Artificial Neural Network Training Research Articles

Optimizing cotton‐picking robotic manipulator and inverse kinematics modeling using evolutionary algorithm‐assisted artificial neural network

AbstractThis study presents a particle swarm optimization (PSO) algorithm‐assisted neural‐network‐based inverse kinematics solution for a 4‐DoF (degree‐of‐freedom) cotton harvesting robot. A novel setup was developed to measure the three‐dimensional locations of in‐field cotton bolls. Dimensional optimization of the manipulator was conducted using the PSO algorithm to minimize torque requirements at joints. With the optimized links’ lengths, the targeted end‐effector positions were achieved effectively (coefficient of determination (R2) > 99.88). The genetic algorithm optimized the neural network architecture to include three hidden layers with [64 64 32] neurons, identifying the Tanh activation function as the optimal configuration. A custom loss function was used during the training of artificial neural network (ANN). Using angles predicted by the trained ANN, the end‐effector reached targeted positions with positioning errors below 13.0 mm. A hybrid model consisting of an ANN and PSO algorithm was developed to further reduce the error. This trained hybrid model resulted in a positioning error below 1.0 mm with inference time of 6.07 s during simulation phase. As compared to the ANN and PSO algorithm, hybrid model reduced the positioning error and inference time (>40.0%), respectively. For hybrid model, the mean percentage errors of 0.25%, 0.39%, and 0.84% were observed along the x‐, y‐, and z‐axis. A positioning error below 9.0 mm occurred during evaluation of the hybrid model with the fabricated manipulator. Hence, the developed hybrid model precisely determines the joint angles, allowing the end‐effector of the cotton harvesting robot to reach at targeted pose with minimum error.

A Statistical Shape Model Pipeline to Enable the Creation of Synthetic 3D Liver Data

Abstract The application of machine learning approaches in medical technology is gaining more and more attention. Due to the high restrictions for collecting intraoperative patient data, synthetic data is increasingly used to support the training of artificial neural networks. We present a pipeline to create a statistical shape model (SSM) using 28 segmented clinical liver CT scans. Our pipeline consists of four steps: data preprocessing, rigid alignment, template morphing, and statistical modeling. We compared two different template morphing approaches: Laplace-Beltrami-regularized projection (LBRP) and nonrigid iterative closest points translational (N-ICP-T) and evaluated both morphing approaches and their corresponding shape model performance using six metrics. LBRP achieved a smaller mean vertex-to-nearest-neighbor distances (2.486±0.897 mm) than N-ICP-T (5.559±2.413 mm). Generalization and specificity errors for LBRP were consistently lower than those of N-ICP-T. The first principal components of the SSM showed realistic anatomical variations. The performance of the SSM was comparable to a state-of-the-art model.

Artificial neural networks based on segmented model for adaptability and stability evaluation of soybean genotypes

Unlike models based on simple linear regressions, segmented models can better assess the adaptability and stability of genotypes, demonstrating a nonlinear response pattern over environmental variation. However, these methods can be under statistical limitations, such as the Type Error II increase and biased estimates. Therefore, this work aimed to transpose the concepts of adaptability and stability from the statistical analysis of a segmented model to the discriminatory potential of an artificial neural network (ANN) and use it to classify soybean (Glycine max (L.) Merr.) genotypes. An ANN training was carried out with the grain yield of 7,200 soybean genotypes simulated in 15 different environments. The ANN topology chosen was the one that had less than 1% of error in the testing phase with 1,800 simulated genotypes. A total of 9,000 simulated soybean genotypes were previously arranged in 18 different classes, which represented the combination of nine classes of adaptability by the Verma and collaborators (VCM) method and two classes of stability (invariability concept) by the Finlay & Wilkinson (FW) method. Finally, the grain production of ten real soybean genotypes was inputted into the ANN-trained model, and the classification regarding adaptability and stability was obtained. There was 90% agreement between the ANN and VCM analyses regarding the adaptability classification and 20% regarding stability. With the methods presented in this work, it was demonstrated that the potential of using ANNs to assess the adaptability of genotypes is strong. In addition, since stability was introduced in the ANN as a different concept from that used to classify the genotypes by the statistical method, such classification needs to be reviewed and further improved

Artificial neural network model of non-Darcy MHD Sutterby hybrid nanofluid flow over a curved permeable surface: Solar energy applications

The conversion of solar radiation to thermal energy has recently much interest as the requirement for renewable heat and power grows. Due to their enhanced ability to promote heat transmission, nanofluids can significantly improve solar-thermal systems' efficiency. This section aims to study the heat transfer behavior of the Sutterby hybrid nanofluid flow of magnetohydrodynamics in the presence of a non-uniform heat source/sink and linear thermal radiation over a non-Darcy curved permeable surface. A novel implementation of an intelligent numerical computing solver based on multi-layer perceptron (MLP) feed-forward back-propagation artificial neural network (ANN) with the Levenberg-Marquard algorithm is provided in the current study. Data were gathered for the ANN model's testing, certification, and training. Established mathematical equations are nonlinear, which are resolved for velocity, the temperature in addition to the skin friction coefficient, and the rate of heat transfer by using the bvp4c with MATLAB solver. The ANN model selects data, constructs and trains a network, then evaluates its efficacy via mean square error. Graphs illustrate the impact of a wide range of physical factors on variables, including pressure, velocity, and temperature. In the entire study, the thermal energy improved by the SiO2 (silicon dioxide) - Au (gold) hybrid nanofluid than the SiO2-TiO2 (titanium dioxide) hybrid nanofluid. The higher internal heat generation/absorption parameter values increase the temperature.

Modern Artificial Intelligence-An Overview

Modern Artificial Intelligence (AI) is a rapidly evolving field that encompasses a range of techniques and approaches, including machine learning, deep learning, natural language processing, computer vision, robotics, and more. The development of AI technologies has enabled unprecedented levels of accuracy in tasks such as image and speech recognition, natural language understanding, and game playing. This has been made possible by the rise of deep learning, which involves training artificial neural networks on vast amounts of data to recognize patterns and make predictions with high accuracy. Other recent advances in modern AI include the development of generative models and reinforcement learning. Despite the significant progress made in modern AI, there are still many challenges that need to be addressed, including issues related to data privacy, fairness, and bias, and the need for more explainable AI systems that can provide clear and transparent reasoning for their decisions. This study provides an overview of modern AI and its applications, as well as the challenges and opportunities that lie ahead in this rapidly evolving field.

An Intelligent Method for Fault Location Estimation in HVDC Cable Systems Connected to Offshore Wind Farms

Large and remote offshore wind farms (OWFs) usually use voltage source converter (VSC) systems to transmit electrical power to the main network. Submarine high-voltage direct current (HVDC) cables are commonly used as transmission links. As they are liable to insulation breakdown, fault location in the HVDC cables is a major issue in these systems. Exact fault location can significantly reduce the high cost of submarine HVDC cable repair in multi-terminal networks. In this paper, a novel method is presented to find the exact location of the DC faults. The fault location is calculated using extraction of new features from voltage signals of cables’ sheaths and a trained artificial neural network (ANN). The results obtained from a simulation of a three-terminal HVDC system in power systems computer-aided design (PSCAD) environment show that the maximum percentage error of the proposed method is less than 1%.

A Machine-Learning-Based Approach for Railway Track Monitoring Using Acceleration Measured on an In-Service Train.

In this paper, a novel railway track monitoring approach is proposed that employs acceleration responses measured on an in-service train to detect the loss of stiffness in the track sub-layers. An Artificial Neural Network (ANN) algorithm is developed that works with the energies of the train acceleration responses. A numerical model of a half-car train coupled with a track profile is employed to simulate the train vertical acceleration. The energy of acceleration signals measured from 100 traversing trains is used to train the ANN for healthy track conditions. The energy is calculated every 15 m along the track, each of which is called a slice. In the monitoring phase, the trained ANN is used to predict the energies of a set of train crossings. The predicted energies are compared with the simulated ones and represented as the prediction error. The damage is modeled by reducing the soil stiffness at the sub-ballast layer that represents hanging sleepers. A damage indicator (DI) based on the prediction error is proposed to visualize the differences in the predicted energies for different damage cases. In addition, a sensitivity analysis is performed where the impact of signal noise, slice sizes, and the presence of multiple damaged locations on the performance of the DI is assessed.

Improved artificial neural networks based whale optimization algorithm

Owing to the increasing interest in artificial neural networks (ANNs) across various fields of study, many studies have focused on enhancing their performance through the utilisation of different learning algorithms. This study examines the use of the Whale Optimization Algorithm (WOA) as a training algorithm to improve the classification accuracy of ANNs. To achieve a high level of classification accuracy with ANN models, it is imperative to ensure that the model is appropriately designed in terms of the employed structure, training algorithm and activation function. In this work, WOA was adopted to train ANN models using 10 well-known datasets sourced from the UCI machine learning repository. The classification accuracy of a WOA-trained ANN was compared with that of a backpropagation-trained ANN, and the results showed that the WOA-trained ANN exhibited superior performance.

Effectiveness of MATLAB and Neural Networks for Solving Nonlinear Equations by Repetitive Methods

Finding solutions to nonlinear equations is not only a matter for mathematicians but is essential in many branches such as physics, statistics, and others. However, some of the nonlinear equations in numerical analysis require a lot of complex calculations to achieve convergence. This leads to many arithmetic errors and is consumed a great effort to solve them. Hence, researchers in numerical analysis use computer programs to find approximate solutions. This study used Matlab and Artificial Neural Networks and applied two different numerical analysis methods. The results from training artificial neural networks by utilizing the Backpropagation algorithm and MATLAB have been compared. The importance of this study lies in shedding light on the capabilities of Matlab and its strength in the field of methods for solving mathematical series, and helps students in mathematics in solving complex equations faster and more accurately, also studying the utilization of Artificial Neural Network algorithms in solving these methods, and clarifying the difference between them and programming Ordinary Matlab and comparing them with ordinary mathematical methods. The findings revealed that Traditional methods need more effort. MATLAB helps. On the other hand, solving numerical analysis problems is easier, faster, more accurate, and more effective. Furthermore, in the case of the Matlab application, the Newton method gave faster and less in the number of steps. Additionally, in training, the neural network based on the Newton method gave results faster depending on the Bisection method.

Damage analysis of Nomex honeycomb sandwich structures using image processing and artificial intelligence approaches

AbstractIn this study, a method was developed to detect impact damage and determine the damage propagation in increasing and repetitive impacts. The effect of nanographene additives on the propagation of delamination damage in the top face layer of a sandwich structure was investigated using the developed method. In addition, the effects of nanographene on visible (VID) or barely visible damage (BVID) formation were investigated, and the stiffness of the sandwich structures was evaluated. An ultrasonic C‐scan nondestructive testing was used to visualize the damage, and the numerical values of the damaged areas of the sandwich structures at different energy levels were determined using an image processing method. The obtained values were used in artificial neural network (ANN) training to estimate the impact energy at which the structural integrity of the sandwich structures was disrupted. It was observed that the nanographene additive was effective in maintaining rigidity of the sandwich structures and therefore, slowed the delamination damage progression in increasing and repeated impacts. The damage areas of the neat sandwich structures (NSS) and graphene nanoplatelets doped sandwich structures (GSS) structures at the first impact 5j energy level was 1004 and 811 mm2, respectively. At the first impact, 20% less damage occurred in the GNP‐doped sandwich structures. The increasing and repetitive last impact energies at the end of the neural network training were 5j + 35j in NSS and 5j + 50j in GSS. The method allows for a detailed examination of the damage progression of sandwich structures under increasing and repeated impact loads.Highlights Low‐velocity impact tests of Nomex honeycomb sandwich structures were performed. Damage formation and propagation in sandwich structures were investigated. The non‐destructive test C‐scan method was used to determine BVID. The damaged areas were determined using an image processing method. The damage areas were estimated using artificial neural networks.

Error correction method for heat flux and a new algorithm employed in inverting wall thermal resistance using an artificial neural network: Based on IN-SITU heat flux measurements

The heat-flux meter is affected by various uncertain factors in practical applications, which brings some difficulties to scholars in postprocessing data. Therefore, a modified method for calculating the heat flux of the building surface is proposed. It can make the simulated heat flux value consistent with the experimental heat flux value, thus verifying the accuracy of the model with the heat flux value. On this basis, through artificial neural network (ANN) training, the heat conduction flux inside the wall is predicted by using the measured heat flux and temperature of inner surfaces. The results show that the root mean square errors obtained by the heat flux error correction method are all less than 1.12 W·m−2. The ANN fitting coefficients are also greater than 0.9. The error correction method proposed in this work comprehensively considers the influence of solar radiation and the total heat transfer coefficient and is also applicable to passive buildings and buildings with PCM in the envelope. The well-performing network built by ANN can predict the thermal conductivity heat flux inside the wall more accurately, providing a new method to measure or check whether the thermal resistance of the wall is qualified in the field.

Analysis of road accidents due to obstruction in sight distance caused by foggy weather using Geographic Information System (GIS) and Artificial Neural Network (ANN) at Jumja along Thimphu-Phuentsholing Highway (Asian Highway 48)

Road accident analysis is a crucial technique for determining what elements contribute to collisions and can help lower the accident rate. The consideration of sight distance is a crucial component of highway geometric design in order to guarantee effective traffic operations with the highest level of safety at an affordable price. Road accidents frequently result from reduced visibility caused by foggy weather, hence there is increased interest in studying this issue. This study aims to determine and correlate sight distance, fog blockage, and their effects on traffic accidents at Jumja along Thimphu-Phuentsholing Highway (Asian Highway 48). Data gathering and prediction was aided by the usage of Geographic Information Systems and Artificial Neural Network technologies. Temperature, relative humidity and wind speed for seven years from 2017 to 2023 were considered for fog occurrence to train Artificial Neural Network. Visibility was trained and predicted for summer and winter seasons since the occurrence of fog is prevalent during these seasons. As a result, a correlation value R of 0.998 was found, proving the findings to be almost accurate. This study offers a thorough investigation of how sight distance contributes to traffic accidents at Jumja region. A conclusive finding addressing the significance of sight distance as a contributing factor to traffic accidents due to fog invisibility along Jumja region was made after completion of this study.

On Performance of Marine Predators Algorithm in Training of Feed-Forward Neural Network for Identification of Nonlinear Systems

Artificial neural networks (ANNs) are used to solve many problems, such as modeling, identification, prediction, and classification. The success of ANN is directly related to the training process. Meta-heuristic algorithms are used extensively for ANN training. Within the scope of this study, a feed-forward artificial neural network (FFNN) is trained using the marine predators algorithm (MPA), one of the current meta-heuristic algorithms. Namely, this study is aimed to evaluate the performance of MPA in ANN training in detail. Identification/modeling of nonlinear systems is chosen as the problem. Six nonlinear systems are used in the applications. Some of them are static, and some are dynamic. Mean squared error (MSE) is utilized as the error metric. Effective training and testing results were obtained using MPA. The best mean error values obtained for six nonlinear systems are 2.3 × 10−4, 1.8 × 10−3, 1.0 × 10−4, 1.0 × 10−4, 1.2 × 10−5, and 2.5 × 10−4. The performance of MPA is compared with 16 meta-heuristic algorithms. The results have shown that the performance of MPA is better than other algorithms in ANN training for the identification of nonlinear systems.

Influences of particle size distribution and fines content on the excess pore water pressure generation of sand-silt mixtures under cyclic loading

The mechanical behaviors of binary mixtures are different from that of uniform soils. However, experimental studies on the dynamic responses of sand-silt mixtures considering the combined effects of fines content (FC) and host sand gradation are limited in the literature. In this paper, an experimental program was designed concerning five binary mixtures with three grain size ratios (Rd) and three sand uniformity coefficients (Cus) for investigating the effects of fines content and host sand gradation on both the liquefaction resistance and the excess pore water pressure (EPWP) generation of sand-silt mixtures. Results show that the effect of FC on the liquefaction resistance of sand-silt mixtures depends on relative density (Dr), grain size ratio, and sand uniformity coefficient. The liquefaction resistance of sand-silt mixture can be uniquely characterized by the equivalent relative density Dr* regardless of FC, Dr, and host sand gradation. Relationship between the EPWP ratio Ru and the cycle ratio N/NL for the sand-silt mixtures can be well expressed by the Seed model, which is significantly affected by cyclic stress ratio (CSR), FC, and Dr but insensitive to the change of host sand gradation. A trained artificial neural networks (ANN) model is conducted to predict the fitting parameter β of the Seed model with a maximum difference of 25% between the predicted and measured values. Moreover, there is an arctangent relationship between Ru and shear strain amplitude γa of each specimen, and the fitting parameter A for the relationship is sensitive to the particle size distribution and stress amplitude. A unique relationship between a density-corrected parameter A/Dr* and CSR is proposed regardless of FC and host sand gradation.

Flow characteristics of microwave treated Indian coal: A deep learning modelling

To meet the escalating energy needs of the growing population, the power stations are gradually shifting towards abundant low-quality coals and lignites. This study proposes a sustainable methodology to utilize such low-quality coals by cutting down the coal pretreatment costs. Instead of utilizing the treated coal as a whole, this study proposes to blend the treated coal with the untreated coal for further utilization. These blended samples are then used to prepare coal water slurry, with concentrations varying in the range of 10 to 60% by mass. The rheology of the blended samples shows a reduction in viscosity by ∼ 23% when the blended sample contains 50% treated coal. The sustainability of the process is verified by a trained artificial neural network, that includes an extensive validation of predicted results with experimental data and Irvine (1988) model. The artificial neural network indicates a maximum of 11.5% reduction in the headloss when the blended sample is transported at 50% solid concentration. The specific enthalpy consumption is found to be the minimum for the blended samples transported at 50% solid concentration.

IDSNN: Towards High-Performance and Low-Latency SNN Training via Initialization and Distillation.

Spiking neural networks (SNNs) are widely recognized for their biomimetic and efficient computing features. They utilize spikes to encode and transmit information. Despite the many advantages of SNNs, they suffer from the problems of low accuracy and large inference latency, which are, respectively, caused by the direct training and conversion from artificial neural network (ANN) training methods. Aiming to address these limitations, we propose a novel training pipeline (called IDSNN) based on parameter initialization and knowledge distillation, using ANN as a parameter source and teacher. IDSNN maximizes the knowledge extracted from ANNs and achieves competitive top-1 accuracy for CIFAR10 (94.22%) and CIFAR100 (75.41%) with low latency. More importantly, it can achieve 14× faster convergence speed than directly training SNNs under limited training resources, which demonstrates its practical value in applications.

Performance diagnostics of gas turbines operating under transient conditions based on dynamic engine model and artificial neural networks

Gas turbine engines are machines of high complexity and non-linearity. Interpreting the vast amount of data from a gas turbine and converting them into customer value requires the combination of domain knowledge and modern computational intelligence tools. Reaching to an accurate and reliable diagnosis of gas turbines is a process that is becoming increasingly complex. The engines are now expected to operate in more dynamic conditions to compensate for the intermittent nature of renewables. Transient operating conditions will accelerate the deterioration of gas turbine components which motivates the development of new methods and tools to cope with such type of information. In this paper, we propose a novel performance diagnostic method for gas turbines that combines a dynamic engine model with artificial neural networks (ANN). An engine model of a two shaft gas turbine has been developed in MATLAB/Simulink and used by a family of ANNs to detect the degradation of the engine operating under transient conditions, when all of its components are experiencing degradation. The conducted case studies consider various degradation scenarios. The advantage of the proposed method is that it deals effectively with both fixed and time-evolving degradation. Furthermore, in cases where there is limited amount of data for training ANNs the model can fill this gap by simulating plethora of scenarios that can potentially extend the applicability of ANNs to gas turbine diagnostics. The proposed method could be used as a tool for supporting the operation and maintenance activities of gas turbines.

Artificial intelligence driven smart operation of large industrial complexes supporting the net-zero goal: Coal power plants

The true potential of artificial intelligence (AI) is to contribute towards the performance enhancement and informed decision making for the operation of the large industrial complexes like coal power plants. In this paper, AI based modelling and optimization framework is developed and deployed for the smart and efficient operation of a 660 MW supercritical coal power plant. The industrial data under various power generation capacity of the plant is collected, visualized, processed and subsequently, utilized to train artificial neural network (ANN) model for predicting the power generation. The ANN model presents good predictability and generalization performance in external validation test with R2 = 0.99 and RMSE =2.69 MW. The partial derivative of the ANN model is taken with respect to the input variable to evaluate the variable’ sensitivity on the power generation. It is found that main steam flow rate is the most significant variable having percentage significance value of 75.3 %. Nonlinear programming (NLP) technique is applied to maximize the power generation. The NLP-simulated optimized values of the input variables are verified on the power generation operation. The plant-level performance indicators are improved under optimum operating mode of power generation: savings in fuel consumption (3 t/h), improvement in thermal efficiency (1.3 %) and reduction in emissions discharge (50.5 kt/y). It is also investigated that maximum power production capacity of the plant is reduced from 660 MW to 635 MW when the emissions discharge limit is changed from 510 t/h to 470 t/h. It is concluded that the improved plant-level performance indicators and informed decision making present the potential of AI based modelling and optimization analysis to reliably contribute to net-zero goal from the coal power plant.

Artificial Neural Network Assisted Probabilistic and Geometric Shaping for Flexible Rate High-Speed PONs

In this paper, we employ artificial neural networks (ANNs) to optimize joint probabilistic shaping (PS) and geometric shaping (GS) for a realistic 50G IM/DD passive optical network (PON) link. Apart from being able to find a generalized mutual information (GMI)-maximizing modulation for channel conditions unseen at the training phase, the compatibility of the ANN training with Monte Carlo simulation also enables us to use a more complicated channel model that more closely resembles a real PON system where fiber dispersion, bandwidth limitation and digital signal processing (DSP) are present. The forward error correction (FEC) requirement that must be satisfied in an actual implementation is imposed on the learned modulation by including a normalized GMI (NGMI) penalty term in the loss function. The proposed scheme is demonstrated with simulations. Results show that the ANN can achieve similar performance compared to a case-by-case optimization while also being capable of generalizing to a wide range of received optical power (ROP) from −30 dBm to −18 dBm and/or a broad range of fiber distance from 0 km to 20 km. About 0.1-bits/symbol GMI improvement is attained compared to uniform modulation.

Modified Taylor Impact Tests with Profiled Copper Cylinders: Experiment and Optimization of Dislocation Plasticity Model.

Current progress in numerical simulations and machine learning allows one to apply complex loading conditions for the identification of parameters in plasticity models. This possibility expands the spectrum of examined deformed states and makes the identified model more consistent with engineering practice. A combined experimental-numerical approach to identify the model parameters and study the dynamic plasticity of metals is developed and applied to the case of cold-rolled OFHC copper. In the experimental part, profiled projectiles (reduced cylinders or cones in the head part) are proposed for the Taylor impact problem for the first time for material characterization. These projectiles allow us to reach large plastic deformations with true strains up to 1.3 at strain rates up to 105 s-1 at impact velocities below 130 m/s. The experimental results are used for the optimization of parameters of the dislocation plasticity model implemented in 3D with the numerical scheme of smoothed particle hydrodynamics (SPH). A Bayesian statistical method in combination with a trained artificial neural network as an SPH emulator is applied to optimize the parameters of the dislocation plasticity model. It is shown that classical Taylor cylinders are not enough for a univocal selection of the model parameters, while the profiled cylinders provide better optimization even if used separately. The combination of different shapes and an increase in the number of experiments increase the quality of optimization. The optimized numerical model is successfully validated by the experimental data about the shock wave profiles in flyer plate experiments from the literature. In total, a cheap, simple, but efficient route for optimizing a dynamic plasticity model is proposed. The dislocation plasticity model is extended to estimate grain refinement and volume fractions of weakened areas in comparison with experimental observations.