Feed-forward Back-propagation Neural Network Research Articles

An Optimized Artificial Neural Network Structure to Predict Clay Sensitivity in a High Landslide Prone Area Using Piezocone Penetration Test (CPTu) Data: A Case Study in Southwest of Sweden

Application of artificial neural networks (ANN) in various aspects of geotechnical engineering problems such as site characterization due to have difficulty to solve or interrupt through conventional approaches has demonstrated some degree of success. In the current paper a developed and optimized five layer feed-forward back-propagation neural network with 4-4-4-3-1 topology, network error of 0.00201 and R2 = 0.941 under the conjugate gradient descent ANN training algorithm was introduce to predict the clay sensitivity parameter in a specified area in southwest of Sweden. The close relation of this parameter to occurred landslides in Sweden was the main reason why this study is focused on. For this purpose, the information of 70 piezocone penetration test (CPTu) points was used to model the variations of clay sensitivity and the influences of direct or indirect related parameters to CPTu has been taken into account and discussed in detail. Applied operation process to find the optimized ANN model using various training algorithms as well as different activation functions was the main advantage of this paper. The performance and feasibility of proposed optimized model has been examined and evaluated using various statistical and analytical criteria as well as regression analyses and then compared to in situ field tests and laboratory investigation results. The sensitivity analysis of this study showed that the depth and pore pressure are the two most and cone tip resistance is the least effective factor on prediction of clay sensitivity.

Artificial neural network modeling for predicting the growth of the microalga Karlodinium veneficum

A feed-forward back-propagation neural network (FBN) was used as a tool for predicting the growth dynamics of the microalga Karlodinium veneficum in a culture medium with any specified concentrations of the key nutrients. A 3-layered FBN configuration of 27-25-1 nodes was used. This FBN satisfactorily represented the nonlinear interactions among all the nutrients of a culture medium containing up to 25 different components. The FBN model was trained using the growth dynamics data from more than 420 batch culture experiments involving different media compositions. The relative impact of individual nutrients, the initial cell concentration and the culture duration on growth profiles were determined through a systematic analysis of the partitioning of the FBN connection weights. Microelements and vitamins together had a higher relative impact on growth compared to the impact of the macronutrients. The trained FBN successfully predicted the cell concentration dynamics in cultures with previously untested initial conditions. The FBN proved to be an excellent tool for predicting the growth curves in the range of culture conditions that were relevant to this study.

The Use of genetic algorithm and neural network to predict rate-dependent tensile flow behaviour of AA5182-O sheets

In this study, the tensile flow behaviour of AA5182-O sheet was experimentally obtained in different material directions (RD, DD, and TD) at strain rates ranging from 0.001 to 1000s−1 and predicted by means of both phenomenological models and neural networks (NNs). Constants in Johnson–Cook (JC), Khan–Huang–Liang (KHL), and modified Voce were calculated using genetic algorithm (GA) and linear regression analysis and used to simulate the uniaxial tension tests. Two types of feed-forward back-propagation neural networks were also trained and validated to predict the rheological behaviour of the alloy without the limitations of a mathematical function. The weights and bias values of each network were then used to simulate uniaxial tensile deformation. Subsequently, the results were compared with experimental flow curves and accuracy parameters were calculated. It was found that the modified Voce constitutive equation was able to predict the flow behaviour of AA5182-O with better accuracy than JC and KHL models. Also, the NN was found to be the most accurate method of predicting the anisotropic rate-dependant behaviour of AA5182-O.

Supervised artificial neural network-based method for conversion of solar radiation data (case study: Algeria)

In this study, a backpropagation artificial neural network (BP-ANN) model is used as an alternative approach to predict solar radiation on tilted surfaces (SRT) using a number of variables involved in physical process. These variables are namely the latitude of the site, mean temperature and relative humidity, Linke turbidity factor and Angstrom coefficient, extraterrestrial solar radiation, solar radiation data measured on horizontal surfaces (SRH), and solar zenith angle. Experimental solar radiation data from 13 stations spread all over Algeria around the year (2004) were used for training/validation and testing the artificial neural networks (ANNs), and one station was used to make the interpolation of the designed ANN. The ANN model was trained, validated, and tested using 60, 20, and 20 % of all data, respectively. The configuration 8-35-1 (8 inputs, 35 hidden, and 1 output neurons) presented an excellent agreement between the prediction and the experimental data during the test stage with determination coefficient of 0.99 and root meat squared error of 5.75 Wh/m2, considering a three-layer feedforward backpropagation neural network with Levenberg–Marquardt training algorithm, a hyperbolic tangent sigmoid and linear transfer function at the hidden and the output layer, respectively. This novel model could be used by researchers or scientists to design high-efficiency solar devices that are usually tilted at an optimum angle to increase the solar incident on the surface.

Gaussian kernel-based FCM segmentation of brain MRI with BPNN classification

In this paper, the unsupervised Modified Gaussian Kernel-Based Fuzzy c-Means (MGKFCM) technique is proposed for the segmentation of magnetic resonance brain image and a Feed Forward Back Propagation Neural Network (FFBPNN) technique is presented for the classification of brain tissues. In this proposed MGKFCM algorithm, the Euclidean distance in the standard fuzzy c-means is replaced by a Gaussian radial basis function with additive bias field. The proposed method segments the given MRI data of the real image with the size of 443 × 708 × 3 automatically by considering the effects of intensity inhomogeneity, partial volume and noise. The performance of the proposed fuzzy technique has been found to be 98%. It has been compared with the existing magnetic resonance image segmentation techniques also. To separate the brain tissues into white matter (WM), gray matter (GM) and Cerebrospinal Fluid (CSF) from the fuzzy segmented output images, a different approach of FFBPNN classification technique has been used. The FFBPNN classifier has been trained already with the fuzzy segmented images. Sensitivity and specificity are the statistical measures used to analyse the performance of the proposed segmentation and selected classification tests.

Classification of lung columnar cells using feed forward back propagation neural network

Early detection of lung cancer is very essential for proper treatment and to reduce the mortality rate. Through this paper we proposed a methodology for the early detection of lung cancer, using lung cytology image analysis. Before the segmentation, the images are pre-processed using Contrast Limited Adaptive Histogram Equalisation (CLAHE). The features are extracted from segmented PAP stained sputum cytology images, using Discrete Wavelet Transforms (DWT). Proper feature extraction gave a good classification result, which is to be obtained through a good feature extraction and classification method. Here, we used DWT to extract the features from sputum cytology images. The enhanced images are segmented using Otsu segmentation method. The segmented images are classified using Feed Forward Back Propagation Neural Network (FFBNN). The experimental results show that FFBNN gave better classification result using features from DWT-based matrix and obtained an accuracy level 92.9%.

Compressor performance prediction using a novel feed-forward neural network based on Gaussian kernel function

In this article, a novel artificial neural network integrating feed-forward back-propagation neural network with Gaussian kernel function is proposed for the prediction of compressor performance map. To demonstrate the potential capability of the proposed approach for the typical interpolated and extrapolated predictions, other two classical data-driven modeling methods including feed-forward back-propagation neural network and support vector machine are compared. An assessment is performed and discussed on the sensitivity of different models to the number of training samples (48 training samples, 32 training samples, and 18 training samples). All the results indicate that the proposed neural network in this article has superior prediction performance to the existing feed-forward back-propagation neural network and support vector machine, especially for the extrapolation with small samples. Furthermore, this study can be utilized in refining the existing performance-based modeling for improved simulation analysis, condition monitoring, and fault diagnosis of gas turbine compressor.

Volumetric gas-liquid mass transfer coefficient in an external-loop airlift reactor with inserted membrane

The effects of the inserted membrane in the downcomer of an external-loop airlift reactor, the gas sparger type (single orifice and sinter plate) and added alcohol (ethanol, n-butanol, or n-hexanol) on the volumetric gas-liquid mass transfer coefficient (kLa) were studied. Due to the presence of the membrane in the downcomer, kLa did not change significantly; the differences were smaller than 10%. The highest values of the kLa were obtained using the sinter plate. It was found that the addition of small amounts of alcohol increased the mass transfer. Using our experimental results and the data of other authors, the feed-forward back propagation neural network for prediction of kLa in external-loop airlift reactors with alcohol solutions was proposed.

Breathomics for gastric cancer classification using back-propagation neural network

Breathomics is the metabolomics study of exhaled air. It is a powerful emerging metabolomics research field that mainly focuses on health-related volatile organic compounds (VOCs). Since the quantity of these compounds varies with health status, breathomics assures to deliver noninvasive diagnostic tools. Thus, the main aim of breathomics is to discover patterns of VOCs related to abnormal metabolic processes occurring in the human body. Classification systems, however, are not designed for cost-sensitive classification domains. Therefore, they do not work on the gastric carcinoma (GC) domain where the benefit of correct classification of early stages is more than that of later stages, and also the cost of wrong classification is different for all pairs of predicted and actual classes. The aim of this work is to demonstrate the basic principles for the breathomics to classify the GC, for that the determination of VOCs such as acetone, carbon disulfide, 2-propanol, ethyl alcohol, and ethyl acetate in exhaled air and stomach tissue emission for the detection of GC has been analyzed. The breath of 49 GC and 30 gastric ulcer patients were collected for the study to distinguish the normal, suspected, and positive cases using back-propagation neural network (BPN) and produced the accuracy of 93%, sensitivity of 94.38%, and specificity of 89.93%. This study carries out the comparative study of the result obtained by the single- and multi-layer cascade-forward and feed-forward BPN with different activation functions. From this study, the multilayer cascade-forward outperforms the classification of GC from normal and benign cases.

Neutral Networks for Fault Classification: Comparison between Feed-Forward Back-Propagation, RBF and LVQ Neural Network

— Neural networks are frequently used as a classifier for tasks in many classifications. However there are disadvantages in terms of amount of training data required, and length of training time. This paper, develop an intelligent diagnosis system for zinc oxide (ZnO) surge arrester fault classification. First the features were extracted from 600 ZnO surge arrester thermal images and leakage currents. Then these features were presented to several neural network architectures to investigate the most suitable network model for classifying the ZnO surge arrester fault condition effectively. Three classification models were used namely feed forward back propagation (FFBP), radial basis function (RBF) and learning vector quantization (LVQ) algorithm. The performance of the networks was compared based on resulted of misclassify and correct rate. The method was evaluated using 24 testing datasets. Comparison results showed that LVQ was the best training algorithm for the ZnO surge arrester fault classification compared to the others system. Also the LVQ is faster than FFBP and RBF.

Human age classification using appearance features and artificial neural network

This paper presents a novel method for human age classification via face images by a computer. The proposed method classifies the human face images into four age groups: child, young, adult and senior adult by using appearance features as ageing features and artificial neural network (ANN) as age classifier. The appearance features consist of both shape and textural features. Only two geometric ratios in combination with newly introduced rotation, scale and translation invariant efficient feature face angle are used as shape features. Local binary pattern histogram (LBPH) of regions of interest in face images are used as textural features. The ANN is designed by using two layer feedforward backpropagation neural networks. The performance of proposed age classification system is evaluated on face images from FG-NET ageing database and achieved greatly improved accuracy of 91.09% and 88.18% for male and female, respectively.

Human age classification using appearance features and artificial neural network

This paper presents a novel method for human age classification via face images by a computer. The proposed method classifies the human face images into four age groups: child, young, adult and senior adult by using appearance features as ageing features and artificial neural network (ANN) as age classifier. The appearance features consist of both shape and textural features. Only two geometric ratios in combination with newly introduced rotation, scale and translation invariant efficient feature face angle are used as shape features. Local binary pattern histogram (LBPH) of regions of interest in face images are used as textural features. The ANN is designed by using two layer feedforward backpropagation neural networks. The performance of proposed age classification system is evaluated on face images from FG-NET ageing database and achieved greatly improved accuracy of 91.09% and 88.18% for male and female, respectively.

Real-Time Monitoring and Fault Diagnosis of a Low Power Hub Motor Using Feedforward Neural Network.

Low power hub motors are widely used in electromechanical systems such as electrical bicycles and solar vehicles due to their robustness and compact structure. Such systems driven by hub motors (in wheel motors) encounter previously defined and undefined faults under operation. It may inevitably lead to the interruption of the electromechanical system operation; hence, economic losses take place at certain times. Therefore, in order to maintain system operation sustainability, the motor should be precisely monitored and the faults are diagnosed considering various significant motor parameters. In this study, the artificial feedforward backpropagation neural network approach is proposed to real-time monitor and diagnose the faults of the hub motor by measuring seven main system parameters. So as to construct a necessary model, we trained the model, using a data set consisting of 4160 samples where each has 7 parameters, by the MATLAB environment until the best model is obtained. The results are encouraging and meaningful for the specific motor and the developed model may be applicable to other types of hub motors. The prosperous model of the whole system was embedded into Arduino Due microcontroller card and the mobile real-time monitoring and fault diagnosis system prototype for hub motor was designed and manufactured.

Geometric Optimization of a PCM-Based Heat Sink—A Coupled ANN and GA Approach

Transient three-dimensional simulations are done for a phase-change material (PCM)-based heat sink subjected to constant heat flux of 1.59 kW/m2 with commercially available FLUENT software. Numerically obtained temperature–time histories are matched up with experimental results obtained earlier by our group to determine the overall heat transfer coefficient from the geometry that reconciles the differences between simulations and experiments. For a constant power level and fixed volumes of heat sink material (aluminum) and PCM (n-eicosane), simulations were done for 30 different geometric configurations to determine the time to reach complete melting. A feed-forward back-propagation neural network was then used in conjunction with a genetic algorithm to perform an exhaustive search and estimate the dimensions of the heat sink cavity for which the melt time and hence the thermal performance of the heat sink is maximum.

Prediction of the Liquid Vapor Pressure Using the Artificial Neural Network-Group Contribution Method

In this paper, vapor pressure for pure compounds is estimated using the Artificial Neural Networks and a simple Group Contribution Method (ANN-GCM). For model comprehensiveness, materials were chosen from various families. Most of materials are from 12 families. Vapor pressure data of 100 compounds is used to train, validate and test the ANN-GCM model. Vapor pressure data were taken from literature for wide ranges of temperature (68.55-559.15 K). Based on results, the best structure for feed-forward back propagation neural network is Levenberg-Marquardt back propagation training algorithm, logsig transfer function for hidden layer and linear transfer function for output layer. The multiplayer network model consists of temperature, acentric factor, critical temperature, critical pressure and the structure of molecules as inputs, 10 neurons in the hidden layer and one neuron in the output layer corresponding to vapor pressure. The weights are optimized to minimize error between experimental and calculated data. Results show that optimum neural network architecture is able to predict vapor pressure data with an acceptable level. The trained network predicts the vapor pressure data with average relative deviation percent of 1.18%.

Robust and accurate feature selection for humanoid push recovery and classification: deep learning approach

This current work describes human push recovery data classification using features that are obtained from intrinsic mode functions by performing empirical mode decomposition on different leg joint angles (hip, knee and ankle). Joint angle data were calculated for both open-eyes and closed-eyes subjects. Four kinds of pushes were applied (small, medium, moderately high, high) during the experiment to analyze the recovery mechanism. The classification was performed based on these different kinds of the pushes using deep neural network (DNN), and 89.28 % overall accuracy was achieved. The first classifier was based on artificial neural network on feed-forward back-propagation neural network (FF-BPNN), and second one was based on DNN. The proposed DNN-based classifier has been applied and evaluated on four types of pushes, i.e., small, medium, moderately high, high. The classification accuracy with a success of 88.4 % has been obtained using fivefold cross-validation approach. The analysis of variance has also been conducted to show the statistical significance of results. The corresponding strategies (hip, knee, and ankle) can be utilized once the categories of pushes (small, medium, moderately high, high) were identified accordingly push recovery (Semwal et al. in International conference on control, automation, robotics and embedded systems (CARE), pp 1---6, 2013).

Development and comparison of regression models and feedforward backpropagation neural network models to predict seasonal indoor PM2.5–10 and PM2.5 concentrations in naturally ventilated schools

A combination of multivariate statistical methods, including multiple linear regression (MLR) and feedforward backpropagation (FFBP) were used to evaluate the influence of seasons on the concentrations of indoor PM2.5−10 and PM2.5 in twelve naturally ventilated schools located in Gaza Strip, Palestine. Samples were collected by using hand held particulate matter sampler during fall, winter and spring from 2011 to 2012. Statistical results revealed that MLR models agree fairly well with the measured data with reasonable coefficients of determination (R2) 0.58, 0.69 and 0.70 for indoor PM2.5 and 0.44, 0.56 and 0.57 for indoor PM2.5−10 during fall, winter and spring, respectively. The FFBP model results performed better than the MLR analysis in determining indoor PMs with R2 values of 0.75, 0.78 and 0.79 for PM2.5 and 0.65, 0.73, and 0.78 for PM2.5−10 during fall, winter, and spring, respectively. The accuracy (R2) models of the FFBP showed an improvement of 12.08%–25.56% and from 26.36% to 38.53% for prediction of indoor PM2.5 and PM2.5−10 compared to MLR models. In addition, FFBP models improved the accuracy by reducing the error (RMSE) as much as 19.35%, 7.41%, and 7.41% during fall, winter, and spring for prediction of indoor PM2.5, and by 32.00%, 7.41%, and 32.00% during fall, winter, and spring, respectively for prediction of indoor PM2.5−10 compared with MLR. Results showed that the artificial neural network approach can be capable of accurately modeling indoor air quality in naturally ventilated buildings.

Multi-scale crown closure retrieval for moso bamboo forest using multi-source remotely sensed imagery based on geometric-optical and Erf-BP neural network models

This article focuses on retrieving the multi-scale crown closure (CC) of Moso bamboo forest using Système Pour l’Observation de la Terre (SPOT5) and Landsat Thematic Mapper (TM) satellite remotely sensed imagery based on the geometric-optical model and the artificial neural network (ANN) model. CC at local scale was first retrieved using the Li-Strahler geometric-optical model (LSGM) and images from an unmanned aerial vehicle (UAV). Then, multi-scale CC was retrieved using the Erf-BP model (a kind of back-propagation (BP) feed-forward neural network, which takes a Gaussian error function (Erf) as an activation function of the hidden layer) based on a combination of SPOT5 and Landsat TM images. The results show that by combining multi-source remotely sensed data, the CC of Moso bamboo forest can be retrieved at the local region, township area, and county scale with high accuracy using the Erf-BP model. Estimated values have a linear relationship with the observed values at a significance level of 0.05. The highest accuracy of the retrieval of CC (referred to as LSGM-UAV-CC) was observed at the local region based on LSGM and UAV, with the coefficient of determination (R2) of 0.63, followed by that at the township area with an R2 of 0.0.55 based on LSGM-UAV-CC and SPOT5 data using the Erf-BP model (Erf-BP-SPOT5-CC), and that at the county scale with an R2 of 0.54 based on Erf-BP-SPOT5-CC and Landsat TM data using the Erf-BP model (Erf-BP-TM-CC).

Numerical Simulation and Artificial Neural Network Modeling for Predicting Welding-Induced Distortion in Butt-Welded 304L Stainless Steel Plates

In the present study, artificial neural network modeling has been employed for predicting welding-induced angular distortions in autogenous butt-welded 304L stainless steel plates. The input data for the neural network have been obtained from a series of three-dimensional finite element simulations of TIG welding for a wide range of plate dimensions. Thermo-elasto-plastic analysis was carried out for 304L stainless steel plates during autogenous TIG welding employing double ellipsoidal heat source. The simulated thermal cycles were validated by measuring thermal cycles using thermocouples at predetermined positions, and the simulated distortion values were validated by measuring distortion using vertical height gauge for three cases. There was a good agreement between the model predictions and the measured values. Then, a multilayer feed-forward back propagation neural network has been developed using the numerically simulated data. Artificial neural network model developed in the present study predicted the angular distortion accurately.

PRELIMINARY MODELING OF AN INDUSTRIAL RECOMBINANT HUMAN ERYTHROPOIETIN PURIFICATION PROCESS BY ARTIFICIAL NEURAL NETWORKS

In the present study a preliminary neural network modelling to improve our understanding of Recombinant Human Erythropoietin purification process in a plant was explored. A three layer feed-forward back propagation neural network was constructed for predicting the efficiency of the purification section comprising four chromatographic steps as a function of eleven operational variables. The neural network model performed very well in the training and validation phases. Using the connection weight method the predictor variables were ranked based on their estimated explanatory importance in the neural network and five input variables were found to be predominant over the others. These results provided useful information showing that the first chromatographic step and the third chromatographic step are decisive to achieve high efficiencies in the purification section, thus enriching the control strategy of the plant.