Abstract: Since the presentation of the backpropagation algorithm, a vast variety of improvements of the technique for training a feed forward neural networks have been proposed. This article focuses on two classes of acceleration techniques, one is known as Local Adaptive Techniques that are based on weightspecific only, such as the temporal behavior of the partial derivative of the current weight. The other, known as Dynamic Adaptation Methods, which dynamically adapts the momentum factors, α, and learning rate, η, with respect to the iteration number or gradient. Some of most popular learning algorithms are described. These techniques have been implemented and tested on several problems and measured in terms of gradient and error function evaluation, and percentage of success. Numerical evidence shows that these techniques improve the convergence of the Backpropagation algorithm.
Abstract: This paper presents an approach for the determination of the optimal cutting parameters (spindle speed, feed rate, depth of cut and engagement) leading to minimum surface roughness in face milling of high silicon stainless steel by coupling neural network (NN) and Electromagnetism-like Algorithm (EM). In this regard, the advantages of statistical experimental design technique, experimental measurements, artificial neural network, and Electromagnetism-like optimization method are exploited in an integrated manner. To this end, numerous experiments on this stainless steel were conducted to obtain surface roughness values. A predictive model for surface roughness is created by using a back propogation neural network, then the optimization problem was solved by using EM optimization. Additional experiments were performed to validate optimum surface roughness value predicted by EM algorithm. It is clearly seen that a good agreement is observed between the predicted values by EM coupled with feed forward neural network and experimental measurements. The obtained results show that the EM algorithm coupled with back propogation neural network is an efficient and accurate method in approaching the global minimum of surface roughness in face milling.
Abstract: Basic ingredients of concrete are cement, fine aggregate, coarse aggregate and water. To produce a concrete of certain specific properties, optimum proportion of these ingredients are mixed. The important factors which govern the mix design are grade of concrete, type of cement and size, shape and grading of aggregates. Concrete mix design method is based on experimentally evolved empirical relationship between the factors in the choice of mix design. Basic draw backs of this method are that it does not produce desired strength, calculations are cumbersome and a number of tables are to be referred for arriving at trial mix proportion moreover, the variation in attainment of desired strength is uncertain below the target strength and may even fail. To solve this problem, a lot of cubes of standard grades were prepared and attained 28 days strength determined for different combination of cement, fine aggregate, coarse aggregate and water. An artificial neural network (ANN) was prepared using these data. The input of ANN were grade of concrete, type of cement, size, shape and grading of aggregates and output were proportions of various ingredients. With the help of these inputs and outputs, ANN was trained using feed forward back proportion model. Finally trained ANN was validated, it was seen that it gave the result with/ error of maximum 4 to 5%. Hence, specific type of concrete can be prepared from given material properties and proportions of these materials can be quickly evaluated using the proposed ANN.
Abstract: Hand gesture is one of the typical methods used in
sign language for non-verbal communication. It is most commonly
used by people who have hearing or speech problems to
communicate among themselves or with normal people. Various sign
language systems have been developed by manufacturers around the
globe but they are neither flexible nor cost-effective for the end
users. This paper presents a system prototype that is able to
automatically recognize sign language to help normal people to
communicate more effectively with the hearing or speech impaired
people. The Sign to Voice system prototype, S2V, was developed
using Feed Forward Neural Network for two-sequence signs
detection. Different sets of universal hand gestures were captured
from video camera and utilized to train the neural network for
classification purpose. The experimental results have shown that
neural network has achieved satisfactory result for sign-to-voice
translation.
Abstract: A concern that researchers usually face in different
applications of Artificial Neural Network (ANN) is determination of
the size of effective domain in time series. In this paper, trial and
error method was used on groundwater depth time series to determine
the size of effective domain in the series in an observation well in
Union County, New Jersey, U.S. different domains of 20, 40, 60, 80,
100, and 120 preceding day were examined and the 80 days was
considered as effective length of the domain. Data sets in different
domains were fed to a Feed Forward Back Propagation ANN with
one hidden layer and the groundwater depths were forecasted. Root
Mean Square Error (RMSE) and the correlation factor (R2) of
estimated and observed groundwater depths for all domains were
determined. In general, groundwater depth forecast improved, as
evidenced by lower RMSEs and higher R2s, when the domain length
increased from 20 to 120. However, 80 days was selected as the
effective domain because the improvement was less than 1% beyond
that. Forecasted ground water depths utilizing measured daily data
(set #1) and data averaged over the effective domain (set #2) were
compared. It was postulated that more accurate nature of measured
daily data was the reason for a better forecast with lower RMSE
(0.1027 m compared to 0.255 m) in set #1. However, the size of input
data in this set was 80 times the size of input data in set #2; a factor
that may increase the computational effort unpredictably. It was
concluded that 80 daily data may be successfully utilized to lower the
size of input data sets considerably, while maintaining the effective
information in the data set.
Abstract: This paper analyzes the patterns of the Monte Carlo
data for a large number of variables and minterms, in order to
characterize the circuit path length behavior. We propose models
that are determined by training process of shortest path length
derived from a wide range of binary decision diagram (BDD)
simulations. The creation of the model was done use of feed forward
neural network (NN) modeling methodology. Experimental results
for ISCAS benchmark circuits show an RMS error of 0.102 for the
shortest path length complexity estimation predicted by the NN
model (NNM). Use of such a model can help reduce the time
complexity of very large scale integrated (VLSI) circuitries and
related computer-aided design (CAD) tools that use BDDs.
Abstract: The ElectroEncephaloGram (EEG) is useful for
clinical diagnosis and biomedical research. EEG signals often
contain strong ElectroOculoGram (EOG) artifacts produced
by eye movements and eye blinks especially in EEG recorded
from frontal channels. These artifacts obscure the underlying
brain activity, making its visual or automated inspection
difficult. The goal of ocular artifact removal is to remove
ocular artifacts from the recorded EEG, leaving the underlying
background signals due to brain activity. In recent times,
Independent Component Analysis (ICA) algorithms have
demonstrated superior potential in obtaining the least
dependent source components. In this paper, the independent
components are obtained by using the JADE algorithm (best
separating algorithm) and are classified into either artifact
component or neural component. Neural Network is used for
the classification of the obtained independent components.
Neural Network requires input features that exactly represent
the true character of the input signals so that the neural
network could classify the signals based on those key
characters that differentiate between various signals. In this
work, Auto Regressive (AR) coefficients are used as the input
features for classification. Two neural network approaches
are used to learn classification rules from EEG data. First, a
Polynomial Neural Network (PNN) trained by GMDH (Group
Method of Data Handling) algorithm is used and secondly,
feed-forward neural network classifier trained by a standard
back-propagation algorithm is used for classification and the
results show that JADE-FNN performs better than JADEPNN.
Abstract: This paper reports work done to improve the modeling of complex processes when only small experimental data sets are available. Neural networks are used to capture the nonlinear underlying phenomena contained in the data set and to partly eliminate the burden of having to specify completely the structure of the model. Two different types of neural networks were used for the application of pulping problem. A three layer feed forward neural networks, using the Preconditioned Conjugate Gradient (PCG) methods were used in this investigation. Preconditioning is a method to improve convergence by lowering the condition number and increasing the eigenvalues clustering. The idea is to solve the modified odified problem M-1 Ax= M-1b where M is a positive-definite preconditioner that is closely related to A. We mainly focused on Preconditioned Conjugate Gradient- based training methods which originated from optimization theory, namely Preconditioned Conjugate Gradient with Fletcher-Reeves Update (PCGF), Preconditioned Conjugate Gradient with Polak-Ribiere Update (PCGP) and Preconditioned Conjugate Gradient with Powell-Beale Restarts (PCGB). The behavior of the PCG methods in the simulations proved to be robust against phenomenon such as oscillations due to large step size.
Abstract: End milling process is one of the common metal
cutting operations used for machining parts in manufacturing
industry. It is usually performed at the final stage in manufacturing a
product and surface roughness of the produced job plays an
important role. In general, the surface roughness affects wear
resistance, ductility, tensile, fatigue strength, etc., for machined parts
and cannot be neglected in design. In the present work an
experimental investigation of end milling of aluminium alloy with
carbide tool is carried out and the effect of different cutting
parameters on the response are studied with three-dimensional
surface plots. An artificial neural network (ANN) is used to establish
the relationship between the surface roughness and the input cutting
parameters (i.e., spindle speed, feed, and depth of cut). The Matlab
ANN toolbox works on feed forward back propagation algorithm is
used for modeling purpose. 3-12-1 network structure having
minimum average prediction error found as best network architecture
for predicting surface roughness value. The network predicts surface
roughness for unseen data and found that the result/prediction is
better. For desired surface finish of the component to be produced
there are many different combination of cutting parameters are
available. The optimum cutting parameter for obtaining desired
surface finish, to maximize tool life is predicted. The methodology is
demonstrated, number of problems are solved and algorithm is coded
in Matlab®.
Abstract: This paper presents an approach which is based on the
use of supervised feed forward neural network, namely multilayer
perceptron (MLP) neural network and finite element method (FEM)
to solve the inverse problem of parameters identification. The
approach is used to identify unknown parameters of ferromagnetic
materials. The methodology used in this study consists in the
simulation of a large number of parameters in a material under test,
using the finite element method (FEM). Both variations in relative
magnetic permeability and electrical conductivity of the material
under test are considered. Then, the obtained results are used to
generate a set of vectors for the training of MLP neural network.
Finally, the obtained neural network is used to evaluate a group of
new materials, simulated by the FEM, but not belonging to the
original dataset. Noisy data, added to the probe measurements is used
to enhance the robustness of the method. The reached results
demonstrate the efficiency of the proposed approach, and encourage
future works on this subject.
Abstract: Physiological control of a left ventricle assist device (LVAD) is generally a complicated task due to diverse operating environments and patient variability. In this work, a tracking control algorithm based on sliding mode and feed forward control for a class of discrete-time single input single output (SISO) nonlinear uncertain systems is presented. The controller was developed to track the reference trajectory to a set operating point without inducing suction in the ventricle. The controller regulates the estimated mean pulsatile flow Qp and mean pulsatility index of pump rotational speed PIω that was generated from a model of the assist device. We recall the principle of the sliding mode control theory then we combine the feed-forward control design with the sliding mode control technique to follow the reference trajectory. The uncertainty is replaced by its upper and lower boundary. The controller was tested in a computer simulation covering two scenarios (preload and ventricular contractility). The simulation results prove the effectiveness and the robustness of the proposed controller
Abstract: Although lots of research work has been done for
human pose recognition, the view-point of cameras is still critical
problem of overall recognition system. In this paper, view-point
insensitive human pose recognition is proposed. The aims of the
proposed system are view-point insensitivity and real-time processing.
Recognition system consists of feature extraction module, neural
network and real-time feed forward calculation. First, histogram-based
method is used to extract feature from silhouette image and it is
suitable for represent the shape of human pose. To reduce the
dimension of feature vector, Principle Component Analysis(PCA) is
used. Second, real-time processing is implemented by using Compute
Unified Device Architecture(CUDA) and this architecture improves
the speed of feed-forward calculation of neural network. We
demonstrate the effectiveness of our approach with experiments on
real environment.
Abstract: Training neural networks to capture an intrinsic
property of a large volume of high dimensional data is a difficult
task, as the training process is computationally expensive. Input
attributes should be carefully selected to keep the dimensionality of
input vectors relatively small.
Technical indexes commonly used for stock market prediction
using neural networks are investigated to determine its effectiveness
as inputs. The feed forward neural network of Levenberg-Marquardt
algorithm is applied to perform one step ahead forecasting of
NASDAQ and Dow stock prices.
Abstract: This paper reports work done to improve the modeling of complex processes when only small experimental data sets are available. Neural networks are used to capture the nonlinear underlying phenomena contained in the data set and to partly eliminate the burden of having to specify completely the structure of the model. Two different types of neural networks were used for the application of Pulping of Sugar Maple problem. A three layer feed forward neural networks, using the Preconditioned Conjugate Gradient (PCG) methods were used in this investigation. Preconditioning is a method to improve convergence by lowering the condition number and increasing the eigenvalues clustering. The idea is to solve the modified problem where M is a positive-definite preconditioner that is closely related to A. We mainly focused on Preconditioned Conjugate Gradient- based training methods which originated from optimization theory, namely Preconditioned Conjugate Gradient with Fletcher-Reeves Update (PCGF), Preconditioned Conjugate Gradient with Polak-Ribiere Update (PCGP) and Preconditioned Conjugate Gradient with Powell-Beale Restarts (PCGB). The behavior of the PCG methods in the simulations proved to be robust against phenomenon such as oscillations due to large step size.
Abstract: This paper uses the radial basis function neural
network (RBFNN) for system identification of nonlinear systems.
Five nonlinear systems are used to examine the activity of RBFNN in
system modeling of nonlinear systems; the five nonlinear systems are
dual tank system, single tank system, DC motor system, and two
academic models. The feed forward method is considered in this
work for modelling the non-linear dynamic models, where the KMeans
clustering algorithm used in this paper to select the centers of
radial basis function network, because it is reliable, offers fast
convergence and can handle large data sets. The least mean square
method is used to adjust the weights to the output layer, and
Euclidean distance method used to measure the width of the Gaussian
function.
Abstract: In this paper, novel techniques in increasing the accuracy
and speed of convergence of a Feed forward Back propagation
Artificial Neural Network (FFBPNN) with polynomial activation
function reported in literature is presented. These technique was
subsequently used to determine the coefficients of Autoregressive
Moving Average (ARMA) and Autoregressive (AR) system. The
results obtained by introducing sequential and batch method of weight
initialization, batch method of weight and coefficient update, adaptive
momentum and learning rate technique gives more accurate result
and significant reduction in convergence time when compared t the
traditional method of back propagation algorithm, thereby making
FFBPNN an appropriate technique for online ARMA coefficient
determination.
Abstract: Vehicle which are turning or maneuvering at high speeds
are susceptible to sliding and subsequently deviate from desired path. In
this paper the dynamics governing the Yaw/Roll behavior of a vehicle
has been simulated. Two different simulations have been used one for
the real vehicle, for which a fuzzy controller is designed to increase its
directional stability property. The other simulation is for a hypothetical
vehicle with much higher tire cornering stiffness which is capable of
developing the required lateral forces at the tire-ground patch contact to
attain the desired lateral acceleration for the vehicle to follow the
desired path without slippage. This simulation model is our reference
model.
The logic for keeping the vehicle on the desired track in the cornering
or maneuvering state is to have some braking forces on the inner or
outer tires based on the direction of vehicle deviation from the desired
path. The inputs to our vehicle simulation model is steer angle δ and
vehicle velocity V , and the outputs can be any kinematical parameters
like yaw rate, yaw acceleration, side slip angle, rate of side slip angle
and so on. The proposed fuzzy controller is a feed forward controller.
This controller has two inputs which are steer angle δ and vehicle
velocity V, and the output of the controller is the correcting moment M,
which guides the vehicle back to the desired track. To develop the
membership functions for the controller inputs and output and the fuzzy
rules, the vehicle simulation has been run for 1000 times and the
correcting moment have been determined by trial and error. Results of
the vehicle simulation with fuzzy controller are very promising
and show the vehicle performance is enhanced greatly over the
vehicle without the controller. In fact the vehicle performance
with the controller is very near the performance of the reference
ideal model.
Abstract: The rate of production of main products of the Fischer-Tropsch reactions over Fe/HZSM5 bifunctional catalyst in a fixed bed reactor is investigated at a broad range of temperature, pressure, space velocity, H2/CO feed molar ratio and CO2, CH4 and water flow rates. Model discrimination and parameter estimation were performed according to the integral method of kinetic analysis. Due to lack of mechanism development for Fisher – Tropsch Synthesis on bifunctional catalysts, 26 different models were tested and the best model is selected. Comprehensive one and two dimensional heterogeneous reactor models are developed to simulate the performance of fixed-bed Fischer – Tropsch reactors. To reduce computational time for optimization purposes, an Artificial Feed Forward Neural Network (AFFNN) has been used to describe intra particle mass and heat transfer diffusion in the catalyst pellet. It is seen that products' reaction rates have direct relation with H2 partial pressure and reverse relation with CO partial pressure. The results show that the hybrid model has good agreement with rigorous mechanistic model, favoring that the hybrid model is about 25-30 times faster.
Abstract: This paper presents Simulation and experimental
study aimed at investigating the effectiveness of an adaptive artificial
neural network stabilizer on enhancing the damping torque of a
synchronous generator. For this purpose, a power system comprising
a synchronous generator feeding a large power system through a
short tie line is considered. The proposed adaptive neuro-control
system consists of two multi-layered feed forward neural networks,
which work as a plant model identifier and a controller. It generates
supplementary control signals to be utilized by conventional
controllers. The details of the interfacing circuits, sensors and
transducers, which have been designed and built for use in tests, are
presented. The synchronous generator is tested to investigate the
effect of tuning a Power System Stabilizer (PSS) on its dynamic
stability. The obtained simulation and experimental results verify the
basic theoretical concepts.
Abstract: Nowadays, hard disk is one of the most popular storage components. In hard disk industry, the hard disk drive must pass various complex processes and tested systems. In each step, there are some failures. To reduce waste from these failures, we must find the root cause of those failures. Conventionall data analysis method is not effective enough to analyze the large capacity of data. In this paper, we proposed the Hough method for straight line detection that helps to detect straight line defect patterns that occurs in hard disk drive. The proposed method will help to increase more speed and accuracy in failure analysis.