding of the tumor motion is required, on one hand, to have a better tumor-targeting efficiency, and on the other hand to avoid as much as possible normal tissues. The four-dimensional computed tomography (4D-CT) allows to quantify tumor motion, but due to artifacts, it introduces biases and errors in tumor localization. Despite this disadvantage, we propose a method to simulate lung motion based on data provided by the 4D-CT for several patients. To reduce uncertainties introduced by the 4D-CT scan, we conveniently treated data using artificial neural networks. More precisely, our approach consists of a data augmentation technique. The data resulting from this processing step are then used to build a training set for another artificial neural network that learns the lung motion. To improve the learning accuracy, we have studied the number of phases required to precisely describe the displacement of each point. Thus, from 1118 points scattered across five patients and defined over 8 or 10 phases, we obtained 5800 points from 50 phases. After training, the network is used to compute the positions of 40 points from five other patients on 10 phases. These points allow to quantify the prediction performance. In comparison with the original data, the ones issued from our treatment process provide a significant increase of the prediction accuracy: an average improvement of 16% can be observed. The motion computed for several points by the neural network that has learnt the lung one exhibits an hysteresis near the one given by the 4D-CT, with an error smaller than 1 mm in the cranio-caudal axis.

Investigation of pool boiling of nanofluids using artificial neural networks and correlation development techniques

The nucleate pool boiling heat transfer characteristics of TiO2 nanofluids are investigated to determine the important parameters' effects on the heat transfer coefficient and also to have reliable empirical correlations based on the neural network analysis. Nanofluids with various concentrations of 0.0001, 0.0005, 0.005, and 0.01vol.% are employed. The horizontal circular test plate, made from copper with different roughness values of 0.2, 2.5 and 4μm, is used as a heating surface. The artificial neural network (ANN) training sets have the experimental data of nucleate pool boiling tests, including temperature differences between the temperatures of the average heater surface and the liquid saturation from 5.8 to 25.21K, heat fluxes from 28.14 to 948.03kWm−2. The pool boiling heat transfer coefficient is calculated using the measured results such as current, voltage, and temperatures from the experiments. Input of the ANNs are the 8 numbers of dimensional and dimensionless values of the test section, such as thermal conductivity, particle size, physical properties of the fluid, surface roughness, concentration rate of nanoparticles and wall superheating, while the outputs of the ANNs are the heat flux and experimental pool boiling heat transfer coefficient from the analysis. The nucleate pool boiling heat transfer characteristics of TiO2 nanofluids are modeled to decide the best approach, using several ANN methods such as multi-layer perceptron (MLP), generalized regression neural network (GRNN) and radial basis networks (RBF). Elimination process of the ANN methods is performed together with the copper and aluminum test sections by means of a 4-fold cross validation algorithm. The ANNs performances are measured by mean relative error criteria with the use of unknown test sets. The performance of the method of MLP with 10-20-1 architecture, GRNN with the spread coefficient 0.7 and RBFs with the spread coefficient of 1000 and a hidden layer neuron number of 80 are found to be in good agreement, predicting the experimental pool boiling heat transfer coefficient with deviations within the range of ±5% for all tested conditions. Dependency of output of the ANNs from input values is investigated and new ANN based heat transfer coefficient correlations are developed, taking into account the input parameters of ANNs in the paper.

IMPROVED SORTING METHODOLOGY OF DATA-PROCESSING INSTRUCTIONS

An improved classification methodology for sorting data-processing instructions for ARM7TDMI CPU core is presented in this paper. Main discussion here is related to the process of creating appropriate training sets for neural network (NN) based estimation of power consumption. We have proposed instructions’ sorting methodology according to the binary instruction representation and the resources being used for the overall system model. Thus separate instructions groups are obtained for NN-based estimation of power consumption. Experimental results of the proposed method confirm successful usage of this sorting methodology for providing higher accuracy estimation of power consumption.

Partial discharge recognition system for current transformer using neural network and 2D wavelet transform

AbstractIn this paper, a neural network and two‐dimensional (2D) wavelet transform are applied to recognize partial discharge (PD) patterns on current transformers (CTs). To avoid the discrepancy between simulated results and real experimental data, we adopted seven cast‐resin CTs that were purposely fabricated with various insulation defects as the PD patterns collected samples to actually emulate the various defects incurred often during their production. All measurements are taken in a shielded lab; the commercial TE571 PD detector is adopted to measure PD patterns to ensure the reliability of the PD signals. Next, we extract the patterns' features via a 2D wavelet transform and use the features as the training set of a backpropagation neural network (BNN) to construct the recognition system for CTs' PD patterns. Finally, we add random noises to the measured PD signals to emulate the field diagnosis under a high‐noise environment. The study results indicate that, under a simulated noise magnitude of 30 pC, the recognition rate of the proposed system still can reach around 80%, signifying a great potential in applying the proposed recognition system in field measurements in the future. © 2011 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Modelling of lead adsorption from industrial sludge leachate on red mud by using RSM and ANN

In this study, response surface methodology (RSM) and artificial neural network (ANN) were employed to develop prediction models for lead removal from industrial sludge leachate using red mud. The leaching characteristics of industrial sludge were observed by Toxicity Characteristics Leaching Procedure (TCLP). Dosage, time and pH were considered as independent experimental factors. Box–Behnken design (BBD) was chosen for the response surface design setup and was also used as Neural Network Training Set for comparison purposes. To evaluate the accuracy of results, several experiments were then conducted. The results of ANN were found to be more reliable than RSM since better statistical parameters were obtained.

Multi-Angle Face Detection Using Back Propagation Neural Network

In machine vision application, the main part to analyze an image is to identify its features which contribute to efficiency of the system. Many applications in vision system and image analysis used face detection as a feature of their whole system development. In application such as video surveillance, fatigue detection and security system, face is a fundamental step in the analysis before proceed to system implementation. It is very challenging to recognize a face from an image due to the wide variety of face and the uncertain of face position. In this paper, we propose a neural network based approach to identify multi-angle face which falls into five categories: all left-side face, half left-side face, positive face, half right-side face, and all right-side face. More than 100 images of each category have been used for training and testing of face detection and its features was extracted to be an input to BP neural network. We analyzed the result of training and testing set of neural network and the best classification achieved was 90.7%.

A parallel neural network approach to prediction of Parkinson’s Disease

Recently the neural network based diagnosis of medical diseases has taken a great deal of attention. In this paper a parallel feed-forward neural network structure is used in the prediction of Parkinson’s Disease. The main idea of this paper is using more than a unique neural network to reduce the possibility of decision with error. The output of each neural network is evaluated by using a rule-based system for the final decision. Another important point in this paper is that during the training process, unlearned data of each neural network is collected and used in the training set of the next neural network. The designed parallel network system significantly increased the robustness of the prediction. A set of nine parallel neural networks yielded an improvement of 8.4% on the prediction of Parkinson’s Disease compared to a single unique network. Furthermore, it is demonstrated that the designed system, to some extent, deals with the problems of imbalanced data sets.

Framework for Automatic Development of Type 2 Fuzzy, Neuro and Neuro-Fuzzy Systems

This paper presents the design and development of generic framework which aids creation of fuzzy, neural network and neuro fuzzy systems to provide expert advice in various fields. The proposed framework is based on neuro fuzzy hybridization. Artificial neural network of the framework aids learning and fuzzy part helps in providing logical reasoning for making proper decisions based on inference of domain expert’s knowledge. Hence by hybridizing neural network and fuzzy logic we obtain advantages of both the fields. Further the framework considered type 2 fuzzy logic for more human like approach. Developing a neuro fuzzy advisory system is tedious and complex task. Much of the time is wasted in developing computational logic and hybridizing the two methodologies. In order to generate a neuro fuzzy advisory system quickly and efficiently, we have designed a generic framework that will generate the advisory system. The resulting advisory system for the given domain is interactive with its user and asks question to generate fuzzy rules. The system also allows provision of training sets for neural network by its users in order to train the neural network. The paper also describes a working prototype implemented based on the designed framework; which can create a fuzzy system, a neural network system or a hybrid neuro fuzzy system according to information provided. The working of the prototype is also discussed with outputs in order to develop a fuzzy system, a neural network system and a hybrid neuro fuzzy system for a domain of course selection advisory. The generated systems through this prototype can be used on web or on desktop as per the user requirement.

Determination of Steel Carburizing Parameters by Using Neural Network

The article discusses the application of neural networks for calculating the laws of complex processes, with carbon diffusion processes in steel carburizing classified among them. Empirical and mathematical models for the Carbomaag carburizing process, which have been proven in practice, are presented for the determination of technological carburizing parameters resulting in a required carbon concentration profile in the carburized layer. A comparison between the mathematical model (MM) and the neural network model (NNM) and the empirical model < EM) with respect to the time required for shallow and great carburizing depths is given special attention in this article. The results of the empirical carburizing model were used as a neural network training set and compared with the results of computer simulation of the MM. A comparison of results obtained at shallow and great carburizing depths shows that the NNM approximates much better the EM than the MM. It is assumed that the carrying out of a larger number of experiments and the repeated neural network training with new sets of experimental data would result in increasingly better solutions. Thus, the disadvantages of theoretical models, i.e., MMs, would be avoided.