Artificial Neural Network Output Research Articles

Limitation of the Artificial Neural Networks Methodology for Predicting the Vertical Swelling Percentage of Expansive Clays

AbstractThe general swelling model has recently been updated in Israel by applying the Excel-solver command (ESC) analysis to new local test results from 897 undisturbed specimens. In this analysis, the goodness-of-fit statistics obtained classify the category of their associated regression only as fair. Thus, it seems necessary to explore the possibility of enhancing the outputs of this regression analysis by applying the artificial neural networks (ANN) methodology to the same 897 undisturbed specimens. However, it is shown that the use of the ANN outputs should be accompanied by an additional check to ensure that they follow the expected physical swelling behavior, as characterized by the index properties of the soil. The ANN methodology applied in this paper is similar to previous studies in geotechnical engineering. Different models were tested using the same database (i.e., the same 897 undisturbed specimens). The statistical fit of the ANN models were clearly found to be superior to the ESC models....

Statistical Downscaling of River Runoff in a Semi Arid Catchment

Linear and non-linear statistical ‘downscaling’ study is applied to relate large-scale climate information from a general circulation model (GCM) to local-scale river flows in west Iran. This study aims to investigate and evaluate the more promising downscaling techniques, and provides a through inter comparison study using Karkheh catchment as an experimental site in a semi arid region for the years of 2040 to 2069. A hybrid conceptual hydrological model was used in conjunction with modeled outcomes from a General Circulation Model (GCM), HadCM3, along with two downscaling techniques, Statistical Downscaling Model (SDSM) and Artificial Neural Network (ANN), to determine how future streamflow may change in a semi arid catchment. The results show that the choice of a downscaling algorithm having a significant impact on the streamflow estimations for a semi-arid catchment, which are mainly, influenced, respectively, by atmospheric precipitation and temperature projections. According to the SDSM and ANN projections, daily temperature will increase up to +0.58 0C (+3.90 %) and +0.48 0C (+3.48 %), and daily precipitation will decrease up to −0.1 mm (−2.56 %) and −0.4 mm (−2.82 %) respectively. Moreover streamflow changes corresponding to downscaled future projections presented a reduction in mean annual flow of −3.7 m^3/s and −9.47 m^3/s using SDSM and ANN outputs respectively. The results suggest a significant reduction of streamflow in both downscaling projections, particularly in winter. The discussion considers the performance of each statistical method for downscaling future flow at catchment scale as well as the relationship between atmospheric processes and flow variability and changes.

Artificial Neural Network based control for PV/T panel to track optimum thermal and electrical power

As solar energy is intermittent, many algorithms and electronics have been developed to track the maximum power generation from photovoltaic and thermal panels. Following technological advances, these panels are gathered into one unit: PV/T system. PV/T delivers simultaneously two kinds of power: electrical power and thermal power. Nevertheless, no control systems have been developed in order to track maximum power generation from PV/T system. This paper suggests a PV/T control algorithm based on Artificial Neural Network (ANN) to detect the optimal power operating point (OPOP) by considering PV/T model behavior. The OPOP computes the optimum mass flow rate of PV/T for a considered irradiation and ambient temperature. Simulation results demonstrate great concordance between OPOP model based calculation and ANN outputs.

Using artificial neural networks to model extrusion processes for the manufacturing of polymeric micro-tubes

In this paper a methodology and an application example are presented aiming to show how Artificial Neural Networks (ANNs) can be used to model manufacturing processes when mathematical models are missing or are not applicable e.g. due to the micro- & nano-scaling, due to non-conventional processes, etc. Besides the ANNs methodology, the results of a Software System developed will be presented, which was used to create ANNs models for micro & nano manufacturing processes. More specifically results of a specific application example will be presented, concerning the modeling of extrusion processes for polymeric micro-tubes. ANNs models are capable for modeling manufacturing processes as far as adequate experimental and/or historical data of processes' inputs & outputs are available for their training. The POLYTUBES ANNs models have been trained and tested with experimental data records of process' inputs and outputs concerning a micro-extrusion process of polymeric micro-tubes for several materials such as: COC, PC, PET, PETG, PP and PVDF. The main ANN model of the extrusion application example has 3 inputs and 9 outputs. The inputs are: tube's inner & outer diameters, and the material density. The model outputs are 9 process parameters, which correspond to the specific inputs e.g. process temperature, die inner & outer diameters, extrusion pressure, draw speed etc. The training of the ANN model was completed, when the errors for the model's outputs, which expressed the difference between the training target values and the ANNs outputs, were minimized to acceptable levels. After the training, the micro-extrusion ANN is capable to simulate the process and can be used to calculate model's outputs, which are the process parameters for any new set of inputs. By this way a satisfactory functional approximation of the whole process is achieved. This research work has been supported by the EU FP7 NMP project POLYTUBES.

ANN based Sensor Faults Detection, Isolation, and Reading Estimates – SFDIRE: Applied in a nuclear process

In this paper a new method is developed based on the capabilities of the Artificial Neural Networks (ANNs) to detect, isolate, and output an estimated reading of faulty sensor/s reading/s that serves in monitoring a complex process. For a number of measured signals working to monitor a process, we allow training of an equal number of ANN’s to those signals. Each ANN has an output signal equal to one corresponding measured signal and the set of inputs is the rest of other measured signals and other networks output signals using two-hidden layer feedforward network with different number of neurons for each. After the measuring the error between the real signal and the ANN output, the faulty sensor can be detected. The reading of this faulty sensor could be isolated and an estimated reading is sent to the monitoring system. This method could be applied for any difficult to model process to increase its availability and in the same time maintaining safety. In this paper, the application considered is the thermal–hydraulics process of Egyptian Training and Research Reactor-2 (ETRR-2) core cooling system. Data from the reactor are used to train the networks using backpropagation algorithm, An approximated model based on the lumped parameters approach for the thermal–hydraulics of the reactor core is built to compare results against ANN based system. The proposed system deals with 6 signals. From comparison, very good results are achieved when testing the design.

Quantification of the predictive uncertainty of artificial neural network based river flow forecast models

The meaningful quantification of uncertainty in hydrological model outputs is a challenging task since complete knowledge about the hydrologic system is still lacking. Owing to the nonlinearity and complexity associated with the hydrological processes, Artificial neural network (ANN) based models have gained lot of attention for its effectiveness in function approximation characteristics. However, only a few studies have been reported for assessment of uncertainty associated with ANN outputs. This study uses a simple method for quantifying predictive uncertainty of ANN model output through first order Taylor series expansion. The first order partial differential equations of non-linear function approximated by the ANN with respect to weights and biases of the ANN model are derived. A bootstrap technique is employed in estimating the values of the mean and the standard deviation of ANN parameters, and is used to quantify the predictive uncertainty. The method is demonstrated through the case study of Upper White watershed located in the United States. The quantitative assessment of uncertainty is carried out with two measures such as percentage of coverage and average width. In order to show the magnitude of uncertainty in different flow domains, the values are statistically categorized into low-, medium- and high-flow series. The results suggest that the uncertainty bounds of ANN outputs can be effectively quantified using the proposed method. It is observed that the level of uncertainty is directly proportional to the magnitude of the flow and hence varies along time. A comparison of the uncertainty assessment shows that the proposed method effectively quantifies the uncertainty than bootstrap method.

Comparison of clinicians and an artificial neural network regarding accuracy and certainty in performance of visual field assessment for the diagnosis of glaucoma

To compare clinicians and a trained artificial neural network (ANN) regarding accuracy and certainty of assessment of visual fields for the diagnosis of glaucoma. Thirty physicians with different levels of knowledge and experience in glaucoma management assessed 30-2 SITA Standard visual field printouts that included full Statpac information from 99 patients with glaucomatous optic neuropathy and 66 healthy subjects. Glaucomatous eyes with perimetric mean deviation values worsethan -10 dB were not eligible. The fields were graded on a scale of 1-10, where 1 indicated healthy with absolute certaintyand 10 signified glaucoma; 5.5 was the cut-off between healthy and glaucoma. The same fields were classified by a previously trained ANN. The ANN output was transformed into a linear scale that matched the scale used in the subjective assessments. Classification certainty was assessed using a classification error score. Among the physicians, sensitivity ranged from 61% to 96% (mean 83%) and specificity from 59% to 100% (mean 90%). Our ANN achieved 93% sensitivity and 91% specificity, and it was significantly more sensitive than the physicians (p < 0.001) at a similar level of specificity. The ANN classification error score was equivalent to the top third scores of all physicians, and the ANN never indicated a high degree of certainty for any of its misclassified visual field tests. Our results indicate that a trained ANN performs at least as well as physicians in assessments of visual fields for the diagnosis of glaucoma.

Application of artificial neural network to predict thermal transmittance of wooden windows

Thermal performance of windows depends on many parameters, such as dimensional characteristics and material properties of the components. The thermal transmittance U can be evaluated by a numerical method based on the CFD approach for the evaluation of the frame U-value (ISO 10077-1, ISO 10077-2) or by experimental campaigns on window prototypes, according to ISO 12657-1; in both cases significant effort and time are required.The paper aims at developing an artificial neural network (ANN) model to predict the U-value of wooden windows, with the same accuracy of the numerical calculation procedure cited above (therefore greater than the one of the simplified method), but in real time and on the basis of a limited number of parameters. In particular, after a preliminary analysis, only 10 main parameters were selected as network inputs: window typology (windows and French windows), wood kind (hardwood and softwood), frame and shutters thickness, glazing spacer, top frame junction characteristics (size and number of small and large non-ventilated air cavities), U-value of the glazing and glazing size; the U-value of the window is the ANN output. Data set for the training and test of the ANN model consist of respectively 256 and 26 wooden window samples (windows and French windows). In such hypothesis, the developed ANN model, based on a multilayer feed-forward architecture, provides in real time the evaluation of the window U-value with an error of about 1% with respect to the results provided by the CFD numerical procedure), obviously when the input parameters vary within appropriate ranges (corresponding to the variation range of the data used for the network training). The ANN model set-up, therefore, allows to easily determine with high accuracy the thermal performance of wooden windows, saving both money and time. A sensitivity analysis of the main design parameters was also carried out.

Sensitivity analysis of the artificial neural network outputs in simulation of the evaporation process at different climatologic regimes

This study follows three aims; firstly to develop and examine three different Artificial Neural Networks (ANNs) viz.: Multi-Layer Perceptron (MLP), Radial Basis Neural Network (RBNN) and Elman network for estimating daily evaporation rate of Tabriz and Urmia cities using measured hydro-meteorological data; second to compare the results of ANN models with three physically-based models include, Energy balance, Aerodynamic, and Penman models and also black-box Multiple Linear Regression (MLR) model; and finally to perform a sensitivity analysis to investigate the effect of each input parameter on the output in terms of magnitude and direction. The used meteorological data set to develop the models for estimation of daily evaporation includes daily air temperature, evaporation, solar radiation, air pressure, relative humidity, and wind speed measured at synoptic stations of Tabriz and Urmia cities which have almost distinct climatologic conditions. The obtained results denote to the superiority of the ANN models on the classic models. Also based on the comparisons, the MLP network performs better than the RBNN and Elman network so that in the next step, sensitivity analysis is performed by the Partial Derivation (PaD) and Weights methods on the MLP outputs. Sensitivity analysis results show although air temperature, solar radiation and the amount of evaporation at previous time step are the effective parameters in estimation of daily evaporation at both regions, due to the climatologic condition wind speed and relative humidity are other predominant parameters in Tabriz and Urmia, respectively.

The use of an artificial neural network for modeling the moisture absorption and thickness swelling of oriented strand board

In this study, an artificial neural network (ANN) approach was employed for modeling the moisture absorption (MA) and thickness swelling (TS) properties of oriented strand board (OSB) in various applications. A series of ANN models were developed for the analysis and prediction of correlations between processing parameters and MA and TS of OSB. An ANN model was found for modeling the effects of OSB treatment variables on the MA and TS. The required data for training and testing of the model were obtained from the experimental results of Salay (2010). In designing this model, the MA and TS of the OSB were determined using OSB treatment variables, including board layup type, resin type, application rate of resin, and wax content. When experimental data and results obtained from the ANN were compared by regression analysis using Matlab, it was determined that both groups of data (test and train) were consistent. It was demonstrated that the well-trained feed forward and back propagation multilayer ANN model is a powerful and sufficient tool for the prediction of MA and TS; therefore, by using ANN outputs, satisfactory results can be estimated, rather than measured and hence time and cost are reduced in all the required experimental activities.

Neural network and classification approach in identifying customer behavior in the banking sector: A case study of an international bank

The customer relationship focus for banks is in development of main competencies and strategies of building strong profitable customer relationships through considering and managing the customer impression, influence on the culture of the bank, satisfactory treatment, and assessment of valued relationship building. Artificial neural networks (ANNs) are used after data segmentation and classification, where the designed model register records into two class sets, that is, the training and testing sets. ANN predicts new customer behavior from previously observed customer behavior after executing the process of learning from existing data. This article proposes an ANN model, which is developed using a six-step procedure. The back-propagation algorithm is used to train the ANN by adjusting its weights to minimize the difference between the current ANN output and the desired output. An evaluation process is conducted to determine whether the ANN has learned how to perform. The training process is halted periodically, and its performance is tested until an acceptable result is obtained. The principles underlying detection software are grounded in classical statistical decision theory.

NEURAL NETWORK AIDED EVALUATION OF LANDSLIDE SUSCEPTIBILITY IN SOUTHERN ITALY

Landslide hazard mapping is often performed through the identification and analysis of hillslope instability factors. In heuristic approaches, these factors are rated by the attribution of scores based on the assumed role played by each of them in controlling the development of a sliding process. The objective of this research is to forecast landslide susceptibility through the application of Artificial Neural Networks. In particular, given the availability of past events data, we mainly focused on the Calabria region (Italy). Vectors of eight hillslope factors (features) were considered for each considered event in this area (lithology, permeability, slope angle, vegetation cover in terms of type and density, land use, yearly rainfall and yearly temperature range). We collected 106 vectors and each one was labeled with its landslide susceptibility, which is assumed to be the output variable. Subsequently a set of these labeled vectors (examples) was used to train an artificial neural network belonging to the category of Multi-Layer Perceptron (MLP) to evaluate landslide susceptibility. Then the neural network predictions were verified on the vectors not used in the training (validation set), i.e. in previously unseen locations. The comparison between the expected output and the artificial neural network output showed satisfactory results, reporting a prediction discrepancy of less than 4.3%. This is an encouraging preliminary approach towards a systematic introduction of artificial neural network in landslide hazard assessment and mapping in the considered area.

A novel soft sensor model based on artificial neural network in the fermentation process

Some crucial process variables in fermentation process could not be measured directly. Soft sensor technology provided an effective way to solve the problem. There has been considerable interest in modeling a soft sensor by using artificial neural network (ANN) in bioprocess. To generate a more efficient soft sensor model, we proposed a novel soft sensor model based on artificial neural network (SS-ANN). By analyzing a grey-box model of fermentation process, the secondary variables were selected. In modeling, on-line measurable variables could be taken as the input of ANN and the output is the derivatives of immeasurable variables. The estimated values of immeasurable variables were calculated by integrating the outputs of the well-trained ANN. The novel SS-ANN is different from the general SS-ANN. Experimental results of erythromycin fermentation process showed the novel soft sensor model could estimate mycelia concentration, sugar concentration and chemical potency with higher accuracy and generalization ability than the general soft sensor based on ANN. The novel soft sensor modeling method provides the theory basis for selecting the secondary variables. The dynamic characteristic of the process is considered, the novel model improves the estimation accuracy and generation ability. It can be concluded that the soft sensor model mentioned in this paper is reasonable and effective. Key words : Soft sensor model, artificial neural network, fermentation process, dynamic characteristics.

Modeling Blanking Process Using Multiple Regression Analysis and Artificial Neural Networks

The design of blanking processes requires the availability of a procedure able to deal with both tooling and mechanical properties of the workpiece material (blank thickness, hardness, ductility, etc.). This research presents the development and comparison of two models to predict the quality of the blanked edge represented by burrs height, the first model is an artificial neural network (ANN) based, while the second model is a multiple regression analysis (MRA) based. Finite Element modeling of the blanking process was used to generate the data for both models. Both ANN and MRA are able to give good prediction results, however, ANN still more accurate because it deals efficiently with hidden nonlinear relations when compared to MRA. The comparison between experimental and model results shows that average absolute relative error in the case of ANN was <2.20% for carbon steel and 4.85% for corrosion-resistant steel (CRES) compared to 15.18% for carbon steel and 14.22% for CRES obtained from the second order MRA. Therefore, by using ANN outputs, satisfactory results can be estimated rather than measured and hence reduce testing time and cost.

A flexible system for initial ship design parameters estimation using a system of neural networks

To initialize ship design process, it is very important to be able to develop an initial estimate of ship parameters to satisfy designer required specifications. For new emerging designs, this estimate has to be made based on a limited available set of examples. Moreover, a practical estimate prediction strategy should be flexible enough having no distinction between input (specified constraints) and outputs (parameters required to be estimated), since these vary from one design case to another. Conventional regression-based techniques, which are usually employed to provide the required estimates, suffer from low accuracy in case of a small number of available examples. In addition to that, they fail to capture the interrelation between different design parameters. To overcome these limitations and others, the present paper proposes a new approach based on a system of artificial neural-networks (ANNs). The new approach not only overcomes regression limitations but is also capable of providing a reliable estimate of initial design offset table based on different ANN outputs. The paper uses a case study for demonstrating the merits of the proposed approach.Keywords: Ship design; regression; ship series; Artificial Neural Networks (ANNs); Multilayer Perceptrons (MLPs); Normalized Gaussian Modified Lagrangian (NGML) doi: http://dx.doi.org/10.3329/jname.v8i2.6945 Journal of Naval Architecture and Marine Engineering 8(2011) 71-82

Predicting subsurface soil layering and landslide risk with Artificial Neural Networks: a case study from Iran

Predicting subsurface soil layering and landslide risk with Artificial Neural Networks: a case study from Iran This paper is concerned principally with the application of Artificial Neural Networks (ANN) in geotechnical engineering. In particular the application of ANN is discussed in more detail for subsurface soil layering and landslide analysis. Two ANN models are trained to predict subsurface soil layering and landslide risk using data collected from a study area in northern Iran. Given the three-dimensional coordinates of soil layers present in thirty boreholes as training data, our first ANN successfully predicted the depth and type of subsurface soil layers at new locations in the region. The agreement between the ANN outputs and actual data is over 90 % for all test cases. The second ANN was designed to recognize the probability of landslide occurrence at 200 sampling points which were not used in training. The neural network outputs are very close (over 92 %) to risk values calculated by the finite element method or by Bishop's method.

Prediction of Compressional, Shear, and Stoneley Wave Velocities from Conventional Well Log Data Using a Committee Machine with Intelligent Systems

Measurement of compressional, shear, and Stoneley wave velocities, carried out by dipole sonic imager (DSI) logs, provides invaluable data in geophysical interpretation, geomechanical studies and hydrocarbon reservoir characterization. The presented study proposes an improved methodology for making a quantitative formulation between conventional well logs and sonic wave velocities. First, sonic wave velocities were predicted from conventional well logs using artificial neural network, fuzzy logic, and neuro-fuzzy algorithms. Subsequently, a committee machine with intelligent systems was constructed by virtue of hybrid genetic algorithm-pattern search technique while outputs of artificial neural network, fuzzy logic and neuro-fuzzy models were used as inputs of the committee machine. It is capable of improving the accuracy of final prediction through integrating the outputs of aforementioned intelligent systems. The hybrid genetic algorithm-pattern search tool, embodied in the structure of committee machine, assigns a weight factor to each individual intelligent system, indicating its involvement in overall prediction of DSI parameters. This methodology was implemented in Asmari formation, which is the major carbonate reservoir rock of Iranian oil field. A group of 1,640 data points was used to construct the intelligent model, and a group of 800 data points was employed to assess the reliability of the proposed model. The results showed that the committee machine with intelligent systems performed more effectively compared with individual intelligent systems performing alone.

Α new void fraction correlation inferred from artificial neural networks for modeling two-phase flows in geothermal wells

A new empirical void fraction correlation was developed using artificial neural network (ANN) techniques. The artificial networks were trained using the backpropagation algorithm and production data obtained from a worldwide database of geothermal wells. Wellhead pressure, steam quality, wellbore diameter, the fluid density and viscosity, and the dimensionless numbers Reynolds, Weber, and Froude were used as main input parameters. The target ANN output was defined by the optimized void fraction values (αopt), which were calculated from the numerical modeling of two-phase flow using GEOWELLS (a wellbore simulator). The Levenberg–Marquardt algorithm, the hyperbolic tangent sigmoid, and the linear activation functions were used for the development of the ANN model. The best ANN learning was achieved with an architecture of six neurons in the hidden layer, which made it possible to obtain a set of void fractions (αANN) with a good accuracy (R2=0.9722). These void fraction estimates were used to obtain the new correlation, which was later coupled into the simulator GEOWELLS for the prediction of pressure gradients in two-phase geothermal wells. The accuracy of the new correlation (αANN) was evaluated by a statistical comparison between simulated pressure gradients and measured field data. These simulation results were also compared with those data calculated by using Duns-Ros and Dix correlations, which were also programmed into GEOWELLS. Pressure gradients predicted with the new αANN correlation showed a better agreement with measured field data, which was also confirmed by the lower values of some statistical parameters (MPE, RMSE, and Theil's U). The statistical evaluation demonstrated the efficiency of the new correlation to predict void fractions and pressure gradients with a better accuracy, in comparison to the other existing correlations. These successful results suggest the use of the new correlation (αANN) for the analysis of two-phase flow mechanisms of geothermal wells.

Nonlinear analysis of load-deflection testing of reinforced one-way slab strengthened by carbon fiber reinforced polymer (CFRP) and using artificial neural network (ANN) for prediction

Load-deflection curve is the most important part of the structural analysis of RC beam and slab. The load-deflection analysis of the RC one-way slab strengthened by CFRP using experimental work, finite element analysis (FEA), artificial neural network (ANN), and a comparison of them together are the important objective of this study. The dimension of the one-way slab was 1800A�400A�120 mm which was strengthened by different length and width of carbon fiber reinforced polymer (CFRP). The experimental results sufficiently adapted with FEA and ANNs output. The feed forward back-propagation (FFB) was the best ANN for prediction of load-deflection curve with minimum error below 1, and maximum correlation coefficient close to 1.

Locating and classifying defects using an hybrid data base

A computational inverse technique was used in the localization and classification of defects. Postulated voids of two different sizes (2 mm and 4 mm diameter) were introduced in PMMA bars with and without a notch. The bar dimensions are 200×20×5 mm. One half of them were plain and the other half has a notch (3 mm × 4 mm) which is close to the defect area (19 mm × 16 mm).This analysis was done with an Artificial Neural Network (ANN) and its optimization was done with an Adaptive Neuro Fuzzy Procedure (ANFIS). A hybrid data base was developed with numerical and experimental results. Synthetic data was generated with the finite element method using SOLID95 element of ANSYS code. A parametric analysis was carried out. Only one defect in such bars was taken into account and the first five natural frequencies were calculated. 460 cases were evaluated. Half of them were plain and the other half has a notch. All the input data was classified in two groups. Each one has 230 cases and corresponds to one of the two sort of voids mentioned above. On the other hand, experimental analysis was carried on with PMMA specimens of the same size. The first two natural frequencies of 40 cases were obtained with one void. The other three frequencies were obtained numerically. 20 of these bars were plain and the others have a notch. These experimental results were introduced in the synthetic data base. 400 cases were taken randomly and, with this information, the ANN was trained with the backpropagation algorithm. The accuracy of the results was tested with the 100 cases that were left. In the next stage of this work, the ANN output was optimized with ANFIS. Previous papers showed that localization and classification of defects was reduced as notches were introduced in such bars. In the case of this paper, improved results were obtained when a hybrid data base was used.