Radial Basis Function-based Neural Networks Research Articles

Driver-Centric Lane-Keeping Assistance System Design: A Noncertainty-Equivalent Neuro-Adaptive Control Approach

Vehicle roadway departure accidents are a major traffic safety concern as they oftentimes result in severe injuries and fatalities. To address such an issue, this article originates a novel driver-centric and neuro-adaptive-control-based lane-keeping assistance system (LKAS). The proposed control strategy synergizes a noncertainty-equivalent adaptive control design scheme, an adaptive radial-basis-function-based neural network (RBFNN) that captures the human driver's lane-keeping steering behavior, and a Gudermannian-function-based smooth parameter projection operator. The benefit and uniqueness of the proposed solution are threefold. First and foremost, the noncertainty-equivalent adaptive control design, which leverages the immersion-and-invariance-like methodology, ensures the asymptotical convergence of the parameter-estimation-error-induced perturbation despite the reference signal's persistency of excitation condition. Second, the LKAS is devised to be driver-centric, i.e., an adaptive RBFNN-based human driver steering model is embedded inside the LKAS's algorithm such that a human driver is assisted in a personalized and adaptive manner. Third, the Gudermannian-function-based smooth parameter projection operator ascertains the prescribed boundedness of the control parameters while maintaining the control action's smoothness. A pilot human-subject study using a high-fidelity moving-base driving simulator is conducted to validate the proposed LKAS. Further, its performance is compared with a baseline certainty-equivalent neuro-adaptive controller.

Stability Analysis and RBF Neural Network Control of Second-Order Nonlinear Satellite System

This paper studies the open-loop stability analysis and radial basis function based neural-network(RBFNN) control of the triangular tethered satellite formation(TTSF) systems which comprise the in-plane and out-plane dynamics. The dynamical equations of TTSF are nonlinear coupled second-order differential equations(DE), whose explicit solutions are hard to work out. Therefore, the dynamics of systems are analyzed by imposing some constraints on initial states, and the norms of proper initialization for such TTSF systems are obtained by solving the DE. As a result, the nonlinear controllers achieving stable tracking to reference states of TTSF systems are designed based on open-loop analysis. Furthermore, when the structure information of the system is uncertain, delay sampled NN(DSNN) as well as offline training NN(OTNN) control algorithms are proposed to train RBFNN, and the optimal weights of NN are computed by least square(LS) method. Then RBFNN control is imposed to the uncertain TTSF system to ensure the stable tracking, and the stability proof of the control is given. Finally, simulations are taken on the specific motions of TTSF system to test the theoretical results.

Observer‐based adaptive backstepping control and the application for a class of multi‐input multi‐output nonlinear systems with structural uncertainties and perturbations

AbstractIn this study, the adaptive backstepping control of a class of multi‐input multi‐output nonlinear systems with immeasurable states, structural uncertainties and periodic perturbations is researched using neural network (NN) based nonlinear observer. Firstly, a series of harmonic components obtained from Fourier series expansion (FSE) are employed to estimate the unknown periodic perturbations. Then treating the estimated perturbations as inputs, a radial basis function‐based neural network (RBFNN) is constructed as a component of the observer, and the observer is proved to be uniformly ultimately bounded (UUB) in estimating the immeasurable system states with the negative‐gradient adaptive laws of NN parameters. Subsequently, the adaptive backstepping control is designed to track reference signals at the outputs of system based on the FSE‐RBFNN observer, and the stability of the closed‐loop system is proved on a finite set of system states. Finally, the proposed methods are applied to a triangular tethered satellite formation model to test the stability of the observer and control, and the effect of FSE in estimating perturbations is comparatively tested as well.

Event-Triggered MFAC of Nonlinear NCSs Against Sensor Faults and DoS Attacks

The event-triggered model-free adaptive control (ET-MFAC) problem for nonlinear networked control systems (NNCSs) with presence of sensor faults and denial-of-service (DoS) attacks is investigated. The attack occured in sensor-to-controller networked channels is supposed to obey the Bernoulli distribution. While sensor faults are formulated as unknown functions which could be approximated by a radial basis function-based neural networks (RBF-NNs). Then an ET-MFAC algorithm is constructed to ensure the tracking performance under sensor faults and DoS attacks based on adaptive estimations. The designed control algorithm is independent with the system structure and only uses the system input and output data. An example with comparison is given to demonstrate the validity of the new ET-MFAC algorithm.

Digital Twin of a Data Center at an Educational Institution

Digital twins are among the most important trends of the fourth industrial revolution. They present a crucial tool for protecting critical mission systems and the development of new services, products, and processes. This paper presents the first digital twin for a data center. The rapid growth of the Internet of things and the areas of modeling and simulation results in high demand for the development of data center digital twins (DCDT) to ensure the safety/protection of critical and costly mission infrastructure and guarantee business continuity, enhance efficiency, and sustain development. This paper presents the design and implementation of a digital twin for a university data center. Different sensory data like temperature, humidity, smoke, and water leakage are analyzed using an intelligent event detection approach, which detects abnormalities using an Extreme Learning Machine (ELM) fed with the minimum ratio between successive real-time data streams. The performance of ELM has outperformed that of both Learning Vector Quantization and Radial Basis Function-based neural network classifiers and proved much faster in abnormal event detection.

High-Precise Bipolar Disorder Detection by Using Radial Basis Functions Based Neural Network

Presently, several million people suffer from major depressive and bipolar disorders. Thus, the modelling, characterization, classification, diagnosis, and analysis of such mental disorders bears great significance in medical research. Electroencephalogram records provide important information to improve clinical diagnosis and are very useful in the scientific community. In this work, electroencephalogram records and patient data from the Hospital Virgen de la Luz in Cuenca (Spain) were processed for a correct classification of bipolar disorders. This work implemented an innovative radial basis function-based neural network employing a fuzzy means algorithm. The results show that the proposed method is an effective approach for discrimination of two kinds of classes, i.e., bipolar disorder patients and healthy persons. The proposed algorithm achieved the best performance compared with other machine learning techniques such as Bayesian linear discriminant analysis, Gaussian naive Bayes, decision trees, K-nearest neighbour, or support vector machine, showing a very high accuracy close to 97%. Therefore, the neural network technique presented could be used as a new tool for the diagnosis of bipolar disorder, considering the possibility of integrating this method into medical software.

Properties of water-based fly ash-copper hybrid nanofluid for solar energy applications: Application of RBF model

The hybrid nanofluids were used as absorber fluids in solar energy applications, which could further increase the efficiency of solar devices. The use of nanofluids in solar devices with the laminar and turbulent flow has received much attention. Presently, the effect of temperature and concentration on thermal conductivity and viscosity of fly ash-copper (80:20% by volume) hybrid nanofluid is investigated. The thermal conductivity and viscosity measurements were carried in the temperature range of 30–60 °C for a concentration range of 0–4.0 vol%. The nanoparticles and nanofluids were characterized by XRF, XRD, SEM, TEM, zeta potential, and DLS techniques. The maximum augmentation in the hybrid nanofluid's dynamic viscosity and thermal conductivity at a concentration of 4 vol% is 45.18% and 49.8%, respectively, at 30 and 60 °C. Correlations to estimate the hybrid nanofluid's dynamic viscosity and thermal conductivity have been proposed considering the results obtained from the present study. A radial basis function-based neural network is used to model nanofluids' effective thermal conductivity and relative viscosity. The outcomes of the experiments were used to calculate the Mouromtseff number and heat transfer efficiency for solar energy applications. • HyNF shows Newtonian behavior in the studied range. • The maximum augmentation in viscosity is 45.2%. • The highest amplification in thermal conductivity is 49.8%. • Radial basis function based neural network was used.

Modeling flexural overstrength factor for steel beams using heuristic soft-computing methods

When evaluating the structural performance of steel moment-resisting frames, significant consideration is given to the flexural behavior of steel beams. This analysis focuses on the primary response parameters, such as the ultimate flexural resistance and rotation capacity that dictate the behavior of the steel beam. In practice, the flexural behavior of steel beams determines the ductile design of steel structures, and a sufficient ductility of steel permits plastic hinge rotations until a full-collapse mechanism is attained. However, it is relatively impossible to guarantee the occurrence of a full-collapse mechanism if the beam’s flexural overstrength is not appropriately quantified by the application of capacity design procedures. Therefore, this study developed analytical formulations to predict the steel beam’s flexural overstrength factor (s) using the multilayer perceptron (MLP)11AMFF: adaptive multilayer feed forward; ANFIS-PSO: adaptive neuro fuzzy system with particle swarm optimization; ANOVA: analysis of variance; GA: genetic algorithm; KKT: Karush–Kuhn–Tucker; LSSVR: least-square support vector regression; M5 Tree: M5 model tree; MAE: mean-absolute errors; MARS: multivariate adaptive regression splines; MLP: multi-layer perceptron; MLR: multiple linear regression; PSO: particle swarm optimization; RBNN: radial basis-function-based neural network; RHS: rectangular hollow section; RMSE: root-mean-square errors; SHS: square hollow sections; SIS: swarm intelligence systems., radial-basis-function-based neural network (RBNN), least-squares support vector regression (LSSVR), adaptive neuro fuzzy system with particle swarm optimization (ANFIS-PSO), multivariate adaptive regression splines (MARS), M5 model tree (M5 Tree), and multiple linear regression (MLR). Numerous cross-sectional geometries were tested with square and rectangular hollow sections, and with I and H sections. The data used for the analysis were sourced from completed experimental studies available in the open literature. The data were categorized as training and testing datasets to develop the models based on the aforementioned methods. For the models, the mechanical properties of the materials, shear length of steel beams, and the geometric properties of the sections constituted the independent variables. Overall, the RBNN demonstrated the best prediction of the steel beam overstrength with a root-mean-square error of 0.099 mm, mean absolute error of 0.066 mm, and a correlation coefficient of 0.825. It was observed that all six machine learning methods provided better estimates than the MLR. The machine learning models developed in this study are expected to serve as a hand tool for engineers and steel structure constructors.

Meta-Models and Genetic Algorithm Application to Approximate Optimization with Discrete Variables for Fire Resistance Design of A60 Class Bulkhead Penetration Piece

A60 class bulkhead penetration piece is a fire-resistance apparatus installed on bulkhead compartments to protect lives and to prevent flame diffusion in case of fire accident in ships and offshore plants. In this study, approximate optimization with discrete variables was carried out for the fire-resistance design of an A60 class bulkhead penetration piece (A60 BPP) using various meta-models and multi-island genetic algorithms. Transient heat transfer analysis was carried out to evaluate the fire-resistance design of the A60 class bulkhead penetration piece, and we verified the results of the analysis via a fire test. The design of the experiment’s method was applied to generate the meta-models to be used for the approximate optimization, and the verified results of the transient heat transfer analysis were integrated with the design of the experiment’s method. The meta-models used in the approximate optimization were response surface model, Kriging, and radial basis function-based neural network. In the approximate optimization, the bulkhead penetration piece length, diameter, material type, and insulation density were applied to discrete design variables, and constraints that were considered include temperature, productivity, and cost. The approximate optimum design problem based on the meta-model was formulated such that the discrete design variables were determined by minimizing the weight of the A60 class bulkhead penetration piece subject to the limit values of constraints. In the context of approximate accuracy, the solution results from the approximate optimization were compared to actual analysis results. It was concluded that the radial basis function-based neural network, among the meta-models used in the approximate optimization, showed the most accurate optimum design results for the fire-resistance design of the A60 class bulkhead penetration piece.

Maximum power point tracking in wind energy conversion system using radial basis function based neural network control strategy

In this paper, a neural network tuned controller for maximum power point tracking (MPPT) in wind energy conversion system (WECS) is proposed. This technique utilizes radial basis function based neural networks (RBF-NN) and tracks the MPP using duty cycle control. The WECS is based on permanent magnet synchronous generator (PMSG). The rectifier output voltage and power are measured and fed into an RBF-NN controller. The MPPT algorithm controls the duty cycle of a DC-DC boost converter to extract the maximum power. The output of the boost converter is tied to the grid using a voltage source inverter (VSI). Unlike conventional methods, MPPT using proposed method does not require knowledge of wind turbine power characteristics, thus minimizes the need for various measuring instruments. The method implemented is also compared with other commonly used MPPT techniques like fuzzy logic control (FLC), perturb & observe (P&O), and back-propagation (BP) based NN. The proposed system provides better results as compared with other relevant results available in the literature. The proposed method is implemented using MATLAB/Simulink and then validated in real-time using digital simulation hardware, OPAL-RT 4510.

Radial basis function neural network algorithm for semi-active control of base-isolated structures

Curved surface slider (CSS) is considered as an effective isolation device for structures subjected to earthquake ground motions. Due to constant frequency, CSS may encounter a resonance problem when subjected to near-fault earthquake ground motions. To overcome this problem, we propose CSS combined with a control device in this study. The control device consists of variable orifice fluid damper, and its damping coefficient is controlled by a radial basis function-based neural network algorithm. Numerical simulations are performed to evaluate the effectiveness of the proposed technique for only one-directional horizontal seismic excitations without any evaluation concerning the durability of CSSs. The results of the investigation demonstrate that the proposed technique is effective to reduce both the base shear and the sliding displacement of the isolated structure. In addition, the response predicted by the proposed technique is almost similar to the response of isolated structure with passive damper at optimum damping ratio.

Explicit MPC-Based RBF Neural Network Controller Design With Discrete-Time Actual Kalman Filter for Semiactive Suspension

Many applications require fast control action and efficient constraint handling, such as in aircraft or vehicle control, where instead of the slow online computation of the model predictive control (MPC) the explicit MPC can be an alternative solution. Explicit MPC controllers consist of several affine feedback gains, each of them valid over a polyhedral region of the state space. The exponential blow-up of the number of regions with increasing the prediction horizon increases the searching time among the regions extremely which together with the requirement of the full state measurement decreases its applicability for real systems. First, discrete-time actual Kalman filter is designed for the semiactive suspension and applied to explicit MPC controller that requires only measurement of the suspension deflection. Second, this paper presents a systematic way to design Gaussian radial basis function-based neural network (NN) approximation of the explicit MPC controller and shows that a well-tuned NN with some neurons can replace the explicit MPC controller. This nonlinear state-feedback controller can ensure the fast control action but price of the approximation is some deterioration of the performance value. The complete novel nonlinear control system with Kalman filter is analyzed in detail. The derived controllers are evaluated through simulations, where shock tests and white noise velocity disturbances are applied to a real quarter car vertical model.

MODELING OF STREAMFLOW- SUSPENDED SEDIMENT LOAD RELATIONSHIP BY ADAPTIVE NEURO-FUZZY AND ARTIFICIAL NEURAL NETWORK APPROACHES (CASE STUDY: DALAKI RIVER, IRAN)

Modeling of stream flow–suspended sediment relationship is one of the most studied topics in hydrology due to itsessential application to water resources management. Recently, artificial intelligence has gained much popularity owing toits application in calibrating the nonlinear relationships inherent in the stream flow–suspended sediment relationship. Thisstudy made us of adaptive neuro-fuzzy inference system (ANFIS) techniques and three artificial neural networkapproaches, namely, the Feed-forward back-propagation (FFBP), radial basis function-based neural networks (RBF),geomorphology-based artificial neural network (GANN) to predict the streamflow suspended sediment relationship. Toillustrate their applicability and efficiency,, the daily streamflow and suspended sediment data of Dalaki River station insouth of Iran were used as a case study. The obtained results were compared with the sediment rating curve (SRC) andregression model (RM). Statistic measures (RMSE, MAE, and R2) were used to evaluate the performance of the models.From the results, adaptive neuro-fuzzy (ANFIS) approach combined capabilities of both Artificial Neural Networks andFuzzy Logic and then reflected more accurate predictions of the system. The results showed that accuracy of estimationsprovided by ANFIS was higher than ANN approaches, regression model and sediment rating curve. Additionally, relatingselected geomorphologic parameters as the inputs of the ANN with rainfall depth and peak runoff rate enhanced theaccuracy of runoff rate, while sediment loss predictions from the watershed and GANN model performed better than theother ANN approaches together witj regression equations in Modeling of stream flow–suspended sediment relationship.

Comparison of numerical and neural network methods for the kinematic modeling of a hybrid structure robot

A combination of parallel and serial mechanisms allows for stiff and dexterous motion of the robotic end-effector but increases the complexity of the kinematic problem. Many of the geometric parameters for such robots are difficult to obtain. This paper presents the numerical and neural network methods to solve the kinematics for such a hybrid robot named Exechon®. Both methods avoid the geometric measurement in the real robot. The geometric parameters used in the numerical model are identified by a particle swarm optimization algorithm. At the same time, a radial basis function-based neural network is trained to approximate the kinematics of the Exechon robot. The resultant models are then compared in terms of modeling accuracy and real-time ability. The presented methods are generic and can be applied to other robots with similar structures.

Comparison of Artificial Neural Network Methods with L-moments for Estimating Flood Flow at Ungauged Sites: the Case of East Mediterranean River Basin, Turkey

A regional flood frequency analysis based on the index flood method is applied using probability distributions commonly utilized for this purpose. The distribution parameters are calculated by the method of L-moments with the data of the annual flood peaks series recorded at gauging sections of 13 unregulated natural streams in the East Mediterranean River Basin in Turkey. The artificial neural networks (ANNs) models of (1) the multi-layer perceptrons (MLP) neural networks, (2) radial basis function based neural networks (RBNN), and (3) generalized regression neural networks (GRNN) are developed as alternatives to the L-moments method. Multiple-linear and multiple-nonlinear regression models (MLR and MNLR) are also used in the study. The L-moments analysis on these 13 annual flood peaks series indicates that the East Mediterranean River Basin is hydrologically homogeneous as a whole. Among the tried distributions which are the Generalized Logistic, Generalized Extreme Vaules, Generalized Normal, Pearson Type III, Wakeby, and Generalized Pareto, the Generalized Logistic and Generalized Extreme Values distributions pass the Z statistic goodness-of-fit test of the L-moments method for the East Mediterranean River Basin, the former performing yet better than the latter. Hence, as the outcome of the L-moments method applied by the Generalized Logistic distribution, two equations are developed to estimate flood peaks of any return periods for any un-gauged site in the study region. The ANNs, MLR and MNLR models are trained and tested using the data of these 13 gauged sites. The results show that the predicting performance of the MLP model is superior to the others. The application of the MLP model is performed by a special Matlab code, which yields logarithm of the flood peak, Ln(QT), versus a desired return period, T.

Comparison of three artificial neural network approaches for estimating of slake durability index

Slake durability index (I d2) is an important engineering parameter to assess the resistance of clay-bearing and weak rocks to erosion and degradation. Standard test sample preparation for slake durability test is difficult for some rock types and the test is time-consuming. The paper reports an attempt to define I d2 using other parameters that are simpler to obtain. In this study, three different artificial neural network approaches, namely feed-forward back propagation (FFBP), radial basis function based neural network (RBNN), and generalized regression neural networks (GRNN) were used for estimating I d2. The determination coefficient (R 2), root mean square error and mean absolute relative error statistics were used as evaluation criteria of the FFBP, RBNN, and GRNN models. The experimental results were compared with these models. The comparison results indicate that the GRNN models are superior to the FFBP and RBNN models in modeling of the slake durability index (I d2).

Design of fuzzy radial basis function-based polynomial neural networks

In this study, we introduce a new design methodology of fuzzy radial basis function-based polynomial neural networks. In many cases, these models do not come with capabilities to deal with granular information. With this regard, fuzzy sets offer several interesting and useful opportunities. This study presents the development of fuzzy radial basis function-based neural networks augmented with virtual input variables. The performance of the proposed category of models is quantified through a series of experiments, in which we use two machine learning data sets and two publicly available software development effort data.

Predicting Breast Screening Attendance Using Machine Learning Techniques

Machine learning-based prediction has been effectively applied for many healthcare applications. Predicting breast screening attendance using machine learning (prior to the actual mammogram) is a new field. This paper presents new predictor attributes for such an algorithm. It describes a new hybrid algorithm that relies on back-propagation and radial basis function-based neural networks for prediction. The algorithm has been developed in an open source-based environment. The algorithm was tested on a 13-year dataset (1995-2008). This paper compares the algorithm and validates its accuracy and efficiency with different platforms. Nearly 80% accuracy and 88% positive predictive value and sensitivity were recorded for the algorithm. The results were encouraging; 40-50% of negative predictive value and specificity warrant further work. Preliminary results were promising and provided ample amount of reasons for testing the algorithm on a larger scale.

Artificial neural network methods for the prediction of framework crystal structures of zeolites from XRD data

Extracting information about the structures of zeolites and other crystalline materials from X-ray diffraction (XRD) data simply by using statistical methods may provide an impetus for the discovery and identification of unknown materials. In this study, the possibility of using artificial neural network methods for relating framework crystal structures to XRD data reported in literature was investigated. Generalized Regression Neural Networks and Radial Basis Function-Based Neural Networks were utilized in the investigations. The results obtained by neural networks, using fivefold cross validation technique, were compared to the actual values as well as to those determined by multilinear regression. The predictions made by these neural network methods were, in general, more reliable than those performed by regression. The best predictions were achieved for the estimation of the framework densities of zeolites, which provided quite small deviations from the actual values.

Compared techniques for the critical submergence of an intake in water flow

Air-entraining vortex at intake is an important problem encountered in hydraulic engineering. Intake submergence depth could result in formation of the air-entraining free surface vortices. Unless the dangerous air entrainment is eliminated, air entraining causes mechanical damage, vibration in pipelines and loss of pump performance. The value of the intake's submergence when the vortex starts entraining air is known as ''critical submergence''. In this study, the critical submergence for a circular intake pipe in still-water and open channel flow for permeable and impermeable bottom was investigated. Experimental results were used to compare with critical spherical sink surface (CSSS), radial basis function based neural network (RBNN) and general linear model (GLM). The CSSS has the same center and discharge as the intake with the critical submergence. The GLM underlies most of the statistical analyses that are used in applied research. And the RBNN is one of the most used network models. The ranking of prediction on critical submergence is obtained as RBNN, GLM and CSSS, respectively.