Batch Back Propagation Research Articles

Scaling an Artificial Neural Network-Based Water Quality Index Model from Small to Large Catchments

Scaling models is one of the challenges for water resource planning and management, with the aim of bringing the developed models into practice by applying them to predict water quality and quantity for catchments that lack sufficient data. For this study, we evaluated artificial neural network (ANN) training algorithms to predict the water quality index in a source catchment. Then, multiple linear regression (MLR) models were developed, using the predicted water quality index of the ANN training algorithms and water quality variables, as dependent and independent variables, respectively. The most appropriate MLR model has been selected on the basis of the Akaike information criterion, sensitivity and uncertainty analyses. The performance of the MLR model was then evaluated by a variable aggregation and disaggregation approach, for upscaling and downscaling proposes, using the data from four very large- and three large-sized catchments and from eight medium-, three small- and seven very small-sized catchments, where they are located in the southern basin of the Caspian Sea. The performance of seven artificial neural network training algorithms, including Quick Propagation, Conjugate Gradient Descent, Quasi-Newton, Limited Memory Quasi-Newton, Levenberg–Marquardt, Online Back Propagation, and Batch Back Propagation, has been evaluated to predict the water quality index. The results show that the highest mean absolute error was observed in the WQI, as predicted by the ANN LM training algorithm; the lowest error values were for the ANN LMQN and CGD training algorithms. Our findings also indicate that for upscaling, the aggregated MLR model could provide reliable performance to predict the water quality index, since the r2 coefficient of the models varies from 0.73 ± 0.2 for large catchments, to 0.85 ± 0.15 for very large catchments, and for downscaling, the r2 coefficient of the disaggregated MLR model ranges from 0.93 ± 0.05 for very large catchments, to 0.97 ± 0.02 for medium catchments. Therefore, scaled models could be applied to catchments that lack sufficient data to perform a rapid assessment of the water quality index in the study area.

Artificial neural network method modeling of microwave-assisted esterification of PFAD over mesoporous TiO2‒ZnO catalyst

An artificial neural network (ANN) was employed to predict biodiesel yield through microwave-assisted esterification of palm fatty acid distillate (PFAD) oil over TiO2‒ZnO mesostructured catalyst. The experimental data of biodiesel content (%) was carried out via changing three input factors (i.e. methanol:PFAD molar ratio, catalyst concentration, and reaction time). The results indicated that ANN is an appropriate approach for modeling and optimizing fatty acid methyl ester (FAME) yield performed over the microwave-assisted esterification process. The network was trained by five different algorithms (i.e. batch backpropagation (BBP), incremental backpropagation (IBP), Levenberg‒Marquardt (LM), genetic algorithm (GA), and quick propagation (QP)). The evaluation disclosed that the QP algorithm gave the least root mean squared error (RMSE), absolute average deviation (AAD), and the highest determination coefficient (R2) for both training and testing data groups. The confirmation test results of the ANN-based on QP-3-10-1 revealed that the RMSE, AAD, and the highest R2 were 0.741, 0.776, and 0.997, correspondingly. All in all, QP‒3‒10‒1 model offered the best possible mathematical qualities amongst all algorithms. Over this method, the FAME yield was determined at 97.45% (relating to the actual FAME yield of 97.33%) which was attained over 3 wt% mesoporous TiO2‒ZnO catalyst, methanol:PFAD molar ratio of 9:1 within 25 min of operating time. The esterification reaction conditions predicted by ANN showed to be potential for modeling and predicting FAME yield with an extremely well precision of 97.06%.

The Use of Artificial Neural Networks to Determine In-Hospital Mortality After Coronary Artery Bypass Surgery

The aim of this study was to present an artificial neural network (ANN) model for the accurate estimation of in-hospital mortality and to demonstrate the validity of the model with real data and a comparison with conventional multiple linear regression models. Retrospective clinical study. University hospital. None. Data were collected from the medical records of 88 patients who had undergone coronary artery bypass graft surgery with an extracorporeal cardiopulmonary pump between January 2018 and March 2020. An ANN approach was used to assess the association between in-hospital mortality and variables from preoperative, intraoperative, and postoperative data garnered retrospectively from patient files. The study examined the data of 88 patients with a mean age of 62.4 ± ten years, 60 (68.1%) of whom were men and 28 (31.8%) of whom were women. An examination of the average success of the training algorithms in the training, validation, and test sets revealed that the quick propagation algorithm ranked first with 97.397%. The algorithm that best matched the present study's dataset was the batch back propagation algorithm, with an average of 99.622 (in other words, this training set accurately estimated 99.622% of every 100 items of data). Furthermore, the rates continuously were greater than 90% when the probability of estimating the estimated output was examined. The ANN model tended to outperform multiple linear regression models in predicting in-hospital mortality among patients who have undergone coronary artery bypass graft surgery. Physicians can make use of this information as an aid in performing treatments and ensuring that more accurate quality of surgical care is achieved.

Neural network approach for optimizing machining parameters of aluminium nano composite

In the present study an integrated approach based on Taguchi modelling and Artificial Neural Networks (ANN) are used for modelling the experiments and output parameters of Abrasive Water Jet Machining (AWJM). For this process Al7075 nano composite is fabricated by using SiC as reinforcement. Composite is fabricated using ultrasonic cavitation method. The fabricated composite is machined by AWJM based on Taguchi L18 Orthogonal Array (OA) approach. For ANN modelling the data was selected according to the Taguchi L18 orthogonal array i.e., the selected input parameters for ANN are Abrasive Jet Speed, Abrasive Flow Rate, Stand of Distance, Pressure and the outputs from the ANN model are Average KERF width, Material Removal Rate (MRR). Batch Back Propagation is used for Training, Validation and Testing of Artificial neural network. From ANN the correlation coefficient R obtained is above 96% that implies the predicted values are 96% in agreement with the experimental values.

OPTIMIZATION OF PROCESS PARAMETERS IN LASER WELDING OF HASTELLOY C-276 USING ARTIFICIAL NEURAL NETWORK AND GENETIC ALGORITHM

In this paper, an effort is made to determine the optimized parameters in laser welding of Hastelloy C-276 using Artificial Neural Network (ANN) and Genetic Algorithm (GA). CO2 Laser welding was performed on a sheet of thickness 1.6[Formula: see text]mm based on Taguchi L27 orthogonal array. Laser power, welding speed and shielding gas flow rate were chosen as input parameters and Bead width, depth of Penetration and Microhardness were measured for assessing the weld quality. ANN was applied for modeling the welding process parameters i.e. heat input, welding speed and gas flow rate. Various learning algorithms such as Batch Back Propagation (BBP), Incremental Back Propagation (IBP), Quick Propagation (QP) and Levenberg–Marquardt (LM) were comprehensively tested for estimating the output parameters and a comparison was also made among them, with respect to prediction accuracy. BBP method was found to be the best learning algorithm. Experimental validation test was performed based on the ANN and GA predicted optimized responses and this welding input parameters provided satisfactory weld metal characteristics in terms of penetration depth, bead width and microhardness.

Lipase-catalysed synthesis of a novel galanthamine derivative: Process optimization by artificial neural networks

Artificial neural networks (ANNs) analysis was carried out to optimize the esterification of galanthamine and acetic acid in a solvent system. To predict performance parameters of the enzymatic reaction conditions, several parameters were studied which were reaction temperature (50–90 °C), enzyme amount (2–5 wt%), reaction time (6–18 h), and substrate molar ratio of galanthamine to acetic acid (2–5:1). The algoritms used in the network were batch back propagation (BBP), incremental back propagation (IBP), genetic algorithm (GA), Levenberg–Marguardt (LM) and quick propagation (QP) algorithms. The configuration of 4 inputs, one hidden layer with 7 nodes, and 1 output using the batch back propagation (BBP) was determined as the optimum algorithm. The predicted and experimental percentage yield value were 60.24% and 60.36%, respectively. These results prove the validity of ANN model.

Role of trochoidal machining process parameter and chip morphology studies during end milling of AISI D3 steel

This study aims at discovering the effect of the trochoidal loop spacing parameter on Surface Roughness (SR), Specific Cutting Energy (SCE) and Temperature (T) during flat end milling operations. Twenty experimental runs were conducted based on the face centered central composite design (CCD) of response surface methodology (RSM). Artificial Neural Network (ANN) prediction modelling was created using four learning algorithms such as Batch Back Propagation Algorithm (BBPA), Quick Propagation Algorithm (QPA), Incremental Back Propagation Algorithm (IBPA) and Legvenberg–Marquardt back propagation Algorithm (LMBPA). The results were compared based on the value of Root mean square (RMSE) obtained for each learning algorithm and it was identified that LMBPA model produced least RMSE value. The predictive LMBPA neural network model was found to be capable of better predictions of surface roughness, temperature and specific cutting energy within the trained range. The Genetic algorithm(GA) gives the optimum parameters for conformation test and they are cutting speed of 41 m/min, feed rate of 136 mm/min and trochoidal loop spacing of 1.3 mm and error percentage between experimental and GA predicted values is 3.60% for surface roughness, 3.15% for specific cutting energy and 3.89% for temperature was found to be minimal. Scratches and serrated boundaries at both side of the chips were observed and laces, chip adhesion and side flow marks were found on machined surface.

Performance evaluation of hot mix asphaltic concrete incorporating cow bone ash (CBA) as partial replacement for filler

Given the current realities of incessant pavement distresses frequently experienced on Nigerian highways due to axle loads from heavy-duty vehicles, coupled with the menace of grave environmental pollution from abattoir solid wastes such as Cow-bones, the mechanistic properties of Cow Bone Ash (CBA) as partial replacement for filler in the production of asphaltic concrete via the Marshall Mix Design Method and Artificial Neural Networks (ANN) were investigated in this research. The conventional filler was partially replaced with CBA at 2.5%, 5%, 7.5%, 10%, 20%, 30%, 40% and 50% respectively, in the total mix. Sequel to the production of the bituminous concrete at the various proportions, the samples were submerged in water, in a water bath for 30 minutes at a temperature of 105° before conducting Marshall Stability and flow tests. Results revealed that the stability and flow of asphaltic concrete containing the CBA were greater than that of the concrete containing the conventional filler. Furthermore, the physical and volumetric properties of the mix also improved as CBA was observed to be finer than the conventional quarry dust, hence reduced the voids present in the mix and stiffened the bitumen film on the aggregate particles. The results obtained further showed that the conventional filler (quarry dust) can be replaced partially with CBA up to 50%. The ANN Model employed was trained and tested by quick propagation (QP) algorithm amongst others such as the Incremental Back Propagation (IBP), Batch Back Propagation (BBP), Levenberg Marquardt (quasi Newton) and genetic algorithm (GA). QP gave the least Root Mean Square Error (RMSE) at the shortest time. The statistical values obtained showed that the ANN model was able to efficiently study and predict the experimental data.

Optimization of cultivation conditions for monoclonal IgM antibody production by M1A2 hybridoma using artificial neural network.

Monoclonal antibody (McAb) has been established as one of the most successful therapeutic strategies for the treatment of cancer. M1A2 (McAb) as a new monoclonal antibody was designed to recognize heat shock protein (HSP60), but its optimum production condition has not been studied. In this study, the cell culture conditions for both Roswell Park Memorial Institute Medium (RPMI 1640) and Dulbecco's Modified Eagle Medium (DMEM) were optimized using artificial neural network (ANN) analysis to obtain maximum production of IgM McAb by hybridoma M1A2 cells. By using a central composite design, an experimental matrix with cultivation parameters of incubation time, temperature and fetal bovine serum (FBS) concentration on IgM McAb production was designed. The results was analysed by ANN network with different learning algorithms. From the analysis, batch back propagation (BBP) trained ANN composed of eight hidden nodes using a hyperbolic tangent sigmoid transfer function was capable to provide the highest McAb production for both RPMI and DMEM media. Under optimum conditions of 12.5% of FBS, at 33°C after 3(1/2) days of incubation, maximum McAb production (1132.69μg/ml) in DMEM was achieved. With PRMI 1640 medium, maximum McAb production (1105.12μg/ml) was achieved at optimum conditions of 11% of FBS, at 33°C after 4days of incubation. The results of this study will provide information for optimum culture conditions of M1A2 McAb production in both DMEM and RPMI 1640 media and also give some clues for the other hybridoma excreting antibodies in the development of in vitro cell culture.

Performance comparison of neural network training algorithms in the modeling properties of steel fiber reinforced concrete

Our study is aimed at modeling the effect of three contributory factors, namely aspect ratio, water cement ratio and cement content on the water intake/absorption, compressive strength, flexural strength, split tensile strength and slump properties of steel fiber reinforced concrete. Artificial neural network (ANN) as a multilayer perceptron normal feed forward network was integrated to develop a predictive model for the aforementioned properties. Five training algorithms belonging to three classes: gradient descent, Levenberg Marquardt (quasi Newton) and genetic algorithm (GA). The ANN configuration consists of the input layer with three nodes, a single hidden layer of ten nodes of the output layer with five nodes. The study also compared the performance of all algorithms with regards to their predicting abilities. The ANN training was done by splitting the experimental data into the training and testing set. The divergence of the RMSE between the output and target values of the test set was monitored and used as a criterion to stop training. Although the convergence speed of GA was far higher than all other training algorithm, it performed better in predicting the water intake/absorption, split tensile strength and slump properties. However, incremental back propagation (IBP) and batch back propagation (BBP) outperformed GA in predicting the compressive strength and flexural strength respectively. The overall performance of the training algorithm was assessed using the coefficient of determination and the absolute fraction of variance obtained for the test data set and GA was found to have the highest value of 0.94 and 0.92 respectively. In determining the properties fiber reinforced concrete according to GA–ANN implementation, the water/cement ratio played slightly more dominant role than the aspect ratio and this was followed by cement content.

Improved Time Training with Accuracy of Batch Back Propagation Algorithm Via Dynamic Learning Rate and Dynamic Momentum Factor

<span lang="EN-US">The man problem of batch back propagation (BBP) algorithm is slow training and there are several parameters needs to be adjusted manually, also suffers from saturation training.</span><span lang="EN-US">The learning rate and momentum factor are significant parameters for increasing the efficiency of the (BBP). In this study, we created a new dynamic function of each learning rate and momentum facor. We present the DBBPLM algorithm, which trains with a dynamic function for each the learning rate and momentum factor.<br /> A Sigmoid function used as activation function. The XOR problem, balance, breast cancer and iris dataset were used as benchmarks for testing the effects of the dynamic DBBPLM algorithm. All the experiments were performed on Matlab 2012 a. The stop training was determined ten power -5. From the experimental results, the DBBPLM algorithm provides superior performance in terms of training, and faster training with higher accuracy compared to the BBP algorithm and with existing works.</span>

Application of Genetic Algorithm Optimization Technique in TIG Welding of 15CDV6 Aerospace Steel

In this work, an attempt is made to identify the optimized parameter combinations in TIG (Tungsten Inert Gas) welding of 15CDV6 steel. 15CDV6 is a high strength low alloy steel which is widely used in aerospace industries. TIG welding was carried out on 15CDV6 steel with current, voltage, welding speed, shielding gas flow rate as input parameters. Tensile strength, hardness, and corrosion current density were measured as performance characteristics. Artificial Neural Network (ANN) was used as modeling technique for predicting the output parameters. Four different learning algorithms, viz. Batch Back Propagation (BBP), Quick Propagation (QP), Incremental Back Propagation (IBP), Legvenberg- Marquardt (LM) back propagation were used to train the model, and it was found that legvenberg- Marquardt (LM) as the best learning algorithm concerning the prediction accuracy. Genetic algorithm (GA) optimization technique was used to identify the optimized parameter combinations, and a confirmation test was carried out with GA predicted optimized parameters. The confirmation test results were in good agreement with GA results.

Artificial neural network prediction of wire electrical discharge machining properties on sintered porous NiTi shape memory alloy

This article presents the characterisation of porous nickel titanium shape memory alloy during wire electrical discharge machining process with different input process parameters such as servo voltage, pulse on time, pulse off time, current and wire speed. The performance of WEDM process on porous NiTi alloy prepared through powder metallurgy route was measured as the output characteristics such as material removal rate (MRR) and surface roughness (Ra). Taguchi design with L18 orthogonal array was used for depicting the WEDM experimental parameters with 5 factors and 2-3 mixed level for optimization. Artificial neural network model was developed to predict the performance of WEDM process under two methods of connections such as multilayer normal feed and full feed forward. ANN model was predicted using three learning algorithms such as Incremental Back Propagation (IBP), Batch Back Propagation (BBP) and Quick prop (QP) with variations of neurons from 5 to 20 in the hidden layer. Genetic algorithm was employed for optimization the WEDM process parameters and output variables those were closely matched with experimental values. Based on the results, servo voltage is the most influencing factor on MRR and Ra. The best algorithms were selected as IBP and BBP with 20 neurons for multilayer normal and full feed forward method respectively that values given as an input fitting function for GA. The confirmation test was executed using GA and those values were coincided with the experimental values. The best WEDM process parameters were optimized using full feed forward method to provide better MRR and product finish.

Neural networks approach to optimization of aerosolized nanocolloidal carrier system containing quercetin for pulmonary delivery of lung cancer

Event Abstract Back to Event Neural networks approach to optimization of aerosolized nanocolloidal carrier system containing quercetin for pulmonary delivery of lung cancer Mohd Basyaruddin Abdul Rahman1*, Noor Hafizah Arbain1, Norazlinaliza Salim1 and Wong Tin Hui2 1 Department of Chemistry, Faculty of Science, Universiti Putra Malaysia, Malaysia 2 Non-Destructive Biomedical and Pharmaceutical Research Centre, Integrative Pharmacogenomics Institute, Universiti Teknologi MARA, Malaysia Background Globally, lung cancer has become the most common type of cancer cases. Quercetin (QT) has been extensively investigated for its pharmacological effects on lung cancer. However, clinical applications of QT are limited due to poor solubility and low stability in aqueous medium. Hence, this study focused on the development of nanocolloidal carrier system to enhance the solubility of QT by pulmonary administration. Methods The artificial neural networks (ANNs) was carried out for development of a predictive model of aerosolized quercetin (QT) nanoemulsions to achieve minimum volume median diameter for targetting lung cancer by pulmonary delivery. The effects of the amount of the mixture of palm oil esters:ricinoleic acid (1.5–4.5 wt.%), lecithin (1.5–2.5 wt.%), Tween 80 (0.5–1 wt.%), glycerol (1.5–3 wt.%), and water (88–94.95 wt.%) on the volume median diameter were considered as inputs to the network. The volume median diameter of the samples with various compositions was measured as an output. The incremental back propagation (IBP), batch back propagation (BBP), quick propagation (QP), and genetic algorithm (GA) were used in the network. Results It was found that the optimal algorithm and topology were the genetic algorithm (GA) and the configuration with 5 inputs, 14 hidden, and 1 output nodes, respectively. The model obtained indicated the high-quality performance of the neural network and its capability to identify the critical composition factors for the nanoemulsion. The formulation containing quercetin was then successfully prepared using optimum composition and volume median diameter of 4.266 μm was obtained. The QT-loaded nanoemulsion induced cytotoxic response with an IC50 value of 300 μg/mL against A549 lung cancer cells at 48 hours. Conclusion These results suggest that the nanocolloidal formulation containing QT could be successful carrier system for pulmonary drug delivery application. Acknowledgements The financial assistances received from Ministry of Higher Education Malaysia through MyBRAIN 15 for Arbain N.H. This work was supported by Universiti Malaya, LRGS NanoMITe-Ministry of Higher Education, Malaysia (RU029-2014/5526306). Keywords: Aerosols, Quercetin, Nanoemulsions, Pulmonary delivery, artificial neural network Conference: International Conference on Drug Discovery and Translational Medicine 2018 (ICDDTM '18) “Seizing Opportunities and Addressing Challenges of Precision Medicine”, Putrajaya, Malaysia, 3 Dec - 5 Feb, 2019. Presentation Type: Poster Presentation Topic: Cancer Citation: Abdul Rahman M, Arbain N, Salim N and Tin Hui W (2019). Neural networks approach to optimization of aerosolized nanocolloidal carrier system containing quercetin for pulmonary delivery of lung cancer. Front. Pharmacol. Conference Abstract: International Conference on Drug Discovery and Translational Medicine 2018 (ICDDTM '18) “Seizing Opportunities and Addressing Challenges of Precision Medicine”. doi: 10.3389/conf.fphar.2018.63.00151 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 19 Nov 2018; Published Online: 17 Jan 2019. * Correspondence: Prof. Mohd Basyaruddin Abdul Rahman, Department of Chemistry, Faculty of Science, Universiti Putra Malaysia, Serdang, Malaysia, basya@upm.edu.my Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Mohd Basyaruddin Abdul Rahman Noor Hafizah Arbain Norazlinaliza Salim Wong Tin Hui Google Mohd Basyaruddin Abdul Rahman Noor Hafizah Arbain Norazlinaliza Salim Wong Tin Hui Google Scholar Mohd Basyaruddin Abdul Rahman Noor Hafizah Arbain Norazlinaliza Salim Wong Tin Hui PubMed Mohd Basyaruddin Abdul Rahman Noor Hafizah Arbain Norazlinaliza Salim Wong Tin Hui Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.

Dynamically-adaptive Weight in Batch Back Propagation Algorithm via Dynamic Training Rate for Speedup and Accuracy Training

The main problem of batch back propagation (BBP) algorithm is slow training and there are several parameters need to be adjusted manually, such as learning rate. In addition, the BBP algorithm suffers from saturation training. The objective of this study is to improve the speed up training of the BBP algorithm and to remove the saturation training. The training rate is the most significant parameter for increasing the efficiency of the BBP. In this study, a new dynamic training rate is created to speed the training of the BBP algorithm. The dynamic batch back propagation (DBBPLR) algorithm is presented, which trains with adynamic training rate. This technique was implemented with a sigmoid function. Several data sets were used as benchmarks for testing the effects of the created dynamic training rate that we created. All the experiments were performed on Matlab. From the experimental results, the DBBPLR algorithm provides superior performance in terms of training, faster training with higher accuracy compared to the BBP algorithm and existing works.

Predicting the optimum compositions of a transdermal nanoemulsion system containing an extract of Clinacanthus nutans leaves (L.) for skin antiaging by artificial neural network model

A nanoemulsion‐based formulation containing leaf extracts of Clinacanthus nutans Lindau (C. nutans) was prepared for therapeutic use and optimized by artificial neural network (ANN). The model was applied to optimize the particle size of the transdermal nanoemulsion system containing an extract of C. nutans leaves for skin antiaging. Five universal learning algorithms—incremental back propagation, batch back propagation, quick propagation, genetic algorithm, and Levenberg‐Marquardt—were used in the ANN to achieve the optimum topologies. The optimum topologies were selected among the learning algorithms trained with lowest root mean square values. Genetic Algorithm‐5‐13‐1 was found to be the optimum topology for the final model to predict the optimum particle size. Under these conditions, the actual particle size of the optimum nanoemulsion compared well with the predicted values with the residual standard error of less than 1.5%. The final formulation was observed to be stable against phase separation under an accelerated stability study and during storage at room temperature and 45°C for 90 days. The dermal irritation assay demonstrated that the developed nanoemulsion was found to be compatible with human skin and could be classified as nonirritant. In vivo ultrasound attributes of the skin study showed that the collagen content increased significantly with the application of the C. nutans nanoemulsion among all 21 volunteers during the 21 days of the treatment period. Hence, ANN was found to be an effective tool for optimizing the nanoemulsion containing C. nutans leaves extract with efficacy to protect collagen from breakdown.

Landslide susceptibility mapping at Ovacık-Karabük (Turkey) using different artificial neural network models: comparison of training algorithms

This study aims to investigate the performances of different training algorithms used for an artificial neural network (ANN) method to produce landslide susceptibility maps. For this purpose, Ovacik region (southeast of Karabuk Province), located in the Western Black Sea Region (Turkey), was selected as the study area. A total of 196 landslides were mapped, and a landslide database was prepared. Topographical elevation, slope angle, aspect, wetness index, lithology, and vegetation index parameters were taken into account for the landslide susceptibility analyses. Two different ANN structures, which were composed of single and double hidden layers, were applied to compare the effects of the ANN. Four different training algorithms, namely batch back-propagation, quick propagation, conjugate gradient descent (CGD), and Levenberg–Marquardt, were used for the training stage of the ANN models. Thus, eight different landslide susceptibility maps were produced for the study area using different ANN structures and algorithms. In order to assess the effects and spatial performances of the considered training algorithms on the ANN models, the relative operating characteristics (ROC) and relation value (rij) approaches were used. The susceptibility map produced by CGD1 has the highest AUC (0.817) and rij values (0.972). Comparison of the susceptibility maps indicated that CGD training algorithm is the slowest one among the other algorithms, but this algorithm showed the highest performance on the results.

A Review on Enhancements to Speed up Training of the Batch Back Propagation Algorithm

Objectives: The present review is focused on determining the efficiency of some of the parameters for enhancing the time and accuracy training in the batch back propagation (BP) algorithm. Methods: Researchers have used many methods, including heuristic methods, flat-spots, Fletcher -Powel and Quasi-Newton methods to enhance the BP algorithm for speeding up time training. The current heuristic method covers two techniques. The first focuses on choosing the suitable value for each training rate and momentum term, either together or individually. The second technique is to create a dynamic training rate with a penalty to avoid the local minimum. Findings: Slow training or fast training depends on the weight adjusted in the BP algorithm. The training rate and momentum are significant parameters for controlling the updated weight, but it is difficult to choose the suitable value to adjust the weight for improving the BP algorithm. If the weights are adjusted too small, the BP algorithm gives slow training; if the weight is over-adjusted, the BP algorithm gives faster training with an oscillating value of error training. The small or large adjustment of the weights is unsuitable for learning of the BP algorithm. Existing studies do not mention the relationship between the values of training rate and momentum term with gross weight. Gross weight leads to saturation training or reduction in training accuracy. This study suggests creating the dynamic training rate with boundary and momentum terms and then establishing the relationship between them to keep the weight adjusted moderate to avoid the gross the weight being updated. Improvements: This study will guide researchers to create a dynamic training rate and momentum term with an inverse relationship or boundary to escape gross weight training and maintain high accuracy training.

Application of Artificial Neural Network Modelling for Optimization of Yb: YAG Laser Welding of Nitinol

This article presents the investigation of Ytterbium: Yttrium aluminium garnet (Yb: YAG) laser welding of NiTinol sheet of thickness 1 mm. Welding speed, shielding gas blown distance, focal position, laser power were selected as input parameters and depth of penetration, bead width, hardness, corrosion current density were measured as performance characteristics. Experiment was designed based on L9 Taguchi design with 4 factors and 3 levels in each factor. Modelling was done using artificial neural network in four learning algorithms namely batch back propagation, quick propagation, incremental back propagation and Legvenberg-Marquardt back propagation. A comparison was made between these learning algorithms and it was found that based on least error, Legvenberg-Marquardt model was the best learning algorithm. Genetic algorithm was implemented for predicting the optimised laser welding parameters in joining NiTinol and the confirmation test results were in good agreement with the predicted results.

The comparison of performance by using alternative refrigerant R152a in automobile climate system with different artificial neural network models

Due to the refrigerant gas R134a which is used in automobile air conditioning systems and has greater global warming impact will be phased out gradually, an alternative gas is being desired to be used without much change on existing air conditioning systems. It is aimed to obtain the easier solution for intermediate values on the performance by creating a Neural Network Model in case of using a fluid (R152a) in automobile air conditioning systems that has the thermodynamic properties close to each other and near-zero global warming impact. In this instance, a network structure giving the most accurate result has been established by identifying which model provides the best education with which network structure and makes the most accurate predictions in the light of the data obtained after five different ANN models was trained with three different network structures. During training of Artificial Neural Network, Quick Propagation, Quasi-Newton, Levenberg-Marquardt and Conjugate Gradient Descent Batch Back Propagation methodsincluding five inputs and one output were trained with various network structures. Over 1500 iterations have been evaluated and the most appropriate model was identified by determining minimum error rates. The accuracy of the determined ANN model was revealed by comparing with estimates made by the Multi-Regression method.