neural network

In this paper, a new intrusion detection system (IDS) is presented to deal with distributed denial of service (DDoS) attacks. A combined algorithm based on Harris Hawks Optimization (HHO) and Dragonfly Algorithm (DA) is proposed to select relevant features and eliminate irrelevant and redundant features from the NSL-KDD dataset. The extracted features are presented to a multilayer perceptron (MLP) neural network. This network (as a classifier) divides the network traffic into two classes, normal and attack categories. Performance of the proposed model is evaluated with two standard and widely-used datasets in the field of intrusion detection: NSL-KDD and UNSW-NB15. The results of the simulations clearly show the superiority of the proposed method compared to the previous methods in terms of critical evaluation criteria such as accuracy, precision, recall, and F-Measure. Specifically, the proposed method exhibited improvements of 96.9%, 97.6%, 96%, and 96.8% in these metrics, respectively (compared to the baseline method). The main reason for these improvements is the ability of the combined algorithm to intelligently select the optimal features and reduce the dimensions of the data. This careful selection of features allows the MLP neural network to focus on critical information, increasing the classification accuracy and ultimately improving the performance of the intrusion detection system. This research showed that combining optimization algorithms and machine learning works well. So, it is effective for tackling DDoS attacks. It can lead to better intrusion detection systems. These systems will be more efficient and accurate.

The control set in most moving vehicles in water faces an interferential and nonlinear system. Specifically, in a hovercraft, due to the assignment of underactuated and insufficient actuators, the control effect is highly interferential in the channels. In this vehicle, the dynamics change significantly in each maneuver (speed). With rudder deflection in sway motion, the surge channel is affected, and similarly, with a change in surge velocity, the sway channel behavior is completely transformed. In this study, by identifying the desired behavior, initially, a sliding mode controller with the requirement of minimal chattering in commands for the surge and sway velocity channels is designed. The sliding mode controller is a robust controller whose stability can be proven. Since this controller is designed for a specific system characteristic of the hovercraft and with conservative variables, it does not necessarily exhibit suitable behavior with small tracking error for all maneuvers and uncertainties. Inevitably, using reinforcement learning and the PPO method, the initial controller is adjusted for most possible states with the constraint of reducing chattering and increasing tracking accuracy. Uncertainties in system characteristics and motion maneuvers are modeled in the simulation program and used in the learning. Learning calculations are performed offline. The result is applied as a trained actor-critic neural network to the initial sliding mode controller. The research results show that by tuning the controller with machine learning, precise commands are executed without large oscillations being introduced into the control. Additionally, the average cumulative reward increases by at least 40%.

Accurate estimation of bond strength between concrete and deformed reinforcing bars is essential for the stability of reinforced concrete structures, especially in critical regions subjected to heavy loads and environmental stresses. Despite intensive experimental studies revealing the complexity of factors influencing bond strength, existing predictive models, often reliant on artificial neural networks, have limitations in accuracy due to constrained datasets and inadequate representation of real-world stress fields. In response, this study pioneers a novel hybrid metaheuristic-optimized neural network model to swiftly and precisely predict bond strength under tensile load. Utilizing a comprehensive dataset comprising 558 valid experimental outcomes, seven metaheuristic algorithms are employed to optimize the ANN architecture. These metaheuristic algorithms include the Weighted Mean of Vectors, Grey Wolf Optimizer, Energy Valley Optimizer, Circle Search Algorithm, Artificial Ecosystem-Augmented Optimization, War Strategy Optimization, and Brown-Bear Optimization Algorithm. Results demonstrate that the developed hybrid models, particularly the artificial neural networks optimized by the Weighted Mean of Vectors algorithm, exhibit superior predictive performance. This model also demonstrated the lowest miscalibration value, followed by Circle Search Algorithm and Energy Valley Optimizer, indicating a high level of reliability. Moreover, comparison with common analytical and empirical formulations revealed significant performance improvements of the proposed model, achieving a 25% reduction in MSE during the testing phase. Additionally, the Shapley Additive explanations and Sobol sensitivity analysis framework was used to interpret the proposed predictive model, highlighting key predictors such as cross-sectional area, development length or splice, reinforcing bar diameter, and concrete compressive strength.

Countries are managed based on accurate and precise laws. Enacting appropriate and timely laws can cause national progress. Each law is a textual term that is added to the set of existing laws after passing a process with the approval of the assembly. In the review of each new law, the relevant laws are extracted and analyzed among the set of existing laws. This paper presents a new solution for extracting the relevant rules for a term from an existing set of rules using semantic similarity and deep learning techniques based on the BERT model. The proposed method encodes sentences or paragraphs of text in a fixed-length vector (dense vector space). Thereafter, the vectors are utilized to evaluate and score the semantic similarity of the sentences with the cosine distance measurement scale. In the proposed method, the machine can understand the meaning and concept of the sentences by using the BERT model coding method. The BERT model considers the position of the entities in the sentences. Then the semantic similarities of documents, calculating the degree of similarity between their documents with a subject, and detecting their semantic similarity are done. The results obtained from the test dataset indicated the precision and accuracy of the method in detecting semantic similarities of legal documents related to the Islamic Consultative Assembly of Iran, as well as the precision and accuracy of performance above 90%.

Inverters have been widely used in renewable energy as a means of converting extracted power to grid standards. However, this power electronics equipment is highly vulnerable to failures due to its complex architecture and components. One of the main sources of failure is semiconductor switches that are critically sensitive to abnormal conditions such as high voltage. With the advent of multilevel inverters, this concern has been raised considerably due to the increase in the number of switches. This paper has proposed a novel method with a neural network that can detect open circuit failure of switches and replace them with some new arrangement in the inverter so that it can run effectively. Simulations with MATLAB/Simulink for a seven level inverter, illustrate that with a switch failure, the multilevel inverter can work successfully. Results also demonstrate that this method is fast and can compensate for the output in less than two cycles. Therefore, it can be used in reliable multilevel applications in which the power flow should be achieved even if a semiconductor switch is broken.

Wind energy has been identified as a critical component in the growth of all countries throughout the world. Nigeria has been identified as having energy issues as a result of poor maintenance of hydro and thermal energy generating stations. As a result, the current study uses some machine learning approaches over wind speed data for energy generation in the country. Machine learning models were employed for wind speed using selected meteorological parameters. Little research was done using some meteorological data and machine learning to investigate wind speed across Nigerian sub-stations, resulting in the need for further research. This research, on the other hand, focuses on a neural network for forecasting, a Long Short-Term Memory (LSTM) network model based on several fire-work algorithms (FWA). The data for this study came from the archive of the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) Web service, which was modeled. The LSTM predicts the wind speed model based on the FWA, which used hyper-parameter optimization and was based on a real-time prediction model that was dependent on the change and dependence of the neural network. The study data was split into two categories: test and training. According to the validation technique, the sample data was reviewed, and the first 80 % of the data was utilized for training, as revealed by the (LSTM) network model. The remaining 20 % of the data was used as forecast data to ensure that the model was accurate. The normalization of the data for the wind speed range of 0 to 1 which illustrates the process data, the high peak in 1985 (a = 0.12 m/s, b = 0.11 m/s, c = 0.13 m/s, d = 0.08 m/s, e = 0.06 m/s, f = 0.10 m/s) was discovered. However, the summary result of the performances of different 11 Machine Learning algorithms of regression type for each of the seven locations in Nigeria has different values. As a result, it is recommended that this study will facilitate the prediction of wind speed for energy generation in Nigeria.

In this paper, the optimal parameters of the FSW welding process to improve the joint's mechanical properties are obtained using robust multi-objective optimization. First, the properties of the weld zone, such as the chemical composition of the weld, are investigated using scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS). The hardness and tensile properties of the weld were investigated to evaluate the mechanical properties of the joint. The results show at the AA7075 side, the highest hardness is observed in the TMAZ, and the hardness is reduced in the SZ. Tensile testing revealed that the joint's mechanical characteristics were superior to those of the basic metals. In order to obtain the relationship between the process input parameters and the mechanical properties of the obtained joint, an artificial neural network model (ANN) was used. The relationship obtained by ANN was then used to obtain the optimal values of process parameters considering uncertainties in a robust optimization algorithm. In this way, using such an obtained feed-forward neural network and the Monte Carlo simulation, a multi-objective genetic algorithm is used for the robust Pareto optimization of the friction stir welding parameters having probabilistic uncertainties in parameters. Finally, the Technique for Order Preference by Similarity to the Ideal Solution (TOPSIS) was used to get the best optimum solution. The robust optimal process parameters were determined by robust multivariate optimization to be 1467 rpm rotational speed and 11 mm/min traverse velocity.

Investigating the behavior of the box-shaped column panel zone has been one of the major concerns of scientists in the field.  In the American Institute of Steel Construction the shear capacity of I-shaped cross- sections with low column thickness is calculated. This paper determines the shear capacity of panel zone in steel columns with box-shaped cross-sections by using artificial neural network (ANN) and genetic algorithm (GA). It also compares ABAQUS finite element software outputs and AISC relations. Therefore, neural networks were trained using parametric information obtained from 510 connection models in ABAQUS software. The results show that the predicted shear capacity of the NN and the GA in comparison with the AISC relations use a wide range of all effective parameters in the calculation of the shear capacity of panel zone. Therefore, the use of artificial intelligence can be a good choice. Finally, the GA, along with optimization of a mathematical relation, has been able to minimize the error in determining the shear capacity of panel zones of steel-based columns, even at high column thicknesses.

Neural network is one of the new soft computing methods commonly used for prediction of the thermodynamic properties of pure fluids and mixtures. In this study, we have used this soft computing method to predict the coefficients of the Antoine vapor pressure equation. Three transfer functions of tan-sigmoid (tansig), log-sigmoid (logsig), and linear were used to evaluate the performance of different transfer functions to redict the coefficients of the Antoine vapor pressure equation. The critical pressure, critical temperature, critical volume, molecular weight, and acentric factor were considered as the input variables and the Antoine equation coefficients showed by the symbols A, B, and C were considered as the output variables. The results of this study indicated that the linear transfer function had a better performance than other transfer functions and the topology of 5-6-3 with Levenberg–Marquardt learning algorithm and linear transfer function had the best performance for prediction of these coefficients.

This study examined and presents an effective method for detection of failure of conductor bars in the winding of rotor of induction motor in low load conditions using neural networks of radial-base functions. The proposed method used Hilbert method to obtain the stator current signal push. The frequency and signal amplitude of the push stator were used as the input of the neural network and the network outputs were rotor fault state, and the number of conductive bars with broken fault. Moreover, particle-swarm optimization algorithm was used to determine the optimal network weights and neuron penetration radius in the neural network. The results obtained from the proposed method showed the optimal and efficient performance of the method in detecting conductive bars broken fault in induction motor in low load conditions.

Classification of various rice types and determination of its quality is a major issue in the scientific and commercial fields associated with modern agriculture. In recent years, various image processing techniques are used to identify different types of agricultural products. There are also various color and texture-based features in order to achieve the desired results in this area. In this paper, the problem of rice categorization and quality detection using compressive sensing concepts is considered. This issue includes sparse representation and dictionary learning techniques to achieve over-complete models and represent the structural content of rice variety. Also, dictionaries are learned in such a way to have the least coherence values to each other. The results of the proposed classifier based on the learned models are compared with the results obtained from the neural network and support vector machine classifiers. Simulation results show that the proposed method based on the combinational features is able to identify the type of rice grain and determine its quality with high accuracy rate.

In this paper sensitivity analysis of a wideband backward-wave directional coupler due to fabrication imperfections is done using Monte Carlo method. For using this method, a random stochastic process with Gaussian distribution by 0 average and 0.1 standard deviation is added to the different geometrical parameters of the coupler and the frequency response of the coupler is estimated. The applied process must be done several times for converging Monte Carlo method. Therefore, a large number of simulations is reqired for the coupler. This may take a long time if one uses High Frequency Structure Simulator (HFSS) as the simulation software. To decrease the required time of analysis, neural network model of the coupler incnjuction with Mone Carlo is used. Results showed that the bandwidth of the coupler, minimum return loss in passband and minimum isolation in passband won’t be change considerably using the sepecified value of random process. The obtained results for a prototype of a backward wave coupler is presented, which confirm the results of the sensitivity analysis.

Super-resolution (SR) is a technique that produces a high resolution (HR) image via employing a number of low resolution (LR) images from the same scene. One of the degradations that attenuates performance of the SR is the blurriness of the input LR images. In many previous works in the SR, the blurriness of the LR images is assumed to be due to the integral effect of the image sensor of the image acquisition device, while in practice, there are some other factors that blur the LR images, such as diffraction, motion of the object and/or acquisition device, atmospheric blurring and defocus blur. To apply the super-resolution process accurately, we need to know the degradation model applied on HR image leading to LR ones. In this paper, we aim to use the LR images blurriness to find the blurring kernel applied on the HR image. Hence we setup a simulation experiment in which the blurring kernel is limited to be one of the predetermined kernels. In the experiment, the blurriness of the LR images is supposed to be unknown, and is estimated using a blur kernel estimation method. Then the estimated blur kernels of the LR images are fed to an artificial neural network (ANN) to determine the blur kernels associated with the HR image. Experiment results show the use of determined blur kernels improves the quality of output HR image.

Software Defined Network is a new architecture for network management and its main concept is centralizing network management in the network control level that has an overview of the network and determines the forwarding rules for switches and routers(the data level).Although this centralized control is the main advantage of SDN, it is also a single point of failure. If this main control is made unreachable for any reason, the architecture of the network is crashed. A DDoS attack is a threat for the SDN controller that can make it unreachable. Most of the previous works in DDoS detection in SDN focus on early detection of DDoS and not enough work have been done on improvement of accuracy in detection. The proposed solution of this research can detect DDoS attack on SDN controller with a noticeable accuracy and prevents serious damage to the controller .For this purpose, fast entropy of each flow is computed at certain time intervals. Then by the use of adaptive threshold, the possibility of a DDoS attack is investigated. In order to achieve more accuracy, another method, computing flow initiation rate, is used alongside. After observation the results of this two methods, according to the conditions described later, the existence of an attack is confirmed or rejected, or this decision is made at the next step of the algorithm, with further study of flow statistics of network switches by the perceptron neural network.

Hysteresis loop curves are highly important for numerical simulations of materials deformation under cyclic loadings. The models mainly take account of only the tensile half of the stabilized cycle in hysteresis loop for identification of the constants which don’t vary with accumulation of plastic strain and strain range of the hysteresis loop. This approach may be quite erroneous particularly if the mean stress is not small and the effect of isotropic hardening is large. A strain dependent cyclic plasticity model which considers the variation of material constants versus strain range and accumulation of plastic strain has been proposed and experimentally investigated by the authors. In this paper it is proved that their proposed model is accurate for simulating all cycles of the hysteresis loop regardless of the strain range of the test. It is shown in this work that artificial neural network (ANN) model, if designed and trained properly, can be used for interpolating and extrapolating the experimental data. The results of this work are compared with two well-known cyclic plasticity models. The results also indicate that there is a remarkable agreement between the proposed model and ANN within and outside the strain ranges used in the experiments.

Nowadays steel balls wear is a major problem in mineral processing industries and forms a significant part of the grinding cost. Different factors are effective on balls wear. It is needed to find models which are capable to estimate wear rate from these factors. In this paper a back propagation neural network (BPNN) and multiple linear regression (MLR) method have been used to predict wear rate of steel balls using some significant parameters including, pH, solid content, throughout of grinding circuit, speed of mill, charge weight of balls and grinding time. The comparison between the predicted wear rates and the measured data resulted in the correlation coefficients (R), 0.977 and 0.955 for training and test data using BPNN model. However, the R values were 0.936 and 0.969 for training and test data by MLR method. In addition, the average absolute percent relative error (AAPE) obtained 2.79 and 4.18 for train and test data in BPNN model, respectively. Finally, Analysis of the predictions shows that the BPNN and MLR methods could be used with good engineering accuracy to directly predict the wear rate of steel balls.

In some statistical process control applications, the process data are not Normally distributed and characterized by the combination of both variable and attributes quality characteristics. Despite different methods which are proposed separately for monitoring multivariate and multi-attribute processes, only few methods are available in the literature for monitoring multivariate-attribute processes. In this paper, we develop discriminant analysis technique for monitoring the mean vector of correlated multivariate-attribute quality characteristics in the first module. Then in the second module, a novelty approach based on the combination of artificial neural network (ANN) and discriminant analysis is proposed for detecting different mean shifts. The proposed approach is also able to diagnose quality characteristic(s) responsible for out-of-control signals after detecting different step mean shifts. A numerical example based on simulation is given to evaluate the performance of the proposed methods for detection and diagnosis purposes. The detecting performance of the second module is also compared with the extended T2 control chart and with the extension of an ANN in the literature. The results confirm that the proposed method outperforms both methods.

In this paper, experimental responses of the clamped mild steel, copper, and aluminium circular plates are presented subjected to blast loading. The GMDH-type (Group Method of Data Handling) neural networks are then used for the modelling of the mid-point deflection thickness ratio of the circular plates using those experimental results. The aim of such modelling is to show how the mid-point deflection varies with the variations of the important parameters. Further, it is shown that the use of dimensionless input variables, rather than the actual physical parameters, in such GMDH type network modelling leads to simpler polynomial expressions which can be used for modelling and prediction purposes. It is also demonstrated that Singular Value Decomposition (SVD) can be effectively used to find the vector of coefficients of quadratic sub-expressions embodied in such GMDH-type networks. Such application of SVD will highly improve the performance of GMDH-type networks to model the nonlinear dynamic behavior of circular plates.

In this research, at first, the natural frequencies of a cracked beam are obtained analytically, then, location and depth of a crack in beam is identified by neural network method. The research is applied on a beam with an open crack for three different boundary conditions. For this purpose, at first, the natural frequencies of the cracked beam are obtained analytically, to get the examples for training neural network. Then, inversely, the neural network which has been trained by obtained the natural frequencies came from analytically analysis, is used for obtaining the location and depth of the crack. The effect of numbers of natural frequencies as input of the network was evaluated on the prediction accuracy. Results and measure of errors show that the neural network is a powerful method to determine the location and depth of crack. Also, increasing the mode numbers of the natural frequencies give rise the prediction accuracy to be increased

Quality inspection is an indispensable part of modern industrial manufacturing. Steel as a major industry requires constant surveillance and supervision through its various stages of production. Continuous casting is a critical step in the steel manufacturing process in which molten steel is solidified into a semi-finished product called slab. Once the slab is released from the casting unit, the surface often has longitudinal or transverse cracks. Being exposed to air, the crack surfaces oxidize and do not weld during rolling. The early detection of these defects on the slab saves significant time, effort and production expense, reduces costs, and prevents wasted processing steps and rolling mill faults. Traditionally, the inspection process has been carried out visually through human inspectors. However, human inspection is subjective, error-prone, tedious and time consuming. This paper presents an initial study to validate the feasibility of automated inspection of continuously cast hot slabs using computer vision techniques. An automated inspection system such as the one described in this paper can inspect a slab coming out of a caster while it is still hot. The image processing techniques applied in this work including wavelet transform, morphological operations, edge detection and clustering are time-efficient and simply applicable in industrial applications which demand online computations. The experimental results with 97.0% sensitivity and 96.0% specificity demonstrated that the proposed algorithm was effective and reliable. To the best of our knowledge, this is the first time that such a computerized algorithm has been applied in Iran’s steel industry for quality inspection of continuously cast hot slabs.