Developed Neural Network Model Research Articles

Investigating the effects of superplasticizer and recycled plastics on the compressive strength of cementitious composites using neural networks

The potential application of neural network (NN) models to estimate the compressive strength (CS) of cementitious composites under a variety of experimental settings and cement mixes was investigated. The data were extensively collected from previous literature, and the bootstrap resampling tests were applied to estimate the statistics of the parameter correlations. We find that the NN model that involves the coarse and fine natural aggregates (CA and FA), superplasticizer (SP) and recycled plastics (RP) as the features can accurately predict the CS (R2 ∼ 0.9), without the need to specify the type of SP and the structure of RP in advance. The developed NN model holds promise for revealing the global dependency of CS on these parameters. It suggested that increasing 100 kg/m3 of CA could increase CS by ∼4 MPa, but the usage of CA more than 700 kg/m3 could negatively affect CS. How the CS varying with FA is apparently nonlinear. Within the optimum limit, adding 1 kg/m3 of SP could enhance the CS by ∼2 MPa. Contrarily, additional 1 kg/m3 of RP results in a decrease of ∼0.2 MPa of CS. The mixture-type independent models developed here would broaden our understanding of the global influential-sensitivity among these variables and help save cost and time in the industrial applications.

Automatic defect detection in a steel sample using frequency-modulated thermal wave imaging

Non-stationary thermal wave imaging (NSTWI) techniques are primarily used to assess material properties and structural integrity without damaging a structure. Frequency-modulated thermal wave imaging (FMTWI) is a well-known NSTWI approach that uses a low-peak power heat source to examine structures in a reasonable experimentation time. Recently, various methods, such as pulse compression, Fourier transform, principal component analysis (PCA) and independent component analysis (ICA), have been introduced to handle the non-linearity of transient thermal signatures. However, handling non-linearity and developing a fully automatic defect detection system remains very challenging due to certain limitations of the aforementioned methods. To overcome these problems, this paper proposes an artificial neural network (ANN) for the identification of subsurface flaws in a mild steel sample investigated using the FMTWI approach. The accuracy and the performance of the proposed neural network (NN) are evaluated through a confusion matrix and region of convergence (ROC) analysis for the classification of defective and healthy pixels in an infrared image sequence. The developed NN model has achieved 99.7% accuracy in classifying the pixels correctly.

Neural-network-based mixed subgrid-scale model for turbulent flow

An artificial neural-network-based subgrid-scale (SGS) model, which is capable of predicting turbulent flows at untrained Reynolds numbers and on untrained grid resolution is developed. Providing the grid-scale strain-rate tensor alone as an input leads the model to predict a SGS stress tensor that aligns with the strain-rate tensor, and the model performs similarly to the dynamic Smagorinsky model. On the other hand, providing the resolved stress tensor as an input in addition to the strain-rate tensor is found to significantly improve the prediction of the SGS stress and dissipation, and thereby the accuracy and stability of the solution. In an attempt to apply the neural-network-based model trained for turbulent flows with a limited range of the Reynolds number and grid resolution to turbulent flows at untrained conditions on untrained grid resolution, special attention is given to the normalisation of the input and output tensors. It is found that the successful generalization of the model to turbulence for various untrained conditions and resolution is possible if distributions of the normalised inputs and outputs of the neural network remain unchanged as the Reynolds number and grid resolution vary. In a posteriori tests of the forced and the decaying homogeneous isotropic turbulence and turbulent channel flows, the developed neural-network model is found to predict turbulence statistics more accurately, maintain the numerical stability without ad hoc stabilisation such as clipping of the excessive backscatter, and to be computationally more efficient than the algebraic dynamic SGS models.

Optimization of a vertical axis wind turbine with a deflector under unsteady wind conditions via Taguchi and neural network applications

Vertical axis wind turbines (VAWTs), so named because of their vertical axis of rotation, are a sustainable, opportune, and versatile means of producing energy. Their operation is not dependent on wind direction, making them suitable for use in settings with turbulent and inconsistent winds (e.g., urban locations), and they can be installed at the bottom of towers for easier installation and maintenance. However, unsteady wind may cause a vertical axis wind turbine (VAWT) to operate under drag-controlled conditions and reduce its performance. The power coefficient of a VAWT under unsteady wind conditions is heavily impacted by the tip speed ratio (TSR). Understanding and optimizing TSR is critical to making VAWTs a more viable and attractive option for sustainable energy production. Deflectors have been shown to improve the aerodynamic performance of wind turbines. In the present study, the Taguchi method is used in the experimental design, and a high-fitting neural network (NN) model based on computational fluid dynamics (CFD) data is adopted to predict the optimal mean TSR for a VAWT operation with a deflector. The amplitude and frequency fluctuations of the mean inlet velocity are used to specify the unsteady wind conditions. The results show that the imposed unsteady wind reduces the average power coefficient (Cp-) of the VAWT. By applying the Taguchi method and NN analysis to the impact of unsteady wind conditions, it is found that the mean TSR (TSRmean) is the factor producing the greatest impact on Cp-. The optimal TSRmean is evaluated by the NN model. In light of the recommendation from the NN predictions, the Cp- value from CFD can be improved by up to 3.58 folds under the optimal TSRmean. Furthermore, the relative errors of predicted Cp- values between the NN and CFD simulation are less than 4%, showing the reliability of predictions of the developed NN model in efficiently calculating the optimal operation for a VAWT.

Neural Network Based Estimation of Service Life of Different Metal Culverts in Arkansas

The Arkansas Department of Transportation (ARDOT) uses different types of metal culverts and cross-drains. Service lives of these culverts are largely influenced by the corrosion of the metals used in these culverts. Corrosion of metallic parts in any soil-water environment is governed by geochemical and electrochemical properties of the soils and waters. Many transportation agencies including ARDOT primarily focus on investigating the physical and mechanical properties of soils rather than their chemical aspects. The main objective of this study is to analyze the geotechnical and geochemical properties of soils in Arkansas to estimate the service lives of different metal pipes in different conditions. Soil resistivity values were predicted after analyzing the United States Department of Agriculture (USDA) soil survey data using neural network (NN) models. The developed NN models were trained and verified by using laboratory test results of soil samples collected from ARDOT, and survey data were obtained from the USDA. The service lives of metal culverts were then estimated based on the predicted soil properties and water quality parameters extracted from the data acquired from the Arkansas Department of Environmental Quality (ADEQ). Finally, Geographic Information System-based corrosion risk maps of three different types of metal pipes were developed based on their estimated service lives. The developed maps will help ARDOT engineers to assess the corrosion potential of the metal pipes before starting the new construction and repair projects and will allow using proper culvert materials to maximize their life spans.

Machine learning aided supercritical water gasification for H2-rich syngas production with process optimization and catalyst screening

Hydrogen production from wet organic wastes through supercritical water gasification (SCWG) promotes sustainable development. However, it is always time-consuming and expensive to achieve optimal SCWG conditions and suitable catalysts for different wastes to produce H2-rich syngas. Herein, we developed a unified machine learning (ML) framework to predict syngas composition from SCWG processes inclusive of non-catalytic and catalytic systems. The neural network (NN) model which was the core of the framework exhibited generalizable and satisfactory accuracy (R2 > 0.85) for all systems evaluated. The model also aided in the screening of catalyst and provided optimal conditions for accelerating experiments to produce H2-rich syngas by maximizing H2 yield and minimizing CO2 yield. ML model-based exploration found that SCWG temperature and solid content in the feedstock were the two most important factors affecting syngas composition. ML-based optimization suggested that Fe-based catalyst exhibited a greater potential to promote SCWG of wet wastes under optimal operational conditions. Besides, a web-based graphic user interface was developed by embedding the developed NN model for free access to the SCWG scientific community.

Neural Network-Assisted Development of High-Entropy Alloy Catalysts: Decoupling Ligand and Coordination Effects

<h2>Summary</h2> High-entropy alloys (HEAs) recently emerged as promising catalysts due to their immense chemical space and tunability. However, the large chemical space presents challenges for comprehensive characterization due to experiments' trial-and-error nature. Here, we leverage neural network (NN) and density functional theory to simultaneously account for ligand effect (spatial arrangement of different elements) and coordination effect (different crystal facets and defects) for predicting the adsorption energy. The developed NN model demonstrates three advantages: (1) high accuracy, with a mean absolute error of 0.09 eV; (2) universality, with applicability to bimetallic catalysts; and (3) simplicity, with 36 NN parameters and its further simplification to a linear scaling model at a slight loss of accuracy. Using the trained NN model validated with experimental literature, we decouple the comparative extents of ligand and coordination effects. Our results endow high practical significance and provide important insights for rational design of HEA catalysts.

From deep learning to transfer learning for the prediction of skeletal muscle forces.

Skeletal muscle forces may be estimated using rigid musculoskeletal models and neural networks. Neural network (NN) approach has the advantages of real-time estimation ability and promising prediction accuracy. However, most of the developed NN models are based on conventional feedforward NNs, which do not take dynamic temporal relationships of the muscle force profiles into consideration. The objectives of this present paper are twofold: (1) to develop a recurrent deep neural network (RDNN) incorporating dynamic temporal relationships to estimate skeletal muscle forces from kinematics data during a gait cycle; (2) then to establish a transfer learning strategy to improve the accuracy of muscle force estimation. A long short-term memory (LSTM) model as a RDNN was developed and evaluated. A weight transfer strategy was established. Three databases were established for training and evaluation purposes. The predictions of rectus femoris, soleus, and tibialis anterior forces with developed LSTM network show root mean square error range of 2.4-84.6N. Relative root mean square error (RMSE) deviations for internal and external validations are less than 5% and 10% for all analyzed muscles respectively. Pearson correlation coefficients (R) range of 0.95-0.999 showed perfect waveform similarity between data and predicted muscle forces for all analyzed muscles. The use of weight transfer leads to an improvement of 1.3% for the relative deviation between simulation outcome and LSMT prediction. This present study suggests that the recurrent deep neural network is a powerful and accurate computational tool for the prediction of skeletal muscle forces. Moreover, the coupling between this deep learning approach and a transfer learning strategy leads to improve the prediction accuracy. In future work, this coupling approach will be incorporated into a developed decision support tool for functional rehabilitation with real-time estimation and tracking of skeletal muscle forces. Graphical abstract ᅟ.

Genotype-driven identification of a molecular network predictive of advanced coronary calcium in ClinSeq\xae and Framingham Heart Study cohorts

BackgroundOne goal of personalized medicine is leveraging the emerging tools of data science to guide medical decision-making. Achieving this using disparate data sources is most daunting for polygenic traits. To this end, we employed random forests (RFs) and neural networks (NNs) for predictive modeling of coronary artery calcium (CAC), which is an intermediate endo-phenotype of coronary artery disease (CAD).MethodsModel inputs were derived from advanced cases in the ClinSeq®; discovery cohort (n=16) and the FHS replication cohort (n=36) from 89th-99th CAC score percentile range, and age-matched controls (ClinSeq®; n=16, FHS n=36) with no detectable CAC (all subjects were Caucasian males). These inputs included clinical variables and genotypes of 56 single nucleotide polymorphisms (SNPs) ranked highest in terms of their nominal correlation with the advanced CAC state in the discovery cohort. Predictive performance was assessed by computing the areas under receiver operating characteristic curves (ROC-AUC).ResultsRF models trained and tested with clinical variables generated ROC-AUC values of 0.69 and 0.61 in the discovery and replication cohorts, respectively. In contrast, in both cohorts, the set of SNPs derived from the discovery cohort were highly predictive (ROC-AUC ≥0.85) with no significant change in predictive performance upon integration of clinical and genotype variables. Using the 21 SNPs that produced optimal predictive performance in both cohorts, we developed NN models trained with ClinSeq®; data and tested with FHS data and obtained high predictive accuracy (ROC-AUC=0.80-0.85) with several topologies. Several CAD and “vascular aging" related biological processes were enriched in the network of genes constructed from the predictive SNPs.ConclusionsWe identified a molecular network predictive of advanced coronary calcium using genotype data from ClinSeq®; and FHS cohorts. Our results illustrate that machine learning tools, which utilize complex interactions between disease predictors intrinsic to the pathogenesis of polygenic disorders, hold promise for deriving predictive disease models and networks.

Assessment of solar data estimation models for four cities in Iran

AbstractThe estimated solar resources are important for designing renewable energy systems since measured data are not always available. The estimation models have been introduced in several studies. These models are mainly dependent on local meteorological data and need to be assessed for different locations and times. The current study compares the results of Angstrom's model and a neural network (NN) model developed for this study with measured data for four cities in Iran. The time resolution for the estimated global horizontal insolation is monthly. The results show that the developed NN model has promising performance and considering the calibration process for Angstrom's model it can be used as an alternative. The NN model uses climatic data to estimate the solar insolation which makes it more flexible in terms of being applicable for different regions. (© 2015 WILEY‐VCH Verlag GmbH &amp; Co. KGaA, Weinheim)

Development of Hyperelastic Model for Natural Rubber Containing Weldlines

Several constitutive models of non-linear large elastic deformation based on strain-energy-density functions have been developed for hyperelastic materials. These models, coupled with the Finite Element Method (FEM), can effectively utilized by design engineers to analyze and design elastomeric products operating under the deformation states. However, due to the complexities of the mathematical formulation which can only obtained at the moderate strain and the assumption of material used for the analysis. Therefore it is formidable task for design engineer to make use of these constitutive relationships. In the present work, the strain-energy-density function of natural rubber part containing a weldline was constructed by using the Neural Network (NN) model. The analytical results were compared to those obtained by Neo-Hookean, Mooney-Rivlin, Ogden models. Good agreement between developed NN model and the existing experimental data was found, especially at very low strain and at very high strain.

Development of Hyperelastic Model for Weldlines Containing Natural Rubber Molded Part

Several constitutive models of non-linear large elastic deformation based on strain-energy-density functions have been developed for hyperelastic materials. These models, coupled with the Finite Element Method (FEM), can effectively utilized by design engineers to analyze and design elastomeric products operating under the deformation states. However, due to the complexities of the mathematical formulation which can only obtained at the moderate strain and the assumption of material used for the analysis. Therefore it is formidable task for design engineer to make use of these constitutive relationships. In the present work, the strain-energy-density function of weldline containing rubber part was constructed by using the Neural Network (NN) model. The analytical results were compared to those obtained by Neo-Hookean, Mooney-Rivlin, Ogden models. Good agreement between developed NN model and the existing experimental data was found, especially at very low strain and at very high strain.

Prediction of stress intensity factors in pavement cracking with neural networks based on semi-analytical FEA

Computation of the stress intensity factors (SIFs) at the crack tip is the basis for pavement crack propagation analysis. Due to the three-dimensional (3-D) nature of cracked pavements and traffic loading, two-dimensional (2-D) finite element analysis (FEA) may be too simple to precisely predict SIFs, and the best choice for calculating the SIFs seems to be 3-D FEA programs. However, the 3-D FEA solutions are often computationally heavy. We had previously developed a semi-analytical FEA with multi-variable regression approach to fill this gap, but its accuracy still needs to be improved. To address this problem, a neural network approach based on semi-analytical FEA is presented in this paper: firstly, a SIFs database was generated through analyzing varieties of pavement structures using elastic semi-analytical FEA program; secondly, from the results in the database, neural network (NN) based SIF equations were developed for practical applications. The determination coefficients (R2) of all the developed NN models were greater than 0.99 and mean square error (MSE) values were less than 1e−4. The comparisons between the prediction results from NN models and multivariable regression models also showed the advantage of NN over multivariable regression on the prediction accuracy. This proposed NN SA-FEA SIF prediction approach has been developed as a pavement crack propagation analysis tool.

Application of neural networks for the prediction of cartilage stress in a musculoskeletal system

Traditional finite element (FE) analysis is computationally demanding. The computational time becomes prohibitively long when multiple loading and boundary conditions need to be considered such as in musculoskeletal movement simulations involving multiple joints and muscles. Presented in this study is an innovative approach that takes advantage of the computational efficiency of both the dynamic multibody (MB) method and neural network (NN) analysis. A NN model that captures the behavior of musculoskeletal tissue subjected to known loading situations is built, trained, and validated based on both MB and FE simulation data. It is found that nonlinear, dynamic NNs yield better predictions over their linear, static counterparts. The developed NN model is then capable of predicting stress values at regions of interest within the musculoskeletal system in only a fraction of the time required by FE simulation.

Neural network prediction of buckling load of steel arch-shells

Calculating the buckling load of structures is one of the main aspects of geometric structural instability. With reference to finite element software (ANSYS) and considering the weight of the structure as a constant load, the paper calculates the buckling load of steal arch shells as the main scope. Then, the article makes a prediction for the buckling load of steel arch shells applying 85 datasets prepared by the above mentioned method and using an artificial neural network by means of NeuroSolution 5.0 software. The radius of the periphery cylinder of an arch shell, the thickness of the shell and its internal angel are considered as inputs of construct models. Next, in order to attain an optimum model calculating and comparing the value of Root Mean Squared Error, RMSE, Mean Absolute Error, MAE and the coefficient of correlation, R2, for all 1 and 2-layer constructed models, a model having architecture 3-7-1 was found to be optimum. A high confirmed correlation with calculated buckling loads employing the finite element method and considering the calculated values of RMSE (0.0126), MAE (0.011) and the obtained value R2 (0.998) demonstrated high efficiency of a new developed neural network model.

KNEE JOINT CONTACT MECHANICS IN A MALALIGNED LIMB: AN INTEGRATED FINITE ELEMENT AND IN VITRO STUDY

Traditional finite element (FE) analysis is computationally demanding. The computational time becomes prohibitively long when multiple loading and boundary conditions need to be considered such as in musculoskeletal movement simulations involving multiple joints and muscles. Presented in this study is an innovative approach that takes advantage of the computational efficiency of both the dynamic multibody (MB) method and neural network (NN) analysis. A NN model that captures the behavior of musculoskeletal tissue subjected to known loading situations is built, trained, and validated based on both MB and FE simulation data. It is found that nonlinear, dynamic NNs yield better predictions over their linear, static counterparts. The developed NN model is then capable of predicting stress values at regions of interest within the musculoskeletal system in only a fraction of the time required by FE simulation.

Estimating the geoeffectiveness of halo CMEs from associated solar and IP parameters using neural networks

Abstract. Estimating the geoeffectiveness of solar events is of significant importance for space weather modelling and prediction. This paper describes the development of a neural network-based model for estimating the probability occurrence of geomagnetic storms following halo coronal mass ejection (CME) and related interplanetary (IP) events. This model incorporates both solar and IP variable inputs that characterize geoeffective halo CMEs. Solar inputs include numeric values of the halo CME angular width (AW), the CME speed (Vcme), and the comprehensive flare index (cfi), which represents the flaring activity associated with halo CMEs. IP parameters used as inputs are the numeric peak values of the solar wind speed (Vsw) and the southward Z-component of the interplanetary magnetic field (IMF) or Bs. IP inputs were considered within a 5-day time window after a halo CME eruption. The neural network (NN) model training and testing data sets were constructed based on 1202 halo CMEs (both full and partial halo and their properties) observed between 1997 and 2006. The performance of the developed NN model was tested using a validation data set (not part of the training data set) covering the years 2000 and 2005. Under the condition of halo CME occurrence, this model could capture 100% of the subsequent intense geomagnetic storms (Dst ≤ −100 nT). For moderate storms (−100 &lt; Dst ≤ −50), the model is successful up to 75%. This model's estimate of the storm occurrence rate from halo CMEs is estimated at a probability of 86%.

Postprocessing of Near-Field Measurement Based on Neural Networks

This paper presents postprocessing based on neural network (NN) models to reconstruct the magnetic near-field profile with an improved spatial resolution for one or different frequencies. The models aim at decreasing the time required to perform near-field electromagnetic compatibility (EMC) measurements. The multilayer perceptron (MLP) NNs are used to determine the magnetic near field radiated by passive devices and power electronics components. An optimization method, called the split-sample method, is implemented to determine the structures of the NN. The results obtained with the proposed method are compared with the measurement results. A graphic interface (GUI) is created to simplify the utilization of the developed NN models.

Triaxial compression behavior of sand and tire wastes using neural networks

Tire waste additions to sand enhance the shear strength of sand for embankments. Granular and fiber shape tire wastes and their mixture with sand under drained and undrained conditions were tested in triaxial compression apparatus and modeled using neural networks (NN). In the experimental study, tire crumb and tire buffings inclusions were used at varying contents as soil reinforcement. Both quick tests and consolidated drained (CD) triaxial tests were performed to analyze the effects of tire content, tire shape, and tire aspect ratio on the shear strength of sand. Then, this extensive experimental database obtained in laboratory was used in training, testing, and prediction phases of three neural network-based soil models. The input variables in the developed NN models are tire wastes content, tire wastes type, test type, effective stress, and axial strain, and the output is the deviatoric stress. The accuracy of proposed models seems to be satisfactory. Furthermore, the proposed models are also presented as simple explicit mathematical functions for further use by researchers.

Neural Networks Modeling of Stress Growth in Asphalt Overlays due to Load and Thermal Effects during Reflection Cracking

Although several techniques have been introduced to reduce reflective cracking, one of the primary forms of distress in hot-mix asphalt (HMA) overlays of flexible and rigid pavements, the underlying mechanism and causes of reflective cracking are not yet well under- stood. Fracture mechanics is used to understand the stable and progressive crack growth that often occurs in engineering components under varying applied stress. The stress intensity factor (SIF) is its basis and describes the stress state at the crack tip. This can be used with the appropriate material properties to calculate the rate at which the crack will propagate in a linear elastic manner. Unfortunately, the SIF is difficult to compute or measure, particularly if the crack is situated in a complex three-dimensional (3D) geometry or subjected to a non- simple stress state. In this study, the neural networks (NN) methodology is successfully used to model the SIF as cracks grow upward through a HMA overlay as a result of both load and thermal effects with and without reinforcing interlayers. Nearly 100,000 runs of a finite-element program were conducted to calculate the SIFs at the tip of the reflection crack for a wide variety of crack lengths and pavement structures. The coefficient of determination (R 2 ) of all the developed NN models except one was above 0.99. Owing to the rapid prediction of SIFs using developed NN models, the overall computer run time for a 20-year reflection cracking prediction of a typical overlay was significantly reduced. DOI: 10.1061/(ASCE)MT.1943-5533.0000153. © 2011 American Society of Civil Engineers. CE Database subject headings: Cracking; Neural networks; Predictions; Models; Finite element method; Thermal factors. Author keywords: Reflective cracking; HMA overlay; Neural networks; Prediction model; Fracture mechanics; Finite element.