Feed-forward Artificial Neural Network Architecture Research Articles

A Comparative Study in Forming Behavior of Different Grades of Steel in Cold Forging Backward Extrusion by Integrating Artificial Neural Network (ANN) with Differential Evolution (DE) Algorithm

The cold forging backward extrusion is employed to produce parts that are characterized by better mechanical strength. However, in this process, punches are often prone to breakages because of the large forces encountered in deforming the steel billets. The service life of the punches is affected majorly by the geometrical attributes, the type of steel undergoing deformation, and hence the present investigation focuses on the applications of natural computing algorithms such as artificial neural network (ANN) and differential evolution (DE) optimization algorithm to study the differential influence on the forming behavior of different grades steel and enhance the punch service life. The AISI steel grades, such as AISI 1010, 1018, and 1045, employed extensively in the production of automotive components, have been compared in terms of forming behavior, such as effective stress, strain, strain rate, and punch force. The multi-layer feed-forward ANN architecture was utilized for process modeling with forming responses of finite element (FE) simulations that are strategically planned through the design of experiments (DoE) approach. Considerable variations were found for the effective stress and punch force amongst the steels, while marginal deviations were observed for effective strain and strain rates. Confirmatory experiments were conducted to validate the results of optimal combinations obtained through the DE optimization technique, and the deviations were observed to be in the acceptable range. The cold forging backward extruded components have also been examined for better mechanical soundness through microstructure and micro-hardness analysis that clearly revealed the mechanical integrity and strength enhancement within the forged components. The proposed study would assist the industries engaged in the production of cold-forged steel components in determining the appropriate values of variables to minimize the forming responses and, thus, help in enhancing the life of the tooling.

Artificial neural network to predict traction performance of autonomous ground vehicle on a sloped soil bin and uncertainty analysis

• ANN models were developed to estimate the AGV’s traction behavior on slopes. • Uncertainty analysis quantified the reliability and robustness of ANN model output. • ANN is a promising, robust, and reliable method to predict off-road vehicle’s traction performance. • Predictive traction models would empower the multi-AGV’s operation on a high slope. A fleet of autonomous ground vehicles (AGV) is envisioned to expand farming to arable land suitable for production except for being too steep for conventional equipment. The success of proposed multi-AGV system largely depends on the traction performance of the individual AGVs on unevenly sloped terrain and optimization of the AGVs control variables. Therefore, the drawbar pull performance of a prototype AGV was evaluated in a soil bin at varying slopes, speeds, and drawbar pull (DP). The AGV’s traction performance was expressed in three metrics: tractive efficiency (TE), travel reduction ratio (TRR), and power number (PN). Optimizing the control variables is intricate and ill-defined, which requires an accurate model to predict the performance of the proposed multi-AGV system. Hence, this study aims to design an artificial neural network (ANN) to estimate the traction behavior of the AGV on a sloped testbed as a function of AGV’s speed, applied DP, and slope. A multi-layer perceptron feed-forward ANN architecture with a single hidden layer trained with a back-propagation algorithm was adopted. A series of ANN models with increasing complexity and different hidden layer activation functions were developed for each response variable, i.e., ANN-TE, ANN-TRR, and ANN-PN. A re-sampling-based method, K-fold cross-validation, was employed to estimate the model generalization error. The model success was evaluated via Mean Squared Error (MSE) and the Coefficient of Determination ( R 2 ) against a test set. The final predictive model was trained on the entire data set, and the observed R 2 was 0.933, 0.882 and 0.858, respectively, for ANN-TE, ANN-TRR, and ANN-PN. Subsequently, a Monte-Carlo Simulation based uncertainty analysis was carried out to demonstrate the model strength and the degree of uncertainty by constructing a 95% prediction interval. This study shows ANN as a promising, robust, and reliable method to predict traction performance in agricultural tillage-traction studies and developed models can empower the multi-AGV system on steep-uneven slope terrain.

Multischeme feedforward artificial neural network architecture for DDoS attack detection

Distributed denial of service attack classified as a structured attack to deplete server, sourced from various bot computers to form a massive data flow. Distributed denial of service (DDoS) data flows behave as regular data packet flows, so it is challenging to distinguish between the two. Data packet classification to detect DDoS attacks is one solution to prevent DDoS attacks and to maintain server resources maintained. The machine learning method especially artificial neural network (ANN), is one of the effective ways to detect the flow of data packets in a computer network. Based on the research that has carried out, it concluded that ANN with hidden layer architecture that contains neuron twice as neuron on the input layer (2n) produces a stable detection accuracy value on Quasi-Newton, Scaled-Conjugate and Resilient-Propagation training functions. Based on the studies conducted, it concluded that ANN Architecture sufficiently affected the Scaled-Conjugate and Resilient-Propagation training functions, otherwise the Quasi-Newton training function. The best detection accuracy achieved from the experiment is 99.60%, 1.000 recall, 0.988 precision, and 0.993 f-measure using the Quasi-Newton training function with 6-(12)-2 neural network architecture.

Prediction of the individual multileaf collimator positional deviations during dynamic IMRT delivery priori with artificial neural network.

Multileaf collimator (MLC) positional accuracy during dynamic intensity modulation radiotherapy (IMRT) delivery is crucial for safe and accurate patient treatment. The deviations of individual leaf positions from its intended positions can lead to errors in the dose delivered to the patient and hence may adversely affect the treatment outcome. In this study, we propose a state-of-the-art machine learning (ML) method based on an artificial neural network (ANN) for accurately predicting the MLC leaf positional deviations during the dynamic IMRT treatment delivery priori using log file data. Data of ten patients treated with sliding window dynamic IMRT delivery were retrospectively retrieved from a single-institution database. The patients' plans were redelivered with no patient on the couch using a Varian linear accelerator equipped with a Millennium 120 HD MLC system. Then the machine recorded log files data, a total of over 400 files containing 360 800 control points, were collected. A total of 14 parameters were extracted from the planning data in the log files such as leaf planned positions, dose fraction, leaf velocity, leaf moving status, leaf gap, and others. Next, we developed a feed-forward ANN architecture mapping the input parameters with the output to predict the MLC leaf positional deviations during the delivery priori. The proposed model was trained on 70% of the total data using the delivered leaf positional data as a target response. The trained model was then validated and tested on 30% of the available data. The model accuracy was evaluated using the mean squared error (MSE), regression plot, and error histogram. The deviations between the individual MLC planned and delivered positions can reach up to a few millimeters, with a maximum deviation of 1.2mm. The predicted leaf positions at control points closely matched the delivered positions for all MLC leaves during the treatment delivery. The ANN model achieved a maximum MSE of 0.0001mm2 (root MSE of 0.0097mm) in predicting the leaf positions at control points of test data for each leaf. The correlation coefficient, that measures the goodness of fit, was perfect (R=0.999) in all plots indicating an excellent agreement between the predicted and delivered MLC positions for the training, validation, and test data. We successfully demonstrated a proposed ANN-based method capable of accurately predicting the individual MLC leaf positional deviations during the dynamic IMRT delivery priori. Our ML model based on ANN outperformed the reported accuracy in the literature of various ML models. The results of this study could be extended to actual application in the dose calculation/optimization, hence enhancing the gamma passing rate for patient-specific IMRT quality assurance.

Modeling and Optimization of Fractal Dimension in Wire Electrical Discharge Machining of EN 31 Steel Using the ANN-GA Approach.

To achieve enhanced surface characteristics in wire electrical discharge machining (WEDM), the present work reports the use of an artificial neural network (ANN) combined with a genetic algorithm (GA) for the correlation and optimization of WEDM process parameters. The parameters considered are the discharge current, voltage, pulse-on time, and pulse-off time, while the response is fractal dimension. The usefulness of fractal dimension to characterize a machined surface lies in the fact that it is independent of the resolution of the instrument or length scales. Experiments were carried out based on a rotatable central composite design. A feed-forward ANN architecture trained using the Levenberg-Marquardt (L-M) back-propagation algorithm has been used to model the complex relationship between WEDM process parameters and fractal dimension. After several trials, 4-3-3-1 neural network architecture has been found to predict the fractal dimension with reasonable accuracy, having an overall R-value of 0.97. Furthermore, the genetic algorithm (GA) has been used to predict the optimal combination of machining parameters to achieve a higher fractal dimension. The predicted optimal condition is seen to be in close agreement with experimental results. Scanning electron micrography of the machined surface reveals that the combined ANN-GA method can significantly improve the surface texture produced from WEDM by reducing the formation of re-solidified globules.

A neural network approach for the blind deconvolution of turbulent flows

We present a single-layer feed-forward artificial neural network architecture trained through a supervised learning approach for the deconvolution of flow variables from their coarse-grained computations such as those encountered in large eddy simulations. We stress that the deconvolution procedure proposed in this investigation is blind, i.e. the deconvolved field is computed without any pre-existing information about the filtering procedure or kernel. This may be conceptually contrasted to the celebrated approximate deconvolution approaches where a filter shape is predefined for an iterative deconvolution process. We demonstrate that the proposed blind deconvolution network performs exceptionally well in the a priori testing of two-dimensional Kraichnan, three-dimensional Kolmogorov and compressible stratified turbulence test cases, and shows promise in forming the backbone of a physics-augmented data-driven closure for the Navier–Stokes equations.

Neural network adaptive real-time optimizing control of industrial processes

<p>Real-time optimization (RTO) has attracted considerable interest among researchers and industries for being able to optimise the plant economics such as product efficiency, product quality and process safety in the wake of increasing global competitions. The success of RTO depends much on the quality of model being used in the optimisation. The present study was carried out to explore the use of artificial neural network (ANN) to improve the quality of the model being used in the modified two step (MTS) technique. The MTS is a real-time optimising control algorithm of the modifier adaptation scheme which is used to determine the optimum steady-state control set-points. The proposed new version of MTS technique will be using process model based on ANN. A laboratory scale process of a two continuous stirred tank heat exchanger in series (2CSTHEs) is used as a case study. The multilayer feed forward ANN architecture 4-10-6 with linear function was used to model the 2CSTHEs and then integrates into the MTS technique, the resulted algorithm will be known as Iterative Neural Network Modified Two Step (INNMTS) technique. Simulation studies were conducted to test the performance of the INNMTS technique on the 2CSTHEs process. The results show that the overall value for the coefficient of determination(R<sup>2</sup>)is equal to one, which indicates adequacy of the model proposed for the prediction of the behavior of 2CSTHEs system. When NN model of 2CSTHEs is applied to the INNMTS technique, the model-plant mismatch is greatly reduced to almost zero, which indicates by significant reduction in the number of iterations to 5which requires by INNMTS compared to 16 iterations by the MTS technique to converge to optimal real solution.</p><p>Chemical Engineering Research Bulletin 19(2017) 129-138</p>

Application of Artificial Intelligence Approach in Modeling Surface Quality of Aerospace Alloys in WEDM Process

In this work, artificial neural network (ANN) model is developed for prediction of surface roughness (SR) in wire-cut electrical discharge machining (WEDM) of Inconel 825 aerospace alloy. Four process parameters viz., pulse on time (Ton), pulse off time (Toff), peak current (Ip) and servo voltage (SV) are investigated using Box Behnken experimental design.Parametric variation shows that improved SR can be obtained at low levels of Ton and SV. A multi-layer feed forward ANN architecture 4-16-1 working on gradient descent back propagation algorithm isfound optimum and is statistically validated by conducting hypothesis tests. The developed model predicted with average 6.38% error and model accuracy is recorded as 93.62%. ANOVA showed that Ton is the most significant factor affecting SR with 76.12% contribution; followed by SV and Toff respectively with 7.18% and 5.3% contributions. The predictive capability of the developed ANN model is found encouraging and can be effectively used for predicting SR in WEDM of Inconel 825.

A comparative investigation of non-linear activation functions in neural controllers for search-based game AI engineering

The creation of intelligent video game controllers has recently become one of the greatest challenges in game artificial intelligence research, and it is arguably one of the fastest-growing areas in game design and development. The learning process, a very important feature of intelligent methods, is the result of an intelligent game controller to determine and control the game objects behaviors' or actions autonomously. Our approach is to use a more efficient learning model in the form of artificial neural networks for training the controllers. We propose a Hill-Climbing Neural Network (HillClimbNet) that controls the movement of the Ms. Pac-man agent to travel around the maze, gobble all of the pills and escape from the ghosts in the maze. HillClimbNet combines the hill-climbing strategy with a simple, feed-forward artificial neural network architecture. The aim of this study is to analyze the performance of various activation functions for the purpose of generating neural-based controllers to play a video game. Each non-linear activation function is applied identically for all the nodes in the network, namely log-sigmoid, logarithmic, hyperbolic tangent-sigmoid and Gaussian. In general, the results shows an optimum configuration is achieved by using log-sigmoid, while Gaussian is the worst activation function.

Metaheuristics for the feedforward artificial neural network (ANN) architecture optimization problem

This article deals with evolutionary artificial neural network (ANN) and aims to propose a systematic and automated way to find out a proper network architecture. To this, we adapt four metaheuristics to resolve the problem posed by the pursuit of optimum feedforward ANN architecture and introduced a new criteria to measure the ANN performance based on combination of training and generalization error. Also, it is proposed a new method for estimating the computational complexity of the ANN architecture based on the number of neurons and epochs needed to train the network. We implemented this approach in software and tested it for the problem of identification and estimation of pollution sources and for three separate benchmark data sets from UCI repository. The results show the proposed computational approach gives better performance than a human specialist, while offering many advantages over similar approaches found in the literature.

A neural network based urban growth model of an Indian city

The aim of the study reported in this paper is to demonstrate that the subjectivity in urban growth modeling and the calibration time can be reduced by using objective techniques like Artificial neural network (ANN). As a case study, the ANN-based model was applied to simulate the urban growth of Saharanpur city in India. In the proposed model, remote sensing and GIS were used to generate site attributes, while ANN was used to reveal the relationships between urban growth potential and the site attributes. Once ANN learnt the relationship, it was then used to simulate the urban growth. Different feed forward ANN architectures were evaluated in this study and finally the most optimum ANN architecture was selected for future growth simulation.

Computational intelligence-based congestion prediction for a dynamic urban street network

This paper develops a hybrid model for single point short term traffic flow forecasting in an urban traffic network. The hybrid model consists of two main modules: a fuzzy input fuzzy output filter (FIFO-filter) and a multi-layer feed-forward artificial neural network architecture optimized using evolution strategies (MLFN-ES). The FIFO-filter performs the data clustering operation and provides a rough forecasted prediction value based on the input data to the MLFN-ES associated with each cluster for modeling the input–output relation to provide accurate short term forecast value. The performance of the proposed model is demonstrated by predicting the traffic flow for an intersection in the central business district (CBD) area of Singapore. The hybrid model proposed in this paper gave a mean absolute percentage error (MAPE) of 8.35% on weekdays and 9.73% on weekends for the test data. A comparison analysis shows improved performance of the proposed hybrid method in short term traffic prediction over popular approaches like ARIMA and artificial neural network based systems.

Evolutionary artificial neural networks for hydrological systems forecasting

The conventional ways of constructing artificial neural network (ANN) for a problem generally presume a specific architecture and do not automatically discover network modules appropriate for specific training data. Evolutionary algorithms are used to automatically adapt the network architecture and connection weights according to the problem environment without substantial human intervention. To improve on the drawbacks of the conventional optimal process, this study presents a novel evolutionary artificial neural network (EANN) for time series forecasting. The EANN has a hybrid procedure, including the genetic algorithm and the scaled conjugate gradient algorithm, where the feedforward ANN architecture and its connection weights of neurons are simultaneously identified and optimized. We first explored the performance of the proposed EANN for the Mackey–Glass chaotic time series. The performance of the different networks was evaluated. The excellent performance in forecasting of the chaotic series shows that the proposed algorithm concurrently possesses efficiency, effectiveness, and robustness. We further explored the applicability and reliability of the EANN in a real hydrological time series. Again, the results indicate the EANN can effectively and efficiently construct a viable forecast module for the 10-day reservoir inflow, and its accuracy is superior to that of the AR and ARMAX models.

Delamination analysis in high speed drilling of carbon fiber reinforced plastics (CFRP) using artificial neural network model

The carbon fiber reinforced plastics (CFRP) are highly promising materials for the applications in aeronautical and aerospace industries. The delamination is a major problem associated with the drilling fiber reinforced composite materials, which reduce the structural integrity of the material. The present work focuses on the analysis of delamination behavior as a function of drilling process parameters at the entrance of the CFRP plates. The delamination analysis in high speed drilling is performed by developing an artificial neural network (ANN) model with spindle speed, feed rate and point angle as the affecting parameters. A multilayer feed forward ANN architecture, trained using error-back propagation training algorithm (EBPTA) is employed for this purpose. Drilling experiments are conducted as per full factorial design using cemented carbide (grade K20) twist drills that serve as input–output patterns for ANN training. The ANN model so developed is validated by presenting training and new testing input patterns. The validated ANN model is then used to generate the direct and interaction effect plots to analyze the delamination behavior. The simulation results illustrate the effectiveness of the ANN models to analyze the effects of drilling process parameters on delamination factor. The analysis also demonstrates the advantage of employing higher speed in controlling the delamination during drilling.

Optimizing feedforward artificial neural network architecture

Despite the fact that feedforward artificial neural networks (ANNs) have been a hot topic of research for many years there still are certain issues regarding the development of an ANN model, resulting in a lack of absolute guarantee that the model will perform well for the problem at hand. The multitude of different approaches that have been adopted in order to deal with this problem have investigated all aspects of the ANN modelling procedure, from training data collection and pre/post-processing to elaborate training schemes and algorithms. Increased attention is especially directed to proposing a systematic way to establish an appropriate architecture in contrast to the current common practice that calls for a repetitive trial-and-error process, which is time-consuming and produces uncertain results. This paper proposes such a methodology for determining the best architecture and is based on the use of a genetic algorithm (GA) and the development of novel criteria that quantify an ANN's performance (both training and generalization) as well as its complexity. This approach is implemented in software and tested based on experimental data capturing workpiece elastic deflection in turning. The intention is to present simultaneously the approach's theoretical background and its practical application in real-life engineering problems. Results show that the approach performs better than a human expert, at the same time offering many advantages in comparison to similar approaches found in literature.

A fast virtual screening filter for cytochrome P450 3A4 inhibition liability of compound libraries

Current virtual screening applications focus not only on biological activity, but also on additional relevant properties of drug candidates, like absorption, distribution, metabolism, and excretion (ADME). In first-pass virtual screening, these prediction systems must be very fast because typically several millions of compounds must be processed. We have developed a linear PLS-based prediction system for binary classification of drug-drug interaction liability caused by cytochrome P450 3A4 inhibition. The system was trained using IC 5 0 values of 311 carefully selected molecules out of a raw data set containing 1152 compounds. It correctly predicts 95% of the training data and 90% of a semi-independent validation data set. The PLS model was calculated from 333 descriptors encoding a molecule. It outperforms an approach utilizing a three layered feed-forward artificial neural network architecture. The average calculation time required for a prediction is less than 0.3 seconds per molecule on a single microprocessor.

Predicting the orbital lifetime of tethered satellite systems

Two models of uncontrolled tethered systems—a lumped-mass model and an orbital element propagation model—are described. Feed-forward artificial neural network architecture and training are discussed. A network training algorithm and hybrid training approach are described and used to train two lifetime prediction networks: one for free tethers and one for upward-deployed tether-trailing satellites. The accuracy of network predictions is demonstrated.