Feed-forward Back-propagation Artificial Neural Network Model Research Articles

Vibro-Fluidized Bed Drying of Pumpkin Seeds: Assessment of Mathematical and Artificial Neural Network Models for Drying Kinetics

Pumpkin seeds were dried in a vibro-fluidized bed dryer (VFBD) at different temperatures at optimized vibration intensity of 4.26 and 4 m/s air velocity. The drying characteristics were mapped employing semiempirical models and Artificial Neural Network (ANN). Prediction of drying behavior of pumpkin seeds was done using semiempirical models, of which, one was preferred as it indicated the best statistical indicators. Two-term model showed the best fit of data with R2 − 0.999, and lowest χ2 − 1.03 × 10−4 and MSE 7.55 × 10−5. A feedforward backpropagation ANN model was trained by the Levenberg–Marquardt training algorithm using a TANSIGMOID activation function with 2-10-2 topology. Performance assessment of ANN showed better prediction of drying behavior with R2 = 0.9967 and MSE = 5.21 × 10−5 for moisture content, and R2 = 0.9963 and MSE = 2.42 × 10−5 for moisture ratio than mathematical models. In general, the prediction of drying kinetics and other drying parameters was more precise in the ANN technique as compared to semiempirical models. The diffusion coefficient, Biot number, and hm increased from 1.12 × 10−9 ± 3.62 × 10−10 to 1.98 × 10−9 ± 4.61 × 10−10 m2/s, 0.51 ± 0.01 to 0.60 ± 0.01, and 1.49 × 10−7 ± 4.89 × 10−8 to 3.10 × 10−7 ± 7.13 × 10−8 m/s, respectively, as temperature elevated from 40 to 60°C. Arrhenius’s equation was used to the obtain the activation energy of 32.71 ± 1.05 kJ/mol.

Decanol proportional effect prediction model as additive in palm biodiesel using ANN and RSM technique for diesel engine

The present examination explores the impact of utilizing palm biodiesel and decanol blends as a ternary blends in compression ignition engine. Artificial Neural network (ANN) and Response surface methodology (RSM) model is developed to predict and optimize decanol proportion in ternary blends. Test engine was operated with decanol mixing of 10%, 20%, and 30% by volume while the diesel percentage was kept 50% in all the samples at different load conditions. With the support of obtained data, regression equation for all the responses are generated based on RSM model with R2 > 0.95. Further to improve the regression model, feed-forward back propagation ANN model is developed to replica the output of brake thermal efficiency, brake specific fuel consumption, oxides of nitrogen, hydrocarbon, smoke opacity and ignition delay period, carbon monoxide, carbon dioxide and exhaust gas temperature and MFB@90%. Based on the results, ANN predicts all the responses with R > 0.99 and the maximum mean absolute average error is 1.879%. The RSM based optimization study prescribes 30% decanol, biodiesel 20% and diesel 50% sufficiently improve performance and decrease emission parameters. Even, the confirmation study with desirability level of 75.4% was carried at 3 trials for the purpose of authentication, maximum error of 5.27% and minimum of 1.33% error was observed.

Application of Multiple Linear Regression and Artificial Neural Networks for the Prediction of the Packing and Capsule Filling Performance of Coated and Plain Pellets Differing in Density and Size.

Plain or coated pellets of different densities 1.45, 2.53, and 3.61 g/cc in two size ranges, small (380–550 μm) and large (700–1200 μm) (stereoscope/image analysis), were prepared according to experimental design using extrusion/spheronization. Multiple linear regression (MLR) and artificial neural networks (ANNs) were used to predict packing indices and capsule filling performance from the “apparent” pellet density (helium pycnometry). The dynamic packing of the pellets in tapped volumetric glass cylinders was evaluated using Kawakita’s parameter a and the angle of internal flow θ. The capsule filling was evaluated as maximum fill weight (CFW) and fill weight variation (FWV) using a semi-automatic machine that simulated filling with vibrating plate systems. The pellet density influenced the packing parameters a and θ as the main effect and the CFW and FWV as statistical interactions with the coating. The pellet size and coating also displayed interacting effects on CFW, FWV, and θ. After coating, both small and large pellets behaved the same, demonstrating smooth filling and a low fill weight variation. Furthermore, none of the packing indices could predict the fill weight variation for the studied pellets, suggesting that the filling and packing of capsules with free-flowing pellets is influenced by details that were not accounted for in the tapping experiments. A prediction could be made by the application of MLR and ANNs. The former gave good predictions for the bulk/tap densities, θ, CFW, and FWV (R-squared of experimental vs. theoretical data >0.951). A comparison of the fitting models showed that a feed-forward backpropagation ANN model with six hidden units was superior to MLR in generalizing ability and prediction accuracy. The simplification of the ANN via magnitude-based pruning (MBP) and optimal brain damage (OBD), showed good data fitting, and therefore the derived ANN model can be simplified while maintaining predictability. These findings emphasize the importance of pellet density in the overall capsule filling process and the necessity to implement MLR/ANN into the development of pellet capsule filling operations.

An artificial neural network to predict resting energy expenditure in obesity

The resting energy expenditure (REE) determination is important in nutrition for adequate dietary prescription. The gold standard i.e. indirect calorimetry is not available in clinical settings. Thus, several predictive equations have been developed, but they lack of accuracy in subjects with extreme weight including obese populations. Artificial neural networks (ANN) are useful predictive tools in the area of artificial intelligence, used in numerous clinical fields. The aim of this study was to determine the relevance of ANN in predicting REE in obesity. A Multi-Layer Perceptron (MLP) feed-forward neural network with a back propagation algorithm was created and cross-validated in a cohort of 565 obese subjects (BMI within 30-50kgm-2) with weight, height, sex and age as clinical inputs and REE measured by indirect calorimetry as output. The predictive performances of ANN were compared to those of 23 predictive REE equations in the training set and in two independent sets of 100 and 237 obese subjects for external validation. Among the 23 established prediction equations for REE evaluated, the Harris & Benedict equations recalculated by Roza were the most accurate for the obese population, followed by the USA DRI, Müller and the original Harris & Benedict equations. The final 5-fold cross-validated three-layer 4-3-1 feed-forward back propagation ANN model developed in that study improved precision and accuracy of REE prediction over linear equations (precision=68.1%, MAPE=8.6% and RMSPE=210kcal/d), independently from BMI subgroups within 30-50kgm-2. External validation confirmed the better predictive performances of ANN model (precision=73% and 65%, MAPE=7.7% and 8.6%, RMSPE=187kcal/d and 200kcal/d in the 2 independent datasets) for the prediction of REE in obese subjects. We developed and validated an ANN model for the prediction of REE in obese subjects that is more precise and accurate than established REE predictive equations independent from BMI subgroups. For convenient use in clinical settings, we provide a simple ANN-REE calculator available at: https://www.crnh-rhone-alpes.fr/fr/ANN-REE-Calculator.

Optimisation of ANN topology for predicting the rehydrated apple cubes colour change using RSM and GA

In this study, an efficient optimisation method by combining response surface methodology (RSM) and genetic algorithm (GA) is introduced to find the optimal topology of artificial neural networks (ANNs) for predicting colour changes in rehydrated apple cubes. A multi-layered feed-forward backpropagation ANN model of algorithms was developed to correlate one output (colour change) to four input variables (drying air temperature, drying air velocity, temperature of distilled water and rehydration time). A predictive model for ANN topology in terms of the best mean squared error (MSE) performance on validation samples was created using RSM. RSM model was integrated with an effective GA to find the optimum topology of ANN. The optimum ANN had minimum MSE when the number of hidden neurons, learning rate, momentum constant, number of epochs and number of training runs were 13, 0.33, 0.89, 3869 and 3, respectively. MSE of optimal ANN topology on validation samples was 0.0072095. It turned out that the optimal ANN topology can be considered as more precise for predicting colour change in the rehydrated apple cubes. Mean absolute error and regression coefficient (R) of the optimal ANN topology were determined as 0.0259 and 0.96475 for training, 0.0399 and 0.95243 for testing and 0.0264 and 0.95151 for validation data sets. The results of the testing model on new samples showed excellent agreement between the actual and predicted data with coefficient of determination R2 = 0.97.

A Comparative Study of ANN and CFD Modelling for Pressure Drop Prediction in a Fluidized Bed with Internals

In the present study, experiments on fluidized beds were conducted by varying hydrodynamic parameters such as air velocity, bed height and internal spacing and these data were used in the development of a model using artificial neural network (ANN) and simulation using computational fluid dynamics (CFD). Staggered rod internals of width 25% and 50% of the column diameter were used. A two-layered feed-forward back propagation ANN model was developed to predict the pressure drop with 15 neurons and it gave a high R2 value of 0.9966. The model prediction and experimental data are in good agreement. The results from CFD studies are also in good agreement with the experimental data. Pressure drop was found to reduce in the presence of internals and it reduced further with a decrease in internal spacing due to continuous splitting of bubbles. The average error in CFD predicted pressure drop is 9.56% and ANN predicted pressure drop is 0.95%.

Artificial neural networks for infectious diarrhea prediction using meteorological factors in Shanghai (China)

Infectious diarrhea is an important public health problem around the world. Meteorological factors have been strongly linked to the incidence of infectious diarrhea. Therefore, accurately forecast the number of infectious diarrhea under the effect of meteorological factors is critical to control efforts. In recent decades, development of artificial neural network (ANN) models, as predictors for infectious diseases, have created a great change in infectious disease predictions. In this paper, a three layered feed-forward back-propagation ANN (BPNN) model trained by Levenberg–Marquardt algorithm was developed to predict the weekly number of infectious diarrhea by using meteorological factors as input variable. The meteorological factors were chosen based on the strongly relativity with infectious diarrhea. Also, as a comparison study, the support vector regression (SVR), random forests regression (RFR) and multivariate linear regression (MLR) also were applied as prediction models using the same dataset in addition to BPNN model. The 5-fold cross validation technique was used to avoid the problem of overfitting in models training period. Further, since one of the drawbacks of ANN models is the interpretation of the final model in terms of the relative importance of input variables, a sensitivity analysis is performed to determine the parametric influence on the model outputs. The simulation results obtained from the BPNN confirms the feasibility of this model in terms of applicability and shows better agreement with the actual data, compared to those from the SVR, RFR and MLR models. The BPNN model, described in this paper, is an efficient quantitative tool to evaluate and predict the infectious diarrhea using meteorological factors.

Design and Test of a Low-Cost Electronic Nose System for Identification of Salmonella enterica in Poultry Manure

Abstract. The objective of this study was to design and evaluate the performance of a metal-oxide sensor-based electronic nose system (e-nose) for detecting Salmonella enterica in poultry manure. The system has hardware and software components for signal acquisition, data processing, and sample classification. An artificial neural network (ANN) model was used to classify manure samples as Salmonella-positive or Salmonella-negative. Seven manure samples were collected from different broiler houses and divided into four portions. Two portions were spiked with 10 3 and 2 A— 10 3 CFU g -1 of Salmonella enterica (ATCC 13311). The third portion was used for determining natural manure microflora, and the fourth portion was sterilized. All portions were incubated at 37°C for 48 h. A total of 84 e-nose readings were recorded at different time intervals from the manure portions. A multilayer, feed-forward back-propagation ANN model was developed (training step) and validated with the e-nose readings. Of the 84 readings, 48 were used to develop the ANN model and the remainder was used to validate model performance. The model was able to classify the remaining 36 manure samples with an accuracy of 96%. In order to test the actual performance of the ANN model, 16 manure samples were collected from different barns and analyzed. The e-nose system was able to determine the Salmonella status of the manure samples with 100% accuracy.

Hot Deformation Behavior and Flow Stress Prediction of TC4-DT Alloy in Single-Phase Region and Dual-Phase Regions

Isothermal compression tests of TC4-DT titanium alloy at the deformation temperature ranging from 1181 to 1341 K covering α + β phase field and β-phase field, the strain rate ranging from 0.01 to 10.0 s−1 and the height reduction of 70% were conducted on a Gleeble-3500 thermo-mechanical simulator. The experimental true stress-true strain data were employed to develop the strain-compensated Arrhenius-type flow stress model and artificial neural network (ANN) model; the predictability of two models was quantified in terms of correlation coefficient (R) and average absolute relative error (AARE). The R and AARE for the Arrhenius-type flow stress model were 0.9952 and 5.78%, which were poorer linear relation and more deviation than 0.9997 and 1.04% for the feed-forward back-propagation ANN model, respectively. The results indicated that the trained ANN model was more efficient and accurate in predicting the flow behavior for TC4-DT titanium alloy at elevated temperature deformation than the strain-compensated Arrhenius-type constitutive equations. The constitutive relationship compensating strain could track the experimental data across the whole hot working domain other than that at high strain rates (≥1 s−1). The microstructure analysis illustrated that the deformation mechanisms existed at low strain rates (≤0.1 s−1), where dynamic recrystallization occurred, were far different from that at high strain rates (≥1 s−1) that presented bands of flow localization and cracking along grain boundary.

Artificial neural networks predictive models. A case study: carbon and bromine concentrations prediction based on chlorination time

Artificial neural networks (ANNs) are being used increasingly to predict water variables. This study
 offers an alternative approach to quantify the relationship between time of chlorination in potable
 water (due to convectional treatment procedure) and chlorination by-products concentration
 (expressed as carbon and bromine) with an ANN model, i.e., capturing non-linear relationships
 among the water quality variables.
 Thus, carbon and bromine concentrations in potable water (the second chosen due to the toxicity of
 brominated trihalomethanes, THMs) were predicted using artificial neural networks (ANNs) based
 mainly on multi-layer perceptrons (MLPs) architecture. The chlorination (detention) time as much as
 58 hours in Athens distributed network, comprised the input variables to the ANNs models.
 Moreover, to develop an ANN model for estimating carbon and bromine, the available data set was
 partitioned into training, validation and test set. In order to reach an optimum amount of hidden
 layers or nodes, different architectures were tested. The quality of the ANN simulations was
 evaluated in terms of the error in the validation sample set for the proper interpretation of the results.
 The calculated sum-squared errors for training, validation and test set were 0.056, 0.039 and 0.060
 respectively for the best model selected.
 Comparison of the results showed that a two-layer feed-forward back propagation ANN model could
 be used as an acceptable model for predicting carbon and bromine contained in potable water
 THMs.

Prediction of flow stress in a wide temperature range involving phase transformation for as-cast Ti–6Al–2Zr–1Mo–1V alloy by artificial neural network

The isothermal compressions of as-cast Ti–6Al–2Zr–1Mo–1V titanium alloy in a wide temperature range of 1073–1323K and strain rate range of 0.01–10s−1 with a reduction of 60% were conducted on a Gleeble-1500 thermo-mechanical simulator. The flow stress shows a complex non-linear intrinsic relationship with strain, strain rate and temperature, meanwhile the strain-softening behavior articulates dynamic recrystallization mechanism in α phase, dynamic recovery mechanism in β phase and their comprehensive function during phase transformation (α+β). Based on the experimental data, an artificial neural network (ANN) was trained with standard back-propagation learning algorithm to generalize the complex deformation behavior characteristics. In the present ANN model, strain and temperature were taken as inputs, and flow stress as output. A comparative study has been made on ANN model and improved Arrhenius-type constitutive model, and their predictability has been evaluated in terms of correlation coefficient (R) and average absolute relative error (ARRE). During α, α+β and β phase regime, R-value and ARRE-value for the improved Arrhenius-type model are 0.9824% and 6.02%, 0.9644% and 21.02%, and 0.9627% and 12.38%, respectively, while the R-value and ARRE-value for the ANN model are 0.9992% and 0.91%, 0.9996% and 1.47%, and 0.9975% and 2.17%, respectively. The predicted strain–stress curves outside of experimental conditions articulate the similar intrinsic relationships with experimental strain–stress curves. The results show that the feed-forward back-propagation ANN model can accurately tracks the experimental data in a wide temperature range and strain rate range associated with interconnecting metallurgical phenomena, and in further it has a good capacity to model complex hot deformation behavior of titanium alloy outside of experimental conditions.

A comparative study on Arrhenius-type constitutive model and artificial neural network model to predict high-temperature deformation behaviour in Aermet100 steel

For predicting high-temperature deformation behaviour in Aermet100 steel, the experimental stress–strain data from isothermal hot compression tests on a Gleeble-3800 thermo-mechanical simulator, in a wide range of temperatures (1073–1473 K) and strain rates (0.01–50 s −1), were employed to develop the Arrhenius-type constitutive model and artificial neural network (ANN) model, and their predictability for high-temperature deformation behaviour of Aermet100 steel was further evaluated. The predictability of two models was quantified in terms of correlation coefficient ( R) and average absolute relative error ( AARE). The R and AARE for the Arrhenius-type constitutive model were found to be 0.9861 and 7.62% respectively, while the R and AARE for the feed-forward back-propagation ANN model are 0.9995 and 2.58% respectively. The breakdown of the Arrhenius-type constitutive model at the instability regimes (i.e. at 1073 K and 1173 K in 0.1, 1, 10 and 50 s −1, and at 1373 K in 50 s −1) is possibly due to that physical mechanisms in the instability regimes, where microstructure exhibits cracking, shear bands and twin kink bands, are far different from that of the stability regimes where dynamic recovery and recrystallization occur. But the feed-forward back-propagation ANN model can accurately track the experimental data across the whole hot working domain, which indicates it has good capacity to model the complex high-temperature deformation behaviour of materials associated with various interconnecting metallurgical phenomena like work hardening, dynamic recovery, dynamic recrystallization, flow instability, etc.

Prediction of the hot deformation behavior for Aermet100 steel using an artificial neural network

Using the experimental data obtained from hot compression tests in the temperature range 800–1200 °C, strain range 0.05–0.90, and strain rate range 0.01–50 s −1 , an artificial neural network (ANN) model is developed to predict the hot deformation behavior of the ultrahigh strength steel of Aermet100. The inputs of the neural network are strain, strain rate and temperature, whereas flow stress is the output. The developed feed-forward back-propagation ANN model is trained with Levenberg–Marquardt learning algorithm. The performance of the ANN model is evaluated using a wide variety of standard statistical indices. Results show that the ANN model can efficiently and accurately predict hot deformation behavior of Aermet100. Finally the extrapolation ability and noise sensitivity of the ANN model are also investigated. It is found that the extrapolation ability is very high in the proximity of the training domain, and the noise tolerance ability very robust.

Application of Artificial Neural Network Model to Study Arsenic Contamination in Groundwater of Malda District, Eastern India

The extent of arsenic contamination in the groundwater has been estimated using artificial neural network (ANN) based on multi-layer perceptron (MLP) architecture. The input data to the ANN comprised samples collected from different arsenic affected blocks of Malda district, West Bengal, India. Each data sample consisted of the amount of arsenic contaminant observed in the groundwater together with other geochemical parameters believed to have some relationship with the corresponding arsenic contaminant. Here, the inputs to the ANN were observed values for pH (acidic-alkalinity ratio), specific conductivity, total dissolved solids (TDS), salinity, dissolved oxygen (DO), redox potential (Eh), and depth of tube well water, while the expected output for training the ANN was the amount of corresponding arsenic contaminant observed in the groundwater. Using the back propagation technique, the ANN model was trained with a subset of the input data. The trained ANN model was then used to estimate the arsenic contamination in groundwater beyond the specified training data. The quality of the ANN simulations was evaluated in terms of three different error measures; namely, the root mean square error, the mean absolute error, and the percent mean relative error for proper interpretation of the results. We have also used two other methods for prediction; namely, multiple linear regression and active set support vector regression. Amongst the three methods, the ANN model exhibited better prediction results for predicting the arsenic contamination in groundwater. Based on this methodology, it is possible to show that a four-layer feed-forward back propagation ANN model could be used as an acceptable prediction model for estimating the arsenic contamination in groundwater.