Artificial Neural Network Training Research Articles

Enhancing Microwave Photonic Interrogation Accuracy for Fiber-Optic Temperature Sensors via Artificial Neural Network Integration

In this paper, an application of an artificial neural network algorithm is proposed to enhance the accuracy of temperature measurement using a fiber-optic sensor based on a Fabry–Perot interferometer (FPI). It is assumed that the interrogation of the FPI is carried out using an optical comb generator realizing a microwave photonic approach. Firstly, modelling of the reflection spectrum of a Fabry–Perot interferometer is implemented. Secondly, probing of the obtained spectrum using a comb-generator model is performed. The resulting electrical signal of the photodetector is processed and is used to create a sample for artificial neural network training aimed at temperature detection. It is demonstrated that the artificial neural network implementation can predict temperature variations with an accuracy equal to 0.018 °C in the range from −10 to +10 °C and 0.147 in the range from −15 to +15 °C.

Physics-informed loss functions for vertical total electron content forecast

AbstractWe design physics-informed loss functions for training Artificial Neural Network (ANN) models to forecast the ionospheric vertical Total Electron Content (vTEC) from 1 to 24 hours in advance. The ANN models exploit our physics-informed loss functions, data provided by the Global Navigation Satellite Systems (GNSS) receiver installed at Tsukuba (36.06$$^{o}$$ o N, 140.05$$^{o}$$ o E), Japan, and external drivers (solar and geomagnetic indices). The time series used span from January 1, 2006, to December 31, 2018, i.e., a full solar cycle. A proper set of external drivers for the ANN models training are selected by ranking their importance in relation to the vTEC dynamics at different forecasting horizons. They result to be the 10.7 cm Solar Flux (F10.7), the magnitude of the Interplanetary Magnetic Field (IMF) $$\boldsymbol{B}_{\boldsymbol{T}}$$ B T , and the Auroral Electrojet (AE) index. Moreover, a second set of indices among those available has been considered as constraints in the design of the physics-informed loss functions. They are the Disturbance Storm Time (Dst) index, the solar wind speed $$\boldsymbol{v}, \boldsymbol{B}_{\boldsymbol{T}}$$ v , B T , and the By and Bz components of the interplanetary magnetic field. To assess the performance of the resulting ANN models, we use the statistical parameter coefficient of determination ($$\boldsymbol{R}^{{\textbf {2}}}$$ R 2 ), the standard deviation (SD), and the Wilcoxon non-parametric signed ranked test. We show that, in the testing period analyzed (from 2017-09-13, at 04:40:00, to 2018-12-31, at 23:55:00), one of our physics-informed loss functions provides a better performance of the ANN with regard to the standard loss function commonly adopted. In particular, when the new loss function is used in the ANN model, the average SD is minimized across all forecasting horizons in the training, validation and test datasets. SD is 0.2560 TECU, 0.3183 TECU and 0.4240 TECU for the training, validation and test dataset respectively, where 1 TECU = $${\textbf {10}}^{{\textbf {16}}}$$ 10 16 electrons/$$\boldsymbol{m}^{{\textbf {2}}}$$ m 2 . The ANN model, incorporating the new loss function and applied to the test dataset, shows a significant improvement according to the Wilcoxon signed ranked test. In fact by selecting a significance level $$\boldsymbol{\alpha }~ \mathbf {= 0.05}$$ α = 0.05 , the probability to obtain results by chance with the new loss function as compared to the standard loss function is 0.01504 (i.e., $$<\alpha $$ < α ), which implies that the new loss function gives a statistical improvement to the forecasting capability of the ANN model. To the best of our knowledge, this is the first time a physics-informed loss function has been designed for the task of forecasting the ionospheric vTEC.

Computer-assisted design of CGC catalysts for ethylene/1-octene copolymerization: A combined DFT and artificial neural network approach

High temperature solution polymerization of ethylene/1-octene is the main means to produce polyolefin elastomer (POE), one of the most important polyolefin materials, and the metallocene catalysts with high-efficiency are the key to achieve POE production. However, there remains a deficiency in a universal approach for creating catalysts for ethylene/1-octene copolymerization. In this document, we outlined a method for designing catalysts for ethylene/1-octene copolymerization by combining density functional theory (DFT) with an artificial neural network (ANN) was proposed. By optimizing the structure and computing electronically a sequence of constrained geometry catalysts (CGC) by DFT, descriptors related to steric and electronic descriptors were gathered to create the dataset. After ANN training, an ANN-based model using for predicting catalyst activity with high R2 value was optimized, and a class of novel highly efficient CGC have been successfully designed. Furthermore, an exploration was conducted into how the structure of catalysts correlates with their performance. The electronic properties of the catalysts were found to have more significantly effect on the catalytic performance than the steric properties. In addition, the effect of the β-site substitution of the cyclopentadienyl group on the catalytic activity of predicted CGC was studied and discussed. The groups with ring-structure, especially with conjugate system, were proved to possess a good promoting effect on the catalytic activity.

An indirect training approach for implicit constitutive modelling using recurrent neural networks and the virtual fields method

Accurate material description is crucial to achieve high-quality results in computational analysis software. Phenomenological constitutive laws generalize the material behaviour observed in simple mechanical tests. The resulting empirical expressions contain parameters that need to be calibrated through an inverse optimization process. Advancements in Digital Image Correlation (DIC) techniques have enabled the extraction of non-uniform multi-axial displacement fields, facilitating the development of heterogeneous mechanical specimens and unlocking access to richer material behaviour data. Despite these advancements, constitutive models are still limited by mathematical formulations and biases due to simplifying assumptions. Artificial Neural Networks (ANNs), as universal function approximators, could potentially drive a paradigm shift in the field. ANNs do not require explicit pre-formulations, hence avoiding the need for identifying unknown parameters. Moreover, ANNs are able to implicitly learn patterns purely from data, allowing to circumvent the biases induced by the simplifying assumptions of first-order principles. However, the training of ANNs for implicit constitutive modelling is not straightforward, given the impossibility of obtaining direct measurements of stress. This paper explores the integration of Recurrent Neural Networks (RNNs) with the Virtual Fields Method (VFM) for material modelling. This approach uses global force and displacement data to indirectly train the neural network, with the equilibrium being evaluated globally through the VFM loss function. The proposed method is (i) computationally efficient by not requiring Finite Element Analysis (FEA), (ii) compatible with both full-field measurements and numerically generated data, and (iii) able to handle experimental boundary conditions.

Estimating droplet size distribution of emulsions using turbidity measurements: A soft sensor based on artificial neural network

Ability to characterise droplet size distribution (DSD) of emulsions in real-time is essential for on-demand production of customised emulsions. In this work, for the first time, we demonstrate a possibility of estimating full DSD of oil in water emulsions from turbidity measurements using a single wavelength light source. We used recently published data of DSD and turbidity measurements of rapeseed oil in water emulsions (oil volume fractions of 0.15, 0.3 and 0.45) produced with vortex based hydrodynamic cavitation device. Measured DSDs are represented by weighted sum of three log-normal distributions. We developed an approach and a methodology based on artificial neural network (ANN) to estimate DSD from a single measurement of turbidity. A mathematical model is developed to simulate measured turbidity profiles using the known DSDs. The validated model was then used to generate simulated data of turbidity and oil volume fraction pairs (105 pairs). This synthetic data was used to train ANN which used turbidity and volume fraction as input and eight parameters of DSD as output. The developed ANN was able to capture the experimentally measured characteristic diameters and DSDs very well for three oil volume fractions and four different number of passes. The presented methodology and results will be useful for developing an in-line soft sensor based on turbidity measurements for real time estimation of full DSDs of emulsions.

Integrated AI-driven optimization of Fenton process for the treatment of antibiotic sulfamethoxazole: Insights into mechanistic approach

Antibiotics, as a class of environmental pollutants, pose a significant challenge due to their persistent nature and resistance to easy degradation. This study delves into modeling and optimizing conventional Fenton degradation of antibiotic sulfamethoxazole (SMX) and total organic carbon (TOC) under varying levels of H2O2, Fe2+ concentration, pH, and temperature removal using both statistical and artificial intelligence techniques including Multiple Regression Analysis (MRA), Support Vector Regression (SVR) and Artificial Neural Network (ANN). In statistical analysis, ANN model showcased an elevated predictive accuracy compared to its counterparts due to having lowest RMSE values of 0.986 and 1.173 for both SMX and TOC, respectively. Sensitivity showcased H2O2/Fe2+ ratio, time and pH as pivotal for SMX degradation, while in simultaneous SMX and TOC reduction, fine tuning the time, pH, and temperature was essential. Leveraging a Hybrid Genetic Algorithm-Desirability Optimization approach, the trained ANN model revealed an optimal desirability of 0.941 out of 1000 solutions which yielded a 91.18% SMX degradation and 87.90% TOC removal under following specific conditions: treatment time of 48.5 min, Fe2+: 7.05 mg L−1, H2O2: 128.82 mg L−1, pH: 5.1, initial SMX: 97.6 mg L−1, and a temperature: 29.8 °C. LC/MS analysis reveals multiple intermediates with higher m/z (242, 270 and 288) and lower m/z (98, 108, 156 and 173) values identified, however no aliphatic hydrocarbon was isolated, because of the low mineralization performance of Fenton process. Furthermore, some inorganic fragments like NH4+ and NO3− were also determined in solution. This comprehensive research enriches AI modeling for intricate Fenton-based contaminant degradation, advancing sustainable antibiotic removal strategies.

Prediction of Hydrodynamic Parameters of the State of the Bottomhole Zone of Wells Using Machine Learning Methods

The relevance of the development of a methodology for the operational assessment of the bottom-hole formation zone (the permeability of the bottom-hole formation zone and the skin factor) is primarily due to economic considerations, since existing approaches to its definition based on hydrodynamic studies lead to shortages and increased risks of failure to ensure the output of the well. In this regard, the use of modern methods of working with big data, such as deep learning of artificial neural networks, will ensure monitoring of the condition of the bottom-hole zone of the well formation without stopping them for hydrodynamic tests, which will reduce losses for oil production enterprises. It will allow for operational analysis for effective and timely application of intensification technologies, enhanced oil recovery. The authors analyzed the existing methods for determining the bottom-hole characteristics of the formation and machine learning approaches in the direction of solving this problem. The article presents a methodology for the operational assessment of the state of the bottom-hole formation zone: the permeability of the near bottomhole zone (NBHZ) and the skin factor using artificial neural network training approaches based on geological, operational data and the results of interpretation of hydrodynamic studies on the example of sandstones of oil fields in the Perm Region. A fully connected neural network was used to predict the NBHZ permeability. The article presents the results of testing various neural network architectures: the number of layers and neurons in layers with the choice of the best one. Some techniques were used to prevent over-training of models. The author’s methodology for assessing the skin factor of wells is proposed using a comprehensive analysis of the constructed statistical models and training models of artificial neural networks to solve the regression problem. In future studies, it is planned to use recurrent and convolutional neural networks to study the dynamic components of the formation of the bottom-hole formation zone and create an integrated approach to solve the problem.

Deep learning with multilayer perceptron for optimizing the heat transfer of mixed convection equipped with MWCNT-water nanofluid

In the modern era, Artificial Intelligence (AI) has emerged as a powerful tool that can rapidly generate highly accurate data, offering tremendous potential for optimizing system performance. This study focuses on harnessing the capabilities of an artificial neural network (ANN) to determine optimal parameter values for heat transfer in a mixed convection mechanism. The first step of this research involved conducting CFD simulations to investigate the impact of varying Grashof numbers (102,103,and104), Reynolds numbers (1,10,and100), and MWCNT volume fractions (0, 0.01, 0.02, and 0.03) on the Nusselt number within an elliptical enclosure containing a centrally located rotational cylinder. A total of 48 simulations were performed, generating a comprehensive dataset for training the ANN-Multilayer Perceptron (MLP). In the second step, the trained ANN was utilized to generate an additional 700 data points with remarkable accuracy. This enabled efficient exploration of the parameter space, providing valuable insights into the system behavior and facilitating optimization efforts. The findings of this study revealed a 0.03 vol fraction of MWCNT into the water in Grashof numbers of 102,103,and104, the average Nusselt number increased by approximately 46%, 31%, and 12%, respectively. The ANN-based approach successfully identified optimal values for the variables that maximize the Nusselt number.

A New Empirical Correlation for Pore Pressure Prediction Based on Artificial Neural Networks Applied to a Real Case Study

Pore pressure prediction is a critical parameter in petroleum engineering and is essential for safe drilling operations and wellbore stability. However, traditional methods for pore pressure prediction, such as empirical correlations, require selecting appropriate input parameters and may not capture the complex relationships between these parameters and the pore pressure. In contrast, artificial neural networks (ANNs) can learn complex relationships between inputs and outputs from data. This paper presents a new empirical correlation for predicting pore pressure using ANNs. The proposed method uses 42 datasets of well log data, including temperature, porosity, and water saturation, to train ANNs for pore pressure prediction. The trained model, with the Bayesian regularization backpropagation function, predicts the pore pressure with an average absolute percentage error (AAPE) and correlation coefficient (R) of 4.22% and 0.875, respectively. The trained ANN is then used to develop a new empirical correlation that relates pore pressure to the input parameters considering the weights and biases of the optimized ANN model. To validate the proposed correlation, it is applied to a blind dataset, where the model successfully predicts the pore pressure with an AAPE of 5.44% and R of 0.957. The results show that the proposed correlation provides accurate and reliable predictions of pore pressure. The proposed method provides a robust and accurate approach for predicting pore pressure in petroleum engineering applications, which can be used to improve drilling safety and wellbore stability.

Artificial neural network (ANN) analysis on thermophysical properties of magnetohydrodynamics flow with radiation in an arc-shaped enclosure with a rotating cylinder

The objective of this research is to examine the thermophysical features of magnetic parameter (Ha) and time step (τ) in a lid-driven cavity using a water-based Al2O3 nanofluid and the efficacy of ANN models in accurately predicting the average heat transfer rate. The Galerkin weighted residual approach is used to solve a set of dimensionless nonlinear governing equations. The Levenberg-Marquardt back propagation technique is used for training ANN using sparse simulated data. The findings of the investigation about the flow and thermal fields are shown. Furthermore, a comparative study and prediction have been conducted on the impact of manipulating factors on the average Nusselt number derived from the numerical heat transfer analysis. The findings of the research indicate that, in the absence of magnetohydrodynamics, a rise in the Hartmann number resulted in a drop in both the fluid velocity profile and magnitude. Conversely, it was observed that the temperature and Nusselt number exhibited an increase under these conditions. The mean temperature of the fluid rises as the Hartmann number drops, reaching a peak value of 0.114 when Ha = 0. The scenario where Ha = 0, representing the lack of magnetohydrodynamics, shows the highest average Nusselt number, whereas the instance with Ha = 45 presents the lowest Nusselt number. The ANN model has a high level of accuracy, as seen by an MSE value of 0.00069 and a MAE value of 0.0175, resulting in a 99% accuracy rate.

A memetic dynamic coral reef optimisation algorithm for simultaneous training, design, and optimisation of artificial neural networks

Artificial Neural Networks (ANNs) have been used in a multitude of real-world applications given their predictive capabilities, and algorithms based on gradient descent, such as Backpropagation (BP) and variants, are usually considered for their optimisation. However, these algorithms have been shown to get stuck at local optima, and they require a cautious design of the architecture of the model. This paper proposes a novel memetic training method for simultaneously learning the ANNs structure and weights based on the Coral Reef Optimisation algorithms (CROs), a global-search metaheuristic based on corals’ biology and coral reef formation. Three versions based on the original CRO combined with a Local Search procedure are developed: (1) the basic one, called Memetic CRO; (2) a statistically guided version called Memetic SCRO (M-SCRO) that adjusts the algorithm parameters based on the population fitness; (3) and, finally, an improved Dynamic Statistically-driven version called Memetic Dynamic SCRO (M-DSCRO). M-DSCRO is designed with the idea of improving the M-SCRO version in the evolutionary process, evaluating whether the fitness distribution of the population of ANNs is normal to automatically decide the statistic to be used for assigning the algorithm parameters. Furthermore, all algorithms are adapted to the design of ANNs by means of the most suitable operators. The performance of the different algorithms is evaluated with 40 classification datasets, showing that the proposed M-DSCRO algorithm outperforms the other two versions on most of the datasets. In the final analysis, M-DSCRO is compared against four state-of-the-art methods, demonstrating its superior efficacy in terms of overall accuracy and minority class performance.

Utilization of Synthetic Near-Infrared Spectra via Generative Adversarial Network to Improve Wood Stiffness Prediction.

Near-infrared (NIR) spectroscopy is widely used as a nondestructive evaluation (NDE) tool for predicting wood properties. When deploying NIR models, one faces challenges in ensuring representative training data, which large datasets can mitigate but often at a significant cost. Machine learning and deep learning NIR models are at an even greater disadvantage because they typically require higher sample sizes for training. In this study, NIR spectra were collected to predict the modulus of elasticity (MOE) of southern pine lumber (training set = 573 samples, testing set = 145 samples). To account for the limited size of the training data, this study employed a generative adversarial network (GAN) to generate synthetic NIR spectra. The training dataset was fed into a GAN to generate 313, 573, and 1000 synthetic spectra. The original and enhanced datasets were used to train artificial neural networks (ANNs), convolutional neural networks (CNNs), and light gradient boosting machines (LGBMs) for MOE prediction. Overall, results showed that data augmentation using GAN improved the coefficient of determination (R2) by up to 7.02% and reduced the error of predictions by up to 4.29%. ANNs and CNNs benefited more from synthetic spectra than LGBMs, which only yielded slight improvement. All models showed optimal performance when 313 synthetic spectra were added to the original training data; further additions did not improve model performance because the quality of the datapoints generated by GAN beyond a certain threshold is poor, and one of the main reasons for this can be the size of the initial training data fed into the GAN. LGBMs showed superior performances than ANNs and CNNs on both the original and enhanced training datasets, which highlights the significance of selecting an appropriate machine learning or deep learning model for NIR spectral-data analysis. The results highlighted the positive impact of GAN on the predictive performance of models utilizing NIR spectroscopy as an NDE technique and monitoring tool for wood mechanical-property evaluation. Further studies should investigate the impact of the initial size of training data, the optimal number of generated synthetic spectra, and machine learning or deep learning models that could benefit more from data augmentation using GANs.

Machine learning based prediction of mechanical properties of WC-Co cemented carbides from magnetic data only

Based on experimental data and extensive experience, magnetic coercivity and saturation moment are traditionally used to estimate the microstructure and quality of cemented carbides, especially in the manufacturing industry. This work demonstrates that predictions of the structural and mechanical properties of manufactured WC-Co elements can be derived in principle from magnetic data alone using an artificial neural network (ANN). A collection of WC-Co pellet samples with a wide variety of powder compositions and processing parameters was produced to cover a wide range of characteristic features for ANN training. The total field distribution, extracted from first-order-reversal-curves, serves as input data for the ANN. Microstructural parameters such as mean grain size and mechanical properties such as hardness and fracture toughness can be derived from the purely magnetic measurements with high accuracy, while the transverse rupture strength shows large errors and cannot be predicted.

Multiobjective Neuro-Fuzzy Controller Design and Selection of Filter Parameters of UPQC Using Predator Prey Firefly and Enhanced Harmony Search Optimization

This research introduces a unified power quality conditioner (UPQC) that integrates solar photovoltaic (PV) system and battery energy systems (SBES) to address power quality (PQ) issues. The reference signals for voltage source converters of UPQC are produced by the Levenberg–Marquardt back propagation (LMBP) trained artificial neural network control (ANNC). This method removes the necessity for conventional dq0, abc complex shifting. Moreover, the optimal choice of parameters for the adaptive neuro-fuzzy inference system (ANFIS) was achieved through the integration of the enhanced harmony search algorithm (EHSA) and the predator-prey-based firefly algorithm (PPFA) in the form of the hybrid metaheuristic algorithm (PPF-EHSA). In addition, the algorithm is employed to optimize the selection of resistance and inductance values for the filters in UPQC. The primary objective of the ANNC with predator-prey-based firefly algorithm and enhanced harmony search algorithm (PPF-EHSA) is to enhance the stability of the DC-link capacitor voltage (DLCV) with reduced settling time amid changes in load, solar irradiation (G), and temperature (T). Moreover, the algorithm seeks to achieve a reduction in total harmonic distortion (THD) and enhance power factor (PF). The method also focuses on mitigating fluctuations such as swell, harmonics, and sag and also unbalances at the grid voltage. The proposed approach is examined through four distinct cases involving various permutations of loads and sun irradiation (G). However, in order to demonstrate the performance of the suggested approach, a comparison is conducted with the ant colony and genetic algorithms, i.e., (ACA) (GA), as well as the standard methods of synchronous reference frame (SRF) and instantaneous active and reactive power theory (p-q). The results clearly demonstrate that the proposed method exhibits a reduced mean square error (MSE) of 0.02107 and a lower total harmonic distortion (THD) of 2.06% compared to alternative methods.

Thermo-environmental multi- investigation and ANN-based optimization of a novel heat integration criteria system integrated with a marine engine generating liquefied hydrogen

Marine transportation is a significant contributor to overall energy consumption among transportation sectors and is responsible for producing a considerable amount of greenhouse gases. A potential solution for alternative applications involves implementing combined heat integration, thereby enabling the production of additional utilities and mitigating emissions. The current work introduces an innovative heat integration process for a marine engine, focusing on implementing an optimal thermal matching technique to minimize overall irreversibility in producing liquefied hydrogen and coolant. The process incorporates an organic flash-bi-evaporator cooling cycle, a humidification dehumidification desalination, a polymer electrolyte membrane water electrolysis process, and a Claude cycle. The generated freshwater is delivered to the electrolyzer to produce gaseous hydrogen. This product and the cooling for freezing are utilized in the Claude cycle for hydrogen liquefaction. The study utilizes an advanced thermo-environmental multi-criteria investigation and optimization, considering sensitivity analysis and optimization based on artificial intelligence method. The optimization process incorporates the training and testing of artificial neural networks, NSGA-II method, and TOPSIS decision-making. The primary objective functions include exergetic efficiency and carbon dioxide emission. The findings demonstrate that the specified objectives are computed to be 0.121 and 2.67 kg/MWh, correspondingly. Besides, this condition exhibits a liquefied hydrogen flow rate of 6.44 L/h and a cooling output of 43.61 kW, showing an energy efficiency of 0.1145. Also, the total exergy destruction rate associated with the arranged structure is 124.5 kW. Furthermore, the optimum state reveals an exergoenvironmental index of 0.840 and an exergetic stability factor of 0.869.

Artificial neural network prediction of particle size of agglomerated polybutadiene latex

AbstractPolymer agglomeration is a useful way to enlarge the particle size of polybutadiene latex (PBL). However, the agglomeration process is very complex and is affected by many factors. Therefore, it is difficult to predict the mean particle size and distribution of agglomerated PBL. This study proposed a novel simulation method based on the artificial neural network (ANN) model. Firstly, the influence of agglomeration conditions on the particle size of agglomerated PBL were investigated, including the particle size of pre‐agglomerate PBL, the amount, the α‐methacrylic acid (α‐MAA) content, the solid content and the particle size of agglomerating agent. It is found that these factors have a close nonlinear relationship with the mean particle size and polydispersion index (PDI) of agglomerated PBL. Then, a model containing 236 data points was built using ANN including these properties as input with mean particle size and PDI of agglomerated PBL as output. The coefficient of determination (R2) values predicted by ANN for training, validation, test and total data are 0.9879, 0.9964, 0.9930, and 0.9902, respectively. It indicates that there is a good agreement between experimental data and predicted results by ANN. This study demonstrates that a simple machine learning method has good performance in predicting the particle size of agglomerated PBL, which further improves the development efficiency of PBL polymer agglomeration.

Strength Optimization of Nanocomposite Cementitious Materials Using Nanoscale Modifications

Represent Volume Element (RVE) is broadly used by investigators to control the properties of nano-cementitious materials. This paper focuses on analyzing a set of RVE data and proposes parametric equations for determining the compressive and flexural strength characteristics (σc and σf). Primarily, a parametric study is performed with RVE analysis. The essential design parameters are used to fit rational equations. The RVE data are also applied to train artificial neural networks of σc and σf. RVE, neural networks, and rational equations are correlated. Regression equations are validated with the experimental study. SEM, TEM, XRD, and FTIR results are carried out in the microscopic and mechanical analysis for carbon nanofiber cement composites, which can lift their strength, constancy, integrity, and density and reinforce the composite microstructure. Lastly, the Pareto-optimal design results are presented with a multi-objective optimization problem.

A novel approach to accelerate training in artificial neural network to detect three-phase flows using a gamma source and a detector

Measuring the percentage of mass fraction of multiphase flow in pipelines is one of the most significant issues in the petrochemical oil industry. In this research, The mass fraction in three-phase flow has been determined with artificial neural network. There are various ways to accelerate the neural network that in this study, changing the activation function and normalizing the data has been considered. There are 5 activation functions and 5 normalization methods for the input data which combine to create 25 states. Then, by using the normalizing method for the input data to the network to (12tanh(1100Xi−avg(Xi)std(Xi))), while changing the hidden network activator function to (f(X)=tanh(X)), the average number of repetitions for network training reached 171 repetitions and the average number of errors was 0.002. While without applying the mentioned cases, i.e., without normalizing the data and using the activation function (f(X)=X), the average value of errors is 5.63 and the average value of repetition is 417.

Relative vectoring using dual object detection for autonomous aerial refueling

AbstractOnce realized, autonomous aerial refueling will revolutionize unmanned aviation by removing current range and endurance limitations. Previous attempts at establishing vision-based solutions have come close but rely heavily on near perfect extrinsic camera calibrations that often change midflight. In this paper, we propose dual object detection, a technique that overcomes such requirement by transforming aerial refueling imagery directly into receiver aircraft reference frame probe-to-drogue vectors regardless of camera position and orientation. These vectors are precisely what autonomous agents need to successfully maneuver the tanker and receiver aircraft in synchronous flight during refueling operations. Our method follows a common 4-stage process of capturing an image, finding 2D points in the image, matching those points to 3D object features, and analytically solving for the object pose. However, we extend this pipeline by simultaneously performing these operations across two objects instead of one using machine learning and add a fifth stage that transforms the two pose estimates into a relative vector. Furthermore, we propose a novel supervised learning method using bounding box corrections such that our trained artificial neural networks can accurately predict 2D image points corresponding to known 3D object points. Simulation results show that this method is reliable, accurate (within 3 cm at contact), and fast (45.5 fps).

Application of Weighted Combinations of Activation Functions to Defect Prediction in Software Development

Software defect prediction (SDP) method aims to identify potential bugs in programs or defective modules in software projects. SDP method can greatly help developers allocate needed testing- and debugging-efforts. Presently, SDP is typically divided into two procedures: extracting features from source code and building a classification model using machine learning methods, such as support vector machine, decision tree, and neural networks, to build a classifier for defect prediction. However, there are still some limitations for SDP. For example, machine learning models require a fixed input size, but the size of each program is mostly inconsistent. Activation functions play an essential role in the training of artificial neural networks, but every kind of activation function has its own particular strengths and inherent constraints. The main purpose of this article is to propose a general framework of combining different activation functions with given weights to improve the effectiveness of SDP. We construct 41 kinds of defect prediction models by deep belief networks (DBNs) built with double-weighted or triple-weighted combination of six most commonly used activation functions to improve the predictions. In the experiment of this study, some real data from open-source projects are selected to evaluate the performance of our proposed weighted combination methods. It is found experimentally that the weighted combinations methods can enhance the accuracy of SDP. It is also noticed that our proposed weighted combination scheme is not restricted to the DBN or a particular kind of activation functions.