Neural Network Sensitivity Analysis Research Articles

Sensitivity analysis and performance evaluation of neural networks for predicting forest stand volume - A case study: District 2, Kacha, Guilan province, Iran

Sensitivity analysis and performance evaluation of neural networks for predicting forest stand volume - A case study: District 2, Kacha, Guilan province, Iran

Neural Network Modeling of a Stereolithography Printed Mesostructure

Abstract This paper addresses the scarcity of comprehensive studies on the collective impact of various parametric lattice designs on mesostructure functionality. Focusing on optimizing the energy absorption of a serpentine mesostructure made using stereolithography, this research leverages a feedforward neural network to explore the interplay between line width, number of turns, and material properties on the energy absorbed by the structure. Compression simulations using a finite element model, covering a range of configurations, provided the dataset for neural network training. The resulting network was used to probe correlations between geometric variables, material, and energy absorption. Additionally, a neural network sensitivity analysis explored the impact of hidden layers and number of neurons on the network's performance, demonstrating the network's robustness. The optimized mesostructure configuration, identified by the neural network, maximized energy absorption. Using foundational mechanics of materials concepts, the discussion explains how the geometry and material of the cellular mesostructure affect structural stiffness.

Investigating the effect of fundamental properties of materials on the mechanisms of thermal cracking of asphalt mixtures

In most of tests and models to evaluate the thermal cracking of asphalt mixture, the consideration of different elements within the asphalt mixture has been neglected to provide a corrective solution. Accordingly, in this research, the evaluation of the mechanisms of thermal cracking involved assessing the impact of surface free energy (SFE) parameters and other fundamental properties of bitumen and aggregate. These parameters included the adhesion and cohesion free energy, the ratio of the base element to the acid element of the aggregate SFE (RBA), the creep stiffness of bitumen, the rate of changes in creep stiffness (m-value), and specific surface area (SSA) of aggregate. To this end, Wilhelmy plate (WP) test, universal sorption device (USD) test, the bending beam rheometer (BBR) test and the pull-off test were performed. In the following, models were presented to evaluate the effect of mentioned parameters as independent variables on the mechanisms of thermal cracking using an artificial neural network (ANN). The sensitivity analysis of the Artificial Neural Network (ANN) model revealed that in the status of adhesive failure, the cohesion free energy and m-value did not affect the Pull-off adhesion strength. On the other hand, the adhesion free energy, RBA, and the SSA were directly proportional and the creep stiffness was inversely proportional to the Pull-off adhesion strength. Also, in the status of cohesive failure, the adhesion free energy, the RBA, and the SSA did not affect the Pull-off cohesion strength. On the other hand, the cohesion free energy and m-value were directly proportional and the creep stiffness was inversely proportional to the Pull-off cohesion strength.

Artificial neural network simulation and sensitivity analysis for optimal thermal transport of magnetic viscous fluid over shrinking wedge via RSM

PurposeThis study aims to model the important flow response quantities over a shrinking wedge with the help of response surface methodology (RSM) and an artificial neural network (ANN). An ANN simulation for optimal thermal transport of incompressible viscous fluid under the impact of the magnetic effect (MHD) over a shrinking wedge with sensitivity analysis and optimization with RSM has yet not been investigated. This effort is devoted to filling the gap in existing literature.Design/methodology/approachA statistical experimental design is a setup with RSM using a central composite design (CCD). This setup involves the combination of values of input parameters such as porosity, shrinking and magnetic effect. The responses of skin friction coefficient and Nusselt number are required against each parameter combination of the experimental design, which is computed by solving the simplified form of the governing equations using bvp4c (a built-in technique in MATLAB). An empirical model for Cfx and Nux using RSM and ANN adopting the Levenberg–Marquardt algorithm based on trained neural networks (LMA-TNN) is attained. The empirical model for skin friction coefficient and Nusselt number using RSM has 99.96% and 99.99% coefficients of determination, respectively.FindingsThe values of these matrices show the goodness of fit for these quantities. The authors compared the results obtained from bvp4c, RSM and ANN and found them all to be in good agreement. A sensitivity analysis is performed, which shows that Cfx as well as Nux are most affected by porosity. However, they are least affected by magnetic parameters.Originality/valueThis study aims to simulate ANN and sensitivity analysis for optimal thermal transport of magnetic viscous fluid over shrinking wedge.

Classification and prediction of the effects of nutritional intake on diabetes mellitus using artificial neural network sensitivity analysis: 7th Korea National Health and Nutrition Examination Survey.

This study aimed to predict the association between nutritional intake and diabetes mellitus (DM) by developing an artificial neural network (ANN) model for older adults. Participants aged over 65 years from the 7th (2016-2018) Korea National Health and Nutrition Examination Survey were included. The diagnostic criteria of DM were set as output variables, while various nutritional intakes were set as input variables. An ANN model comprising one input layer with 16 nodes, one hidden layer with 12 nodes, and one output layer with one node was implemented in the MATLAB® programming language. A sensitivity analysis was conducted to determine the relative importance of the input variables in predicting the output. Our DM-predicting neural network model exhibited relatively high accuracy (81.3%) with 11 nutrient inputs, namely, thiamin, carbohydrates, potassium, energy, cholesterol, sugar, vitamin A, riboflavin, protein, vitamin C, and fat. In this study, the neural network sensitivity analysis method based on nutrient intake demonstrated a relatively accurate classification and prediction of DM in the older population.

Sensitivity Analysis of Artificial Neural Networks Identifying JWH Synthetic Cannabinoids Built with Alternative Training Strategies and Methods

This paper presents the alternative training strategies we tested for an Artificial Neural Network (ANN) designed to detect JWH synthetic cannabinoids. In order to increase the model performance in terms of output sensitivity, we used the Neural Designer data science and machine learning platform combined with the programming language Python. We performed a comparative analysis of several optimization algorithms, error parameters and regularization methods. Finally, we performed a new goodness-of-fit analysis between the testing samples in the data set and the corresponding ANN outputs in order to investigate their sensitivity. The effectiveness of the new methods combined with the optimization algorithms is discussed.

Application of Artificial Neural Network Sensitivity Analysis to Identify Key Determinants of Harvesting Date and Yield of Soybean (Glycine max [L.] Merrill) Cultivar Augusta

Genotype and weather conditions play crucial roles in determining the volume and stability of a soybean yield. The aim of this study was to identify the key meteorological factors affecting the harvest date (model M_HARV) and yield of the soybean variety Augusta (model M_YIELD) using a neural network sensitivity analysis. The dates of the start of flowering and maturity, the yield data, the average daily temperatures and precipitation were collected, and the Selyaninov hydrothermal coefficients were calculated during a fifteen-year study (2005–2020 growing seasons). During the experiment, highly variable weather conditions occurred, strongly modifying the course of phenological phases in soybean and the achieved seed yield of Augusta cultivar. The harvesting of mature soybean seeds took place between 131 and 156 days after sowing, while the harvested yield ranged from 0.6 t·ha−1 to 2.6 t·ha−1. The sensitivity analysis of the MLP neural network made it possible to identify the factors which had the greatest impact on the tested dependent variables among all the analyzed factors. It was revealed that the variables assigned ranks 1 and 2 in the sensitivity analysis of the neural network forming the M_HARV model were total rainfall in the first decade of June and the first decade of August. The variables with the highest impact on the Augusta soybean seed yield (model M_YIELD) were the mean daily air temperature in the second decade of May and the Seljaninov coefficient values calculated for the sowing–flowering date period.

Shape optimization of airfoil in ground effect based on free-form deformation utilizing sensitivity analysis and surrogate model of artificial neural network

Wings operate in proximity to surfaces for using ground effect to enhance lift-to-drag ratio, but the stability meets challenges. Changing airfoil shape could satisfy the requirement of stability and maximize lift-to-drag ratio, like performing single-objective optimization under a constraint. This research uses free-form deformation technique to adjust the airfoil curve by control points, and reduces the dimension of control variables by sensitivity analysis. Sampling airfoils and corresponding aerodynamics feeds an artificial neural network. Then, the neural network plays as a surrogate to predict fitness value for genetic algorithm execution. It is found that lift-to-drag ratio and static stability height are sensitive to vertical adjustments near the leading edge, one quarter chord point and trailing edge. The trained two-hidden-layer MLP generalizes well. The deformed optimum airfoil with S-shape camber line reduces lift-to-drag ratio to gain adequate static stability. The cause is that the aerodynamic center of pitch and that of altitude are moved upstream together, while the former’s interval is longer than the latter’s. The procedure provides reference for the optimization of airfoil under more states in ground effect zone, and the S-type deformation offers guidance for refinement on wing-in-ground stability.

Coupled effects of fly ash and calcium formate on strength development of cemented tailings backfill

Cemented tailings backfill (CTB) is widely adopted to ensure the safety of underground goafs and mitigate environmental risks. Fly ash (FA) and calcium formate (CF) are common industrial by-products that improve the mechanical performance of CTB. How the coupling of the two components affects the strength development is not yet well-understood. Neural network modelling was conducted to predict the strength development, including the static indicator of uniaxial compressive strength (UCS) and the dynamic indicator of ultrasonic pulse velocity (UPV). Sobol' sensitivity analysis was carried out to reveal the contributions of FA, CF and curing time to CTB strength. SEM microstructure investigation on CTB samples was implemented to reveal the mechanism of strength development and justify the predictions by neural network modelling and sensitivity analysis. Results show that the combination of FA content, CF content and curing time can be used to predict both UCS and UPV while providing adequate accuracy. The maximum of UCS of 6.1215MPa is achieved at (FA content, CF content, curing time) = (13.78 w%, 3.76 w%, 28days), and the maximum of UPV of 2.9887km/s is arrived at (FA content, CF content, curing time) = (11.67 w%, 3.08 w%, 10days). It is also implicated that prediction of UCS using UPV alone, although common in field application is not recommended. However, UPV measurement, in combination with the information of FA dosage, CF dosage and curing time, could be used to improve UCS prediction. The rank of variable significance for UCS is curing time > FA content > CF content, and for UPV is FA content > curing time > CF content; variable interaction is strongest for FA with CF for UCS development, and for FA with curing time for UPV evolution. Influence of FA on CTB strength development is due to improved polymerisation and consumption of Ca(OH)2. Influence of CF on strength development is a result of accelerated hydration and increased combined-water content in calcium silicate hydrate (CSH). Effect of curing time is attributed to the evolution of CSH product and pore-water content during cement hydration.

Investigation of average asphalt film thickness of dense graded asphalt mixtures with compaction effects

The asphalt mixtures with inadequate durability may become too lean, resulting in low rutting and fatigue resistance. The asphalt film thickness (AFT) has been used as a crucial property for the durability of asphalt mixtures. However, the relationship between AFT and volumetric properties is ambiguous with the absence of compaction effects in the standard film thickness equation. This study aims to investigate the relationship between AFT and volumetric properties of dense-graded asphalt mixtures with consideration of the compaction degree. The compaction-property-dependent (CPD) AFT model considering compaction effects was proposed and compared with the standard AFT equation. Then, asphalt mixtures with 112 different gradations were used to calculate the AFT with and without compaction effects. The relationship between new AFT and volumetric parameters (Pb, VMA, VFA, Vc, Vfine and Vfiller) was investigated by linear correlation analysis. After that, two regression models and a back-propagation (BP) neural network were used to predict the new AFT with volumetric parameters. By comparing the prediction accuracy of these models, a reasonable model with higher prediction accuracy was selected. Results show that the new AFT is 2% to 30% less than the AFT in the standard model, and the surface area of air void boundary has a negative effect on the new AFT. The correlation analysis indicates that the new AFT of dense-graded asphalt mixtures has a rough correlation with changes in VMA, binder content and VFA. The prediction accuracy of the BP neural network has the highest R-squared. The relative importance and sensitivity analysis of the BP neural network verified a relatively good relationship between VMA and new AFT. These findings can give a considerable interest for the tendency of using the new AFT to supplement the current specified VMA criteria.

Predicting of acid red 14 removals from synthetic wastewater in the advanced oxidation process using artificial neural networks and fuzzy regression

The titanium dioxide (TiO2) photocatalyst reactor is a widely used approach for eliminating non-biodegradable compounds in water and wastewater. Predicting the performance of the reactor in removing acid red 14 (AR14) was investigated using radial basis function (RBF), Adaptive Neuro-Fuzzy Inference System (ANFIS), and fuzzy regression analysis. The input variables in the models were radiation, irradiation time, the initial concentration of nanoparticle TiO2 to cement ratio (nTiO2/c), pH, initial concentration of AR14, flow rate, and initial concentration of oxidizing peroxydisulfate. In addition, for reducing the overfitting of the training process, we used K-fold cross-validation. The results of using the three methods indicated that the ANFIS performance was more suitable in comparison with the fuzzy regression models, such that the coefficient of determination (R2), index of agreement (IA), Nash–Sutcliffe efficiency (E) for model efficiency, Mean Squared Error (MSE), and Mean Bias Error (MBE) for training between the observed data and predicted data reached 0.965, 0.991, 0.73, 0.0132 and 0.019, respectively. Normalized input data improved slightly the training processes and performance of RBF compared with non-normalized input data. The ANFIS neural network for training process and model performance slightly achieved better results than the RBF neural network. The performance of the fuzzy regression approach was less suitable in comparison with RBF and ANFIS modeling. The sensitivity analysis of the RBF neural network depicted that the initial concentration of peroxydisulfate (S2O8) was the most effective factor (25.6%) in predicting the removal of AR14 from synthetic wastewater. After that, the effective factors were irradiation time (12.44%), initial concentration of AR14 (10.37%), pH (10.37), flow rate (8.55%), and the weight ratio of the initial concentration of nanoparticle TiO2 to cement ratio (nTiO2/c) (1.75%), respectively.

NeuralSens: Sensitivity Analysis of Neural Networks

This article presents the NeuralSens package that can be used to perform sensitivity analysis of neural networks using the partial derivatives method. The main function of the package calculates the partial derivatives of the output with regard to the input variables of a multi-layer perceptron model, which can be used to evaluate variable importance based on sensitivity measures and characterize relationships between input and output variables. Methods to calculate partial derivatives are provided for objects trained using common neural network packages in R, and a 'numeric' method is provided for objects from packages which are not included. The package also includes functions to plot the information obtained from the sensitivity analysis. The article contains an overview of techniques for obtaining information from neural network models, a theoretical foundation of how partial derivatives are calculated, a description of the package functions, and applied examples to compare NeuralSens functions with analogous functions from other available R packages.

Uncertainty propagation and sensitivity analysis of an artificial neural network used as wind turbine load surrogate model

Recent studies have shown the advantage of replacing aeroelastic simulations with regression models based on Artificial Neural Networks (ANNs), which can be used as surrogate models for fast and efficient wind turbine load assessments. Once trained on a high-fidelity load simulation database covering a broad range of conditions, the surrogate model can be applied to predict loads for any site with wind climate falling within the range covered by the database. The aim of this study is to quantify the uncertainty propagation through such an ANN and to analyse how much the selected input variables influence the variance of the fatigue blade load estimations by means of a global sensitivity analysis. Results confirm that the selected ANN architecture seems suitable for this task resulting in small output uncertainties. Furthermore, the sensitivity analysis shows that the turbulence is mainly responsible for the blade load estimation, followed by the wind shear and the wind speed. The contributions of the turbulence length scale, turbulence anisotropy factor and wind veer angle are comparatively low. Comparing three different methods for sensitivity analysis shows that the partial derivative algorithm, Sobol variance decomposition and Shapley effect result in similar sensitivity measures.

Sensitivity analysis of the artificial neural networks in a system for durability prediction of forging tools to forgings made of C45 steel

The article presents the results of a sensitivity analysis of artificial neural networks developed for a system which predicts the durability of forging tools used in the selected hot die forging process. The developed system makes it possible to calculate the geometric loss of the examined tool for the given values of its operating parameters (number of forgings, tool temperature at selected points, type of the applied protective layer, pressure and path of friction) and estimates the intensity of the occurrence of typical mechanisms of tool destruction, i.e. thermo-mechanical fatigue, mechanical wear, abrasive wear and plastic deformation. Nine neural networks operate in the developed system. Five of them determine the geometric loss of the material used for tools operating with protective layers, including a nitrided layer, a pad welded layer and three hybrid layers, i.e. AlCrTiSiN, Cr/CrN and Cr/AlCrTiN. Four networks make calculations determining the intensity of the occurrence of typical destructive mechanisms. The developed sensitivity analysis allows for each neural network to show which input parameters are most important and have the greatest impact on the explained variables. This is determined based on the network error analysis in the case of elimination of individual variables from the input data. The greater the network error calculated after rejecting an input variable relative to the error obtained for the network with all the input variables, the more sensitive the network to the lack of this variable. The best compliance was obtained for the first developed set of networks regarding the geometric loss of material, while the lowest compliance was obtained for the second developed set of networks regarding the applied protective layers, and in particular for plastic deformation and mechanical fatigue, probably due to the smallest size of these sets in the knowledge base. The obtained results of this analysis are important for the system operation, i.e. supporting the technologist’s decision in the selection of such process parameter values that will increase the die’s lifetime.

Sensitivity Analysis of Radial Basis Function Networks for River Stage Forecasting

Sensitivity analysis of neural networks to input variation is an important research area as it goes some way to addressing the criticisms of their black-box behaviour. Such analysis of RBFNs for hydrological modelling has previously been limited to exploring perturbations to both inputs and connecting weights. In this paper, the backward chaining rule that has been used for sensitivity analysis of MLPs, is applied to RBFNs and it is shown how such analysis can provide insight into physical relationships. A trigonometric example is first presented to show the effectiveness and accuracy of this approach for first order derivatives alongside a comparison of the results with an equivalent MLP. The paper presents a real-world application in the modelling of river stage shows the importance of such approaches helping to justify and select such models.

An ecologically constrained procedure for sensitivity analysis of Artificial Neural Networks and other empirical models.

Sensitivity analysis applied to Artificial Neural Networks (ANNs) as well as to other types of empirical ecological models allows assessing the importance of environmental predictive variables in affecting species distribution or other target variables. However, approaches that only consider values of the environmental variables that are likely to be observed in real-world conditions, given the underlying ecological relationships with other variables, have not yet been proposed. Here, a constrained sensitivity analysis procedure is presented, which evaluates the importance of the environmental variables considering only their plausible changes, thereby exploring only ecological meaningful scenarios. To demonstrate the procedure, we applied it to an ANN model predicting fish species richness, as identifying relationships between environmental variables and fish species occurrence in river ecosystems is a recurring topic in freshwater ecology. Results showed that several environmental variables played a less relevant role in driving the model output when that sensitivity analysis allowed them to vary only within an ecologically meaningful range of values, i.e. avoiding values that the model would never handle in its practical applications. By comparing percent changes in MSE between constrained and unconstrained sensitivity analysis, the relative importance of environmental variables was found to be different, with habitat descriptors and urbanization factors that played a more relevant role according to the constrained procedure. The ecologically constrained procedure can be applied to any sensitivity analysis method for ANNs, but obviously it can also be applied to other types of empirical ecological models.

Dimension reduction and classification of hyperspectral images based on neural network sensitivity analysis and multi-instance learning

Hyperspectral remote image sensing is a rapidly developing integrated technology used widely in numerous areas. The rich spectral information from hyperspectral images aids in recognition and classification of many types of objects, but the high dimensionality of these images leads to information redundancy. In this paper, we used sensitivity analysis for dimension reduction. However, another challenge is that hyperspectral images identify objects as either a "different body with the same spectrum" or "same body with a different spectrum." Therefore, it is difficult to maintain the correct correspondence between ground objects and samples, which hinders classification of the images. This issue can be addressed using multi-instance learning for classification. In our proposed method, we combined neural network sensitivity analysis with a multiinstance learning algorithm based on a support vector machine to achieve dimension reduction and accurate classification for hyperspectral images. Experimental results demonstrated that our method provided strong overall classification effectiveness when compared with prior methods.

Sensitivity analysis of artificial neural networks for just-suspension speed prediction in solid-liquid mixing systems: Performance comparison of MLPNN and RBFNN

Just-suspension speed (Njs) is an important parameter for stirred tank design using a solid-liquid mixing system in the chemical process industry. However, current correlations for Njs suffer from uncertainty from limited experimental databases and limitations due to many parameters that play an important role in Njs determination. A comprehensive computation of the radial basis function neural network (RBFNN) was developed based on solid-liquid mixing experiments, which contain 935 datasets for the prediction of Njs. The Njs values were obtained experimentally using Zwietering correlation with different solid loading percentages, solid particle density, solid particle diameter, mixing solvent density, number of impeller blades, impeller diameter, impeller blade hub angle, impeller blade tip angle, the width of the impeller blade and the ratio of the clearance between the impeller and the bottom of the tank with the tank diameter. The RBFNN proved to have a much better ability to accurately predict the desired Njs compared to MLPNN even after decreasing the number of input variables from 11 to 8. Thus, the computational RBFNN model results will be useful for extending the application of a solid-liquid mixing system for estimating the just-suspension speed for stirred tank design.

Artificial Neural Network Modeling and Sensitivity Analysis of Performance and Emissions in a Compression Ignition Engine Using Biodiesel Fuel

In the present research work, a neural network model has been developed to predict the exhaust emissions and performance of a compression ignition engine. The significance and novelty of the work, with respect to existing literature, is the application of sensitivity analysis and an artificial neural network (ANN) simultaneously in order to predict the engine parameters. The inputs of the model were engine load (0, 25, 50, 75 and 100%), engine speed (1700, 2100, 2500 and 2900 rpm) and the percent of biodiesel fuel derived from waste cooking oil in diesel fuel (B0, B5, B10, B15 and B20). The relationship between the input parameters and engine cylinder performance and emissions can be determined by the network. The global sensitivity analysis results show that all the investigated factors are effective on the created model and cannot be ignored. In addition, it is found that the most emissions decreased while using biodiesel fuel in the compression ignition engine.

Hyperspectral Image Recognition Based on Artificial Neural Network

This paper aims to reduce the dimensions of the data in remote sensing hyperspectral image (HIS), and solve the information redundancy caused by the numerous bands in the image. To this end, the neural network sensitivity analysis (NNSA) was introduced to simplify the dimensionality reduction process. Meanwhile, the convolutional neural network (CNN) was adopted as the classification algorithm for the HIS, seeking to prevent the complex data reconstruction of feature extraction and classification. Then, the proposed method was contrasted with several other classification methods in several experiments. The results show that the proposed method outperformed the contrast plans in classification accuracy. Thus, the artificial neural network (ANN) is good at reducing the dimensions of remote sensing HSI and the CNN is a reliable classification tool. The research findings shed new light on remote sensing image processing and other related operations.