Suitable Artificial Neural Network Research Articles

Optimized deep CNN for detection and classification of diabetic retinopathy and diabetic macular edema

Diabetic Retinopathy (DR) and Diabetic Macular Edema (DME) are vision related complications prominently found in diabetic patients. The early identification of DR/DME grades facilitates the devising of an appropriate treatment plan, which ultimately prevents the probability of visual impairment in more than 90% of diabetic patients. Thereby, an automatic DR/DME grade detection approach is proposed in this work by utilizing image processing. In this work, the retinal fundus image provided as input is pre-processed using Discrete Wavelet Transform (DWT) with the aim of enhancing its visual quality. The precise detection of DR/DME is supported further with the application of suitable Artificial Neural Network (ANN) based segmentation technique. The segmented images are subsequently subjected to feature extraction using Adaptive Gabor Filter (AGF) and the feature selection using Random Forest (RF) technique. The former has excellent retinal vein recognition capability, while the latter has exceptional generalization capability. The RF approach also assists with the improvement of classification accuracy of Deep Convolutional Neural Network (CNN) classifier. Moreover, Chicken Swarm Algorithm (CSA) is used for further enhancing the classifier performance by optimizing the weights of both convolution and fully connected layer. The entire approach is validated for its accuracy in determination of grades of DR/DME using MATLAB software. The proposed DR/DME grade detection approach displays an excellent accuracy of 97.91%.

Decision support system for ship energy efficiency management based on an optimization model

This paper introduces an innovative decision support system for improving ship energy efficiency. It combines an engine optimization model and an artificial neural network (ANN). Real-time data regarding the ship and engine performance is obtained along a specific ship navigation, and an engine optimization model is built using Ricardo Wave software. This model is validated with high accuracy. Ship-specific parameters are derived, and fuel consumption is estimated using ANN models with different structures. The study identifies the most suitable ANN model with 1 hidden layer and 5 neurons based on error metric evaluation where the 0.99697 R2, 0.00035 RMSE, and 0.06470 MAPE scores are calculated. This approach offers a cost-effective solution for shipping companies to monitor critical engine parameters in real-time without investing in sensors or data collection systems. Thus, it contributes to dealing with the problem of the scarcity of analyzable data in the maritime literature, which is one of the significant issues in the papers related to energy efficiency, machine learning, and deep learning. It is an innovative approach since the parameters that are related to real-time operations rather than considering only instruction book information are aggregated, and the produced data is analyzed by an intelligent tool.

Performance Evaluation of Artificial Neural Network Modelling to a Ploughing Unit in Various Soil Conditions

Abstract The specific objective of this study is to find a suitable artificial neural network model for estimating the operation indicators (disturbed soil volume, effective field capacity, draft force, and energy requirement) of ploughing units (tractor disc) in various soil conditions. The experiment involved two different factors, i.e., (Ι) soil texture index and (ΙΙ) field work index, and included soil moisture content, tractor engine power, soil bulk density, tillage speed, tillage depth, and tillage width, which were linked to one dimensionless index. We assessed the effectiveness of artificial neural network and multiple linear regression models between the values predicted and the actual values using the mean absolute error criterion to test data points. When the artificial neural network model was applied, the mean absolute error values for disturbed soil volume, effective field capacity, draft force, and energy requirement were 69.41 m3·hr−1, 0.04 ha·hr−1, 1.24 kN, and 1.95 kw·hr·ha−1, respectively. In order to evaluate the behaviour of new models, the coefficient R 2 was used as a criterion, where R 2 values in artificial neural network were 0.9872, 0.9553, 0.9948, and 0.9718, respectively, for the aforementioned testing dataset. Simultaneously, R 2 values in multiple linear regression were 0.7623, 0.696, 0.492, and 0.5572, respectively, for the same testing dataset. Based on these comparisons, it was clear that predictions using the artificial neural network models proposed are very satisfactory.

Neuromodel of an Eddy Current Brake for Load Emulation

The eddy current brake (ECB) is an electromechanical energy conversion device that can be used as a load emulator to load a motor according to the intended load scenario. However, conducting an analysis in the time domain is difficult due to its complex behavior involving mechanical, electrical, and magnetic phenomena. The challenges with the time domain analysis of the ECB require new modeling approaches that provide reliability, robustness, and controllability over a wide speed interval. If the ECB can be modeled with high accuracy, it can be controlled like a load emulator that can simulate nonlinear industrial loads. This paper describes a neuromodeling approach taken to develop an ECB. The nonlinear characteristic of the brake system was modeled with a high performance by using an artificial neural network (ANN), which is a potent nonlinear system identification tool. Several characteristics of a designed and optimized brake system undergoing various excitation currents in whole speed regions are described and verified experimentally. Eventually, an electromechanical brake system is proposed that aims to provide the required linear or nonlinear load model dynamics throughout an emulation process in line with the obtained neuromodel. To identify the most suitable ANN architecture for the problem, various ANN configurations, ranging from 1 neuron to 20 neurons in the hidden layer, as well as a statistical approach that differs from the existing literature, are presented. Additionally, the suggested model’s scalability is discussed.

Simulating the pyrolysis interactions among hemicellulose, cellulose and lignin in wood waste under real conditions to find the proper way to prepare bio-oil

The distribution of cellulose, hemicellulose, and lignin in the biomass structure was uneven. To explore the interactions of three components under real conditions during pyrolysis and find the proper way to prepare bio-oil, chemical methods were used to remove them in poplar wood and the remainders were pyrolyzed separately. The coke produced by hemicellulose pyrolysis inhibited the pyrolysis of lignin and cellulose, while the promoting synergistic effects between cellulose and lignin reduced the pyrolysis temperature (up to 23.2 °C) and activation energy (up to 15.07%). The content of light oil in the pyrolysis oil was higher and more conducive to combustion. XPS analysis showed that during the pyrolysis of lignin and cellulose, oxygen atoms were effectively fixed into the biochar and more benzene ring substances were in the bio-oil, thereby improving the quality of the pyrolysis oil. Finally, the most suitable artificial neural network (ANN) model was selected, and the pyrolysis behavior of the remaining substances after the removal of hemicellulose was studied in depth. It also showed that removing part of hemicellulose was more beneficial to the preparation of high-quality bio-oil.

Forward and backward mixed-mode crack estimation using artificial neural network

PurposeIn this manuscript, the authors aimed to demonstrate the influences of influential parameters in mixed-mode crack propagation phenomenon. The authors attempted to cover almost all surrounding issues of this subject as the authors know simulating of propagating cracks as internal strong discontinuity is a complicated issue.Design/methodology/approachIn this manuscript, the authors demonstrated the influences of influential parameters in mixed-mode crack propagation phenomenon. The authors attempted to cover almost all surrounding issues of this subject as the authors know simulating of propagating cracks as internal strong discontinuity is a complicated issue. Furthermore, three different scenarios for crack growth are considered. In reality, edge-cracked plate, center-cracked plate and cracked plate in the presence of void and inclusion are studied. In fact, by designing suitable artificial neural network's (ANN) architectures all the three aforementioned conditions are trained and estimated through those architectures with very good agreement with input data. Also by conducting a series of sensitivity analysis, the most affecting factors in mixed-mode crack propagation in different situations are demonstrated. The obtained results are very interesting and useful for other researchers and also the authors hope the results would be cited by researchers.FindingsThe influential parameters on mixed-mode crack propagation were found in this paper.Originality/valueThe computer code using MATLAB was prepared to study the mixed-mode crack paths. Also using ANNs toolbox, the crack path estimation was investigated.

Bearing Aluminum-Based Alloys: Microstructure, Mechanical Characterizations, and Experiment-Based Modeling Approach.

Due to the engine's start/stop system and a sudden increase in speed or load, the development of alloys suitable for engine bearings requires excellent tribological properties and high mechanical properties. Including additional elements in the Al-rich matrix of these anti-friction alloys should strengthen their tribological properties. The novelty of this work is in constructing a suitable artificial neural network (ANN) architecture for highly accurate modeling and prediction of the mechanical properties of the bearing aluminum-based alloys and thus optimizing the chemical composition for high mechanical properties. In addition, the study points out the impact of soft and more solid phases on the mechanical properties of these alloys. For this purpose, a huge number of alloys (198 alloys) with different chemical compositions combined from Sn, Pb, Cu, Mg, Zn, Si, Ni, Bi, Ti, Mn, Fe, and Al) were cast, annealed, and tested for determining their mechanical properties. The annealed sample microstructure analysis revealed the formation of soft structural inclusions (Sn-rich, Sn-Pb, and Pb-Sn phases) and solid phase inclusions (strengthened phase, Al2Cu). The mechanical properties of ultimate tensile strength (σu), Brinell hardness (HB), and elongation to failure (δ) were used as control responses for constructing the ANN network. The constructed network was optimized by attempting different network architecture designs to reach minimal errors. Besides the excellent tribological characteristics of the designed set of alloys, soft inclusions based on Sn and Pb and solid-phase Cu inclusions fulfilled the necessary level of mechanical properties for anti-friction alloys; the maximum mechanical properties reached were: σu = 197 ± 7 MPa, HB = 77 ± 4, and δ = 20.3 ± 1.0%. The optimal ANN architecture with the lowest errors (correlation coefficient (R) = 0.94, root mean square error (RMSE) = 3.5, and average actual relative error (AARE) = 1.0%) had two hidden layers with 20 neurons. The model was validated by additional experiments, and the characteristics of the new alloys were accurately predicted with a low level of errors: R ≥ 0.97, RMSE = 1-2.65, and AARE ˂ 10%.

Fiber-Optic Gyroscope Thermal Calibration through Two-Dimensional N-Order Polynomial for Landslide Displacement Monitoring

A fiber-optic gyroscope (FOG) with lower precision but higher cost advantage is typically selected according to working conditions and engineering budget. Thermal drift is the main factor affecting FOG precision. External thermal calibration methods by algorithms can effectively weaken the influence of thermal drift. This paper presents a thermal calibration method of a two-dimensional N-order polynomial (TDNP) and compares it with artificial neural network (ANN) methods to determine a software FOG thermal calibration method for landslide displacement monitoring. The TDNP thermal calibration coefficient matrix was established, and the thermal calibration capability of the TDNP method with different orders N was evaluated on the basis of error analysis. The ANN model with 1 to 18 hidden neural layers was established on the basis of LM, BR, and SCG algorithms to choose a suitable ANN. Finally, the mean absolute errors of FOG thermal calibration through the TDNP with different orders and the LM were compared. This method was applied in the Huangtupo landslide area, China. The results highlight that the TDNP method with order 5 had better performance and satisfied the requirements of landslide displacement monitoring. The research results can compensate for the lack of adaptability of the FOG thermal calibration method in landslide displacement monitoring.

Design of Artificial Neural Networks for Damage Estimation of Composite Laminates: Application to Delamination Failures in Ply Drops

This work presents a data-driven approach based on Artificial Neural Networks (ANN), that benefits from parametric non-linear finite element analyses, in order to provide a “cheaper” numerical alternative to the more expensive experimental testing of advanced composite laminates. The chosen subject of study is the damage evolution and associated delamination failures of ply drops. As such, physical-based modeling of these phenomena are used as training data for obtaining the most suitable ANN capable of estimating the damage evolution using only the prescribed initial conditions (i.e., laminate geometry, material orientation, applied loading). Additionally, this work aims at providing first-hand experience into the procedure that leads to obtaining such ANN. In order to do so, we detail each of the required steps, such as choosing a network architecture, defining a custom loss function, as well as deciding on the better learning parameters. Each stage on this process is guided by intuitive statistical analyses and simple criteria, such as selecting the “simpler” model over whenever a more complex one “improves” on the performance but only by a narrow margin. Indeed, for every step, multiple numerical experiences are provided so that the reader can also get a deeper understanding on the inner-workings of these ANN models. With this goal in mind, we employ dimensionality reduction techniques (PCA and t-SNE) to also propose a geometrical visualization of the nonlinear transformations performed by the ANN. Finally, additional tests regarding the network ability to generalize to unseen data showed that the optimal well-trained ANN is accurate and robust enough for near real-time predictions of the various damage evolution patterns, and outperforms other data-driven methods under comparison.

Optimisation of convolutional neural network architecture using genetic algorithm for the prediction of adhesively bonded joint strength

The classical method of optimising structures for strength is computationally expensive due to the requirement of performing complex non-linear finite element analysis (FEA). This study aims to optimise an artificial neural network (ANN) architecture to perform the task of predicting the strength of adhesively bonded joints in place of non-linear FEA. A manual multi-objective optimisation was performed to find a suitable ANN architecture design space. Then a genetic algorithm optimisation of the reduced design space was conducted to find an optimum ANN architecture. The generated optimum ANN architecture predicts efficiently the strength of adhesively bonded joints to a high degree of accuracy in comparison with the legacy method using FEA with a 93% savings in computational cost.

Data-driven recognition and modelling of deterioration patterns in the US National Bridge Inventory: A genetic algorithm-artificial neural network framework

Preserving infrastructures in a functioning and safe state requires maintenance or rehabilitation actions for the structures’ future needs. Thus, reliable estimations of deterioration rates are necessary for prioritizing structural interventions and assisting decision-makers in optimally allocating budgets. Among various approaches employed to model structural deterioration, statistical modelling using infrastructure databases is an increasingly utilized practice. Information contained within infrastructure databases usually concerns structural and operational attributes of the inventoried structures. Incorporating additional deterioration factors (e.g. environmental conditions) requires extracting content from other data sources. This process may result in large databases that entail complex nonlinear relationships among the included factors. The use of expert judgment or performing a subjective selection of deterioration factors from a small pre-specified set of factors has led to the generation of more manageable data collections, facilitating information processing. This study proposes an unbiased data-driven framework that combines Genetic Algorithms (GAs) and Artificial Neural Networks (ANNs) to simultaneously: (a) optimize the architecture of a suitable ANN model predicting infrastructure condition status and (b) perform a feature selection of essential deterioration factors among explanatory variables contained in the data collection used. The combined GA-ANN framework is applied to a dataset for 223,369 bridge structures located across the conterminous US region, which includes 53 variables with data from the National Bridge Inventory (NBI) and other reliable data sources. The results of the framework's application are assessed for their accuracy, efficiency and relevance, showing that artificial intelligence-based pattern recognition renders promising prospects for objectively capturing essential bridge deterioration information.

Artificial neural networks in the prediction of fraud in integral milk powder by adding whey powder

ABSTRACT: This research was performed to ascertain the most suitable Artificial Neural Network (ANN) model to quantify the degree of fraud in powdered milk through the addition of powdered whey via regular standard physicochemical analyses. In this study, an evaluation was done on 103 samples with different quantities of added whey powder to whole milk powder. Using Fourier Transform Infrared Spectroscopy the fat, cryoscopy, total solids, defatted dry extract, lactose, protein and casein were analyzed. The hyperbolic tangent transformation function was used with 45 topologies, and the Holdback and K-fold validation methods were tested. In the Holdback method, 75% of the database was employed for training, while 25% was used for validation. In the K-fold method, the database was categorized into five equal sized subsets, which alternated between training and validation. Of the two methods, the K-fold method was proven to have superior efficiency. Next, analysis was done on three models of multilayer perceptron networks with feedforward architecture. In Model 1, the input layer contained all the physicochemical analyses conducted, in model 2 the casein analysis was excluded, and in model 3 the routine analyses performed for dairy products was done (fat, defatted dry extract, cryoscopy and total solids). From Model 3 an ANN was derived which could satisfactorily predict fraud calculated from using the routine and standard analyses for dairy products, containing 64 nodes in the hidden layer, with R2 of 0.9935 and RMSE of 0.5779 for training, and R2 of 0.9964 and RMSE of 0.4358 for validation.

A Comparison of Three Neural Network Approaches for Estimating Joint Angles and Moments from Inertial Measurement Units.

The application of artificial intelligence techniques to wearable sensor data may facilitate accurate analysis outside of controlled laboratory settings—the holy grail for gait clinicians and sports scientists looking to bridge the lab to field divide. Using these techniques, parameters that are difficult to directly measure in-the-wild, may be predicted using surrogate lower resolution inputs. One example is the prediction of joint kinematics and kinetics based on inputs from inertial measurement unit (IMU) sensors. Despite increased research, there is a paucity of information examining the most suitable artificial neural network (ANN) for predicting gait kinematics and kinetics from IMUs. This paper compares the performance of three commonly employed ANNs used to predict gait kinematics and kinetics: multilayer perceptron (MLP); long short-term memory (LSTM); and convolutional neural networks (CNN). Overall high correlations between ground truth and predicted kinematic and kinetic data were found across all investigated ANNs. However, the optimal ANN should be based on the prediction task and the intended use-case application. For the prediction of joint angles, CNNs appear favourable, however these ANNs do not show an advantage over an MLP network for the prediction of joint moments. If real-time joint angle and joint moment prediction is desirable an LSTM network should be utilised.

Investment strategies applied to the Brazilian stock market: A methodology based on Sentiment Analysis with deep learning

The Efficient Market Hypothesis states that stock market changes reflect the arrival of new information through external events and news. Thus, many recent studies in the literature evaluate the impact of Sentiment Analysis (SA) applied to social media and news in the stock market. However, these studies generally do not present investment strategies that take advantage of sentiments in new publications considering the correlation between news and the stock market, specially when news are written in Portuguese. This paper proposes investment strategies based on Sentiment Analysis of financial news applied to the Brazilian stock market. For such, the following activities were performed: (i) identifying the most suitable Artificial Neural Network (ANN) architecture to perform Sentiment Analysis in financial news in Brazilian Portuguese; (ii) studying the correlation between the predominant sentiment in financial news of three major Brazilian news portals through the Granger causality test; (iii) proposing two categories of investment strategies based on Sentiment Analysis, considering both negative and positive financial news; and (iv) applying the proposed strategies to the Brazilian stock market. Experiments were conducted with financial news from the most popular Brazilian online news sources and the results showed: (i) the most appropriate ANN to perform SA in Portuguese is the Convolutional Neural Network; (ii) there is a significant influence of the predominant daily news sentiment in the stock market; and (iii) investment strategies based on Sentiment Analysis can bring profitability both in short and in long term, surpassing the strategies Random Walk and Buy & Hold.

Multivariate regression and artificial neural network to predict phenolic content in Schinus terebinthifolius stem bark through TLC images

This study aims to determine a suitable artificial neural network (ANN) regression model for quantification of polyphenols from herbal matrices through thin-layer chromatography (TLC) images. Gallic acid was used as a standard for the development and evaluation of this analytical approach. Analyses were performed on samples extracted from the stem bark of Schinus terebinthifolius, an important species in traditional Brazilian medicine. Calibration curves were obtained by TLC for image acquisition and assessment. High-performance liquid chromatography was used to verify the relationship between the data. The image features were associated with artificial intelligence techniques to determine an ANN configuration. This approach provides an innovative and low-cost analytical strategy with satisfactory performance for the prediction of phenolic content through TLC plate images.

Association between Cardiometabolic risk factor and responsiveness to vitamin D supplementation: a new approach using artificial neural network analysis

BackgroundThere are increasing data highlighting the effectiveness of vitamin D supplementation in the treatment of vitamin D deficiency. But individuals vary in their responsiveness to vitamin D supplementation. In this study, the association between several cardiometabolic risk factors and the magnitude of response to vitamin D supplementation (change in vitamin D level) was investigated using a novel artificial neural networks (ANNs) approach.MethodsSix hundred eight participants aged between 12 to 19 years old were recruited to this prospective interventional study. Nine vitamin D capsules containing 50,000 IU vitamin D/weekly were given to all participants over the 9 week period. The change in serum 25(OH) D level was calculated as the difference between post-supplementation and basal levels. Suitable ANNs model were selected between different algorithms in the hidden and output layers and different numbers of neurons in the hidden layer. The major determinants for predicting the response to vitamin D supplementation were identified.ResultsThe sigmoid in both the hidden and output layers with 4 hidden neurons had acceptable sensitivity, specificity and accuracy, assessed as the area under the ROC curve, was determined in our study. Baseline serum vitamin D (30.4%), waist to hip ratio (10.5%), BMI (10.5%), systolic blood pressure (8%), heart rate (6.4%), and waist circumference (6.1%) were the most important factors in predicting the response to serum vitamin D levels.ConclusionWe provide the first attempt to relate anthropometric specific recommendations to attain serum vitamin D targets. With the exception of cardiometabolic risk factors, the relative importance of other factors and the mechanisms by which these factors may affect the response requires further analysis in future studies (Trial registration: IRCT201509047117N7; 2015-11-25; Retrospectively registered).

A Conceptual Artificial Neural Network Model in Warehouse Receiving Management

The purpose of this research is to explore a suitable Artificial Neural Network (ANN) method applying to warehouse receiving management. A conceptual ANN model is proposed to perform identification and counting of components. The proposed model consists of a standard image library, an ANN system to present objects for identification from the real-time images and to count the number of objects in the image. The authors adopted four basic mechanical design shapes as the attributes of images for shape analysis and pre-defined features; the joint probability from Bayes theorem and image pixel values for object counting is applied in this research. Compared to other ANNs, the proposed conceptual model is straightforward to perform classification and counting. The model is tested by employing a mini image dataset which is industrial enterprise relevant. The initial result shows that the proposed model has achieved an accuracy rate of 80% in classification and a 97% accuracy rate in counting. The development of the model is associated with a few challenges, including exploring algorithms to enhance the accuracy rate for component identification and testing the model in a larger dataset.

A Deep Neural Network Model for Hybrid Spectrum Sensing in Cognitive Radio

Spectrum sensing (SS) is an essential task of the secondary user (SU) in a cognitive radio system. SS monitors the primary user (PU) activity in order to avoid any collision with SU, as the latter should be silent when PU is active on a given channel. Hybrid SS (HSS) is one of the powerful methods used to monitor PU activity. It consists of using different detectors together to make a final decision on the PU status. In this manuscript, artificial neural networks (ANN) are used to perform HSS. Since our data is composed from the test statistics (TSs) of several detectors, thus it can be modeled as tabular. Fully connected neural networks become the most suitable ANN model. We applied cutting-edge techniques in the field of deep learning in order to get the best possible accurate neural network model in our application. These techniques boil down to: embedding, regularization, batch normalization and smart learning rate selection. With the help TSs related to several detectors, ANN is trained to distinguish between two hypotheses, $$H_0$$ : PU is absent and $$H_1$$ : PU is active. Numerical results show the effectiveness of our proposed ANN-based HSS, as it outperforms the classical ANN-based energy detector and proves its capability to detect PU signal at very low SNR.

RUNOFF ESTIMATION IN URBAN CATCHMENT USING ARTIFICIAL NEURAL NETWORK MODELS

Many types of physical models have been developed for runoff estimation with successful results. However, accurate estimation of runoff remains a challenging problem owing to the lack of field data and the complexity of its hydrological process. In this paper, a machine learning method for runoff estimation is presented as an alternative approach to the physical model. Various types of input variables and artificial neural network (ANN) architectures were examined in this study. Results showed that a two-layer network with the tansig activation function and the Levenberg–Marquardt learning algorithm performed the best. For this architecture, the most effective input vector consists of a catchment perimeter, canal length, slope, runoff coefficient, and rainfall intensity. However, results of multivariate analysis of variance indicated the significant interaction effect of input data and the ANN architecture. Thus, to create a suitable ANN model for runoff estimation, a systematic determination of the input vector is necessary

Estimativa de escoamento em captação urbana utilizando modelos de rede neural artificial

Many types of physical models have been developed for runoff estimation with successful results. However, accurate runoff estimation remains a challenging problem owing to the lack of field data and the complexity of its hydrological process. In this paper, a machine learning method for runoff estimation is presented as an alternative approach to the physical model. Various types of input variables and Artificial Neural Network (ANN) architectures were examined in this study. Results showed that a two-layer network with the tansig activation function and the Levenberg–Marquardt learning algorithm had the best performance. For this architecture, the most effective input vector consists of a catchment perimeter, canal length, slope, runoff coefficient, and rainfall intensity. However, results of multivariate analysis of variance indicated the significant interaction effect of input data and the ANN architecture. Thus, to create a suitable ANN model for runoff estimation, a systematic determination of the input vector is necessary.