Developed Neural Network Model Research Articles

A neural network based estimation method for magnetic shielding at extremely low frequencies

The attenuation of extremely low-frequency magnetic fields is important in reducing electromagnetic interference on electric and electronic equipment. In this paper, an innovative method is presented for shielded magnetic field level estimation at power frequencies by a neural network (NN) technique which uses experimental data. The utilized NN is applied to cylindrical shields (transformer-grade iron, copper, and aluminum) in various shield arrangements. Using the developed NN model, the mitigated magnetic field of multilayered shields is measured and evaluated to predict the magnetic field at any distance apart from the magnetic source. The NN, which is based on a feed-forward neural network (FNN), is trained with scaled conjugate gradient, gradient descent with momentum and adaptive learning back propagation, and Levenberg–Marquardt algorithms to compute the shielded magnetic field. Results have shown that the developed FNN trained with the Levenberg–Marquardt algorithm is better than the other training algorithms in predicting the shielded magnetic field value accurately even in the presence of various shield arrangements.

The variability and predictability of the IRI B0, B1 parameters over Grahamstown, South Africa

The International Reference Ionosphere (IRI) parameters B 0 and B 1 provide a representation of the thickness and shape, respectively, of the F2 layer of the bottomside ionosphere. These parameters can be derived from electron density profiles that are determined from vertical incidence ionograms. This paper aims to illustrate the variability of these parameters for a single mid latitude station and demonstrate the ability of the Neural Network (NN) modeling technique for developing a predictive model for these parameters. Grahamstown, South Africa (33.3°S, 26.5°E) was chosen as the mid latitude station used in this study and the B 0 and B 1 parameters for an 11 year period were determined from electron density profiles recorded at that station with a University of Massachusetts Lowell Center for Atmospheric Research (UMLCAR) Digisonde. A preliminary single station NN model was then developed using the Grahamstown data from 1996 to 2005 as a training database, and input parameters known to affect the behaviour of the F2 layer, such as day number, hour, solar and magnetic indices. An analysis of the diurnal, seasonal and solar variations of these parameters was undertaken for the years 2000, 2005 and 2006 using hourly monthly median values. Comparisons between the values derived from measured data and those predicted using the two available IRI-2001 methods (IRI tables and Gulyaeva, T. Progress in ionospheric informatics based on electron density profile analysis of ionograms. Adv. Space Res. 7(6), 39–48, 1987.) and the newly developed NN model are also shown in this paper. The preliminary NN model showed that it is feasible to use the NN technique to develop a prediction tool for the IRI thickness and shape parameters and first results from this model reveal that for the mid latitude location used in this study the NN model provides a more accurate prediction than the current IRI model options.

Modeling viscosity behavior of crumb rubber modified binders

Accurately predicting the viscous properties of crumb rubber modified (CRM) binders has proven difficult, especially as these properties tend to vary with changing crumb rubber concentrations and temperatures. This study explores the utilization of the statistical regression and neural network (NN) approaches in predicting the viscosity values of CRM binder at various temperatures (135 °C and greater). A total of 53 CRM binder combinations were prepared from two different rubber types (ambient and cryogenic), three different binder sources, four rubber concentrations (0%, 5%, 10%, and 15%), and five crumb rubber gradations (ADOT, SCDOT, 0.18 mm, 0.425 mm, and 0.85 mm). The results indicated that the regression model is easy to use and can be used for viscosity prediction, similarly NN-based models also provided accurate for predictions for the viscosity values of CRM binders regardless of rubber type and can easily be implemented in a spreadsheet. In addition, the developed NN model can be used to predict viscosity values of other types of CRM binders efficiently. Furthermore, the sensitivity analysis of input variables indicated that the changes of viscosity are significant as the changes of asphalt binder grade, test temperature, and rubber content. The results also show that these three independent variables are the most important factors in the developed NN models in comparison with other variables.

Failure Pressure Prediction of Composite Cylinders for Hydrogen Storage Using Thermo-mechanical Analysis and Neural Network

Safe installation and operation of high-pressure composite cylinders for hydrogen storage are of primary concern. It is unavoidable for the cylinders to experience temperature variation and significant thermal input during service. The maximum failure pressure that the cylinder can sustain is affected due to the dependence of composite material properties on temperature and complexity of cylinder design. Most of the analysis reported for high-pressure composite cylinders is based on simplifying assumptions and does not account for complexities like thermo-mechanical behavior and temperature dependent material properties. In the present work, a comprehensive finite element simulation tool for the design of hydrogen storage cylinder system is developed. The structural response of the cylinder is analyzed using laminated shell theory accounting for transverse shear deformation and geometric nonlinearity. A composite failure model is used to evaluate the failure pressure under various thermo-mechanical loadings. A back-propagation neural network (NNk) model is developed to predict the maximum failure pressure using the analysis results. The failure pressures predicted from NNk model are compared with those from test cases. The developed NNk model is capable of predicting the failure pressure for any given loading condition.

Predicting the thermal behaviour of engine oils using artificial neural networks

A multi-layer neural network (NN) was developed to analyse experimental boiling data obtained for five engine oils at bulk temperatures ranging from 40 to 150 °C and heat fluxes ranging from 30 to 400 kW/m2. The inputs to the NN were the oil chemical composition (nine elements) along with the wall superheats for different oil bulk temperatures and the NN output was the corresponding heat fluxes. The developed NN model predicted boiling curves that are in close agreement with experimental data ( R2≈ 1). The NN was then trained on the experimental data of four of these oils and allowed to predict the boiling behaviour of the fifth. The NN, therefore, demonstrated its ability to predict the boiling characteristics of untested oils, provided their chemical compositions fall within the range of the training data. In addition, for oils with constituents outside the considered range, the NN was able to predict their boiling characteristics at all bulk temperatures when trained on a sample of experimental measurement. The success of such a technique provides researchers and developers of oils with the thermal performance of oils (based on chemical constituents) without testing or with very limited measurements. This, as a result, has the advantages of saving experimental time and cost of oil testing.

Genetic algorithm-based neural network learning parameter selection for ore grade evaluation of limestone deposit

Ore grade evaluation is a decisive task in mine planning operations. A reliable estimate can represent the deposit close to reality. A neural network (NN) model was used to develop a reliable estimate of ore grades of a study mine. The learning parameters for the NN model were selected by genetic algorithm. The optimum hidden layer nodes were selected based on minimum mean square error of a validation data set. The performance of the model was tested and compared with an ordinary kriging model. The comparative result revealed that the NN model performed better for all four attributes of the case study mine than the ordinary kriging model. The developed NN model incorporates lithological information as input parameters. Therefore, lithological maps of the deposit were generated using sequential indicator simulation. Ten different ore grade maps were generated for calculation of the uncertainty associated with the model. The grade tonnage curve along with uncertainty will help mine management by improving understanding about the deposit.

Sensitivity analysis of neural network parameters to improve the performance of electricity price forecasting

This paper presents a sensitivity analysis of neural network (NN) parameters to improve the performance of electricity price forecasting. The presented work is an extended version of previous works done by authors to integrate NN and similar days (SD) method for predicting electricity prices. Focus here is on sensitivity analysis of NN parameters while keeping the parameters same for SD to forecast day-ahead electricity prices in the PJM market. Sensitivity analysis of NN parameters include back-propagation learning set (BP-set), learning rate (η), momentum (α) and NN learning days (dNN). The SD parameters, i.e. time framework of SD (d=45 days) and number of selected similar price days (N=5) are kept constant for all the simulated cases. Forecasting performance is carried out by choosing two different days from each season of the year 2006 and for which, the NN parameters for the base case are considered as BP-set=500, η=0.8, α=0.1 and dNN=45 days. Sensitivity analysis has been carried out by changing the value of BP-set (500, 1000, 1500); η (0.6, 0.8, 1.0, 1.2), α (0.1, 0.2, 0.3) and dNN (15, 30, 45 and 60 days). The most favorable value of BP-set is first found out from the sensitivity analysis followed by that of η and α, and based on which the best value of dNN is determined. Sensitivity analysis results demonstrate that the best value of mean absolute percentage error (MAPE) is obtained when BP-set=500, η=0.8, α=0.1 and dNN=60 days for winter season. For spring, summer and autumn, these values are 500, 0.6, 0.1 and 45 days, respectively. MAPE, forecast mean square error and mean absolute error of reasonably small value are obtained for the PJM data, which has correlation coefficient of determination (R2) of 0.7758 between load and electricity price. Numerical results show that forecasts generated by developed NN model based on the most favorable case are accurate and efficient. Copyright © 2008 John Wiley & Sons, Ltd.

Modelling damping ratio and shear modulus of sand–mica mixtures using neural networks

This study proposes two neural network (NN) models for damping ratio and shear modulus of sand–mica mixtures based on experimental results. The experimental database used for NN modelling is based on a laboratory study of dynamic properties of saturated coarse rotund sand and mica mixtures with various mix ratios under different effective stresses. In the tests, shear modulus, and damping ratio of the geomaterials have been measured for a strain range of 0.001% up to 0.1% using a Stokoe Resonant Column testing apparatus. The input variables in the developed NN models are the mica content, effective stress and strain, and the outputs are damping ratio and shear modulus. The performance of accuracies of proposed NN models are quite satisfactory ( R 2 = 0.97 for damping ratio and R 2 = 0.99 for shear modulus). Moreover the proposed NN models are also presented as simple explicit mathematical functions for further use by researchers.

Data Mining of Experimental Corrosion Data Using Neural Network

A supervised Neural Network (NN) method was used to predict the corrosion behavior (general and localized corrosion) and in turn, the life of metals and alloys over extended periods of time in specific environments, such that the developed NN model learns the underlying laws that map the alloy characteristics and environments to the alloy's corrosion behavior. The approach used in this work is the mining of existing experimental databases for both DC and AC corrosion experiments on metallic glasses, and grade-2 Titanium. The data mining results allow us to categorize and prioritize certain parameter sets (e.g. pH, temperature, time, electrolyte & metal compositions) that impact the alloy's corrosion rate and electrochemical potential. The developed NN models were tested by validating the predicted life of metals/alloys against available experimental data. The NN models were then used to predict future corrosion rates for user-specified environmental conditions and time frames.

Development and Application of a Rut Prediction Model for Flexible Pavement

A four-layer neural network (NN) was constructed and applied to determine mapping associating factors in the design and testing of asphalt samples with their performance in repetitive rutting tests. A total of 1,586 samples (two samples per data point) were tested to determine rutting with the use of an asphalt pavement analyzer. Test results and mix volumetric properties were used to train the NN model. Preprocessing and principal component analysis were used, and the network was trained with the Levenberg-Marquardt algorithm. With randomly generated weighting factors to initialize the training algorithm, histograms were compiled, and outputs were estimated. Excellent agreement was observed between simulations and test data. The developed NN was used to estimate the optimum asphalt content of a Superpave® mix, and the results were satisfactory. The developed NN model will be a useful tool in the study of asphalt construction and wear.

Development and Application of a Rut Prediction Model for Flexible Pavement

A four-layer neural network (NN) was constructed and applied to determine mapping associating factors in the design and testing of asphalt samples with their performance in repetitive rutting tests. A total of 1,586 samples (two samples per data point) were tested to determine rutting with the use of an asphalt pavement analyzer. Test results and mix volumetric properties were used to train the NN model. Preprocessing and principal component analysis were used, and the network was trained with the Levenberg–Marquardt algorithm. With randomly generated weighting factors to initialize the training algorithm, histograms were compiled, and outputs were estimated. Excellent agreement was observed between simulations and test data. The developed NN was used to estimate the optimum asphalt content of a Superpave® mix, and the results were satisfactory. The developed NN model will be a useful tool in the study of asphalt construction and wear.

Roll Force and Torque Prediction Using Neural Network and Finite Element Modelling

Roll force and torque prediction play a critical role in modern rolling schedule control and optimisation. For a given steel grade, the roll force and torque will be determined by the stock temperature, mill reduction, roll speed, friction and heat transfer, etc. A finite element (FE) model is first developed for the modelling of the rolling process, and gives prediction of stock temperature, strain, strain rate, and stress profiles during rolling. The average roll force and torque were then calculated by post-processing based on the distribution of the local variables of strain, strain rate, temperature, etc. A series of rolling experiments has been conducted to validate the developed FE model before the FE model is then used to generate the training data for developing the neural network (NN) models. An important innovation here is the integration of orthogonal design techniques for training data generation at minimal computational cost with the help of a validated FE model. This effectively overcomes the difficulty of obtaining good training data with sufficient excitations and balanced distribution via data collection. Novel techniques have been adopted in the NN modelling, including structure optimisation and double-loop iterative training to achieve an accurate and robust NN model based on the existing process data. The resulting NN models are feasible for on-line control and optimisation of the rolling process, which is impossible for the FE model due to the heavy computation involved. Also, ensemble modelling techniques have been adopted to improve the model prediction, which can also give confidence bounds for the predictions. The developed NN models can be easily extended to cover different steel grades and stock sizes in a straight-forward way by generating the corresponding training data from an FE model.