AN AERODYNAMIC PERFORMANCE OPTIMIZATION METHOD APPLIED TO THE INTAKE GRILLE OF ACTIVE CLEARANCE CONTROL SYSTEM

The objectives of the research presented in this viewgraph presentation are to 1) Design a mechanical ACC system for HPT tip seal clearance management; 2) Design a test rig to evaluate ACC system concepts. We have focused our efforts on designing mechanical ACC systems that articulate the seal shroud via mechanical linkages connected to actuators that reside outside the extreme environment of the HPT. We opted for this style of design due to a lack of high temperature/low profile actuators that are presently available. We have also selected multiple hydraulic actuators for this first generation ACC system. Fuel-draulic actuators are already a well established technology.

In this paper, an improved aerodynamic performance optimization method for 3-D low Reynolds number (Re) rotor blade is proposed. A conventional optimization procedure of blade is usually divided into three parts, such as the parameterization method, the fitness value evaluation and the optimization algorithm. This work is mainly focused on the first two parts. The parametrization method, Camber-FFD, is presented based on the camber parametrization method and the free-form deformation algorithm (FFD). The shape of 3-D blade is parameterized by the incidence angles and the coordinates of the maximum camber points. The fitness value evaluation has been realized with the help of an adaptive topological back propagation multi-layer forward artificial neural network (BP-MLFANN). During the training of BP-MLFANN, the hybrid particle swarm optimization method combined with the modified very fast simulate annealing algorithm (HPSO-MVFSA) is adopted to determine the neural network topology adaptively. To verify the effectiveness of this aerodynamic optimization method, the aerodynamic performance of a 3-D low-Re blade, such as Blade D900, is optimized, and the results are compared and analyzed based on the experiments and simulations. It is proved that this aerodynamic optimization method is feasible.

<span lang="EN-US">The proposed study employed a method for identifying the main contents (category/class) that a Remote Sensing Image (RSI) belongs to, as well as the percentage contribution if the image comprises a significant number of different content types. Histogram based approach has been used to extract the pixel density distribution and its normalized form helps to make solution independent from image physical characteristics. A multilayer feedforward Artificial Neural Network (ANN) design has been used to address the classification problem. The architecture included an adaptive form of transfer function, whose slope characteristics changes along with weights as learning progresses. The approach of solution design is computation efficient because it doesn’t apply extensive pre-processing.</span>

In this study, an E-Nose system was realized for the anesthetic dose level prediction. For this purpose, sevoflurane anesthetic agent was measured using the E-Nose system implemented with sensor array of quartz crystal microbalances (QCM). In surgeries, anesthetic agents are given to the patients with carrier gases of oxygen (02) and nitrous oxide (N20). Frequency changes on QCM sensors to the eight sevoflurane anesthetic dose levels were recorded via RS-232 serial port. A multilayer feed forward artificial neural network (MLNN) structure was used to provide the relationship between the frequency change and the anesthetic dose level. The MLNNs were trained with the measured data using Levenberg-Marquardt algorithm. Then, the trained MLNNs were tested with random data. The results have showed that, acceptable anesthetic dose level predictions have been obtained successfully.

The research is aimed at studying the features of strategy formulation in Formula One auto racing, the factors that influence this process, and identifying ways to increase the effectiveness of the strategy through the use of artificial intelligence methods. During Formula One races, teams face a large number of challenges related to various aspects, in particular tire wear and degradation. Teams deal with them through the implementation of the strategy – determining the correct pit stop moment and choosing the appropriate type of tires. Based on the determination of factors influencing the rate of tire wear and degradation, an original feature space was formed. Analysis of the strategies of race winners in previous seasons showed that the maximum number of pit stops was three. Using the principal component analysis, a study of the original feature space was conducted, the least relevant features were identified and subsequently removed from the original data set. To solve the problem, a system was proposed and built, consisting of four modules, each of which is a multilayer feedforward artificial neural network. Using the inequality proposed by Widrow and the sample-based estimation of the Lipschitz constant, the minimum required number of neurons of the hidden layers for each neural network module was determined. During the training process, their number was specified to achieve acceptable prediction results. AdaMax was used as an optimization algorithm, and the Huber loss function was chosen to calculate the error of the networks output. The mean squared error of the resulting system prediction on the test set was 0.1. The use of such system will reduce the decision-making time of teams when formulating a racing strategy, which in turn will contribute to achieving higher results in races.

Statistical modeling of nuclear data provides a novel approach to nuclear systematics complementary to established theoretical and phenomenological approaches based on quantum theory. Continuing previous studies in which global statistical modeling is pursued within the general framework of machine learning theory, we implement advances in training algorithms designed to improved generalization, in application to the problem of reproducing and predicting the halflives of nuclear ground states that decay 100% by the beta^- mode. More specifically, fully-connected, multilayer feedforward artificial neural network models are developed using the Levenberg-Marquardt optimization algorithm together with Bayesian regularization and cross-validation. The predictive performance of models emerging from extensive computer experiments is compared with that of traditional microscopic and phenomenological models as well as with the performance of other learning systems, including earlier neural network models as well as the support vector machines recently applied to the same problem. In discussing the results, emphasis is placed on predictions for nuclei that are far from the stability line, and especially those involved in the r-process nucleosynthesis. It is found that the new statistical models can match or even surpass the predictive performance of conventional models for beta-decay systematics and accordingly should provide a valuable additional tool for exploring the expanding nuclear landscape.

This paper presents a method for incorporating a priori information about an uncertain nonlinear system into the structure of a multilayer feedforward artificial neural network. Known information is incorporated into the activation function of the network output layer. An algorithm is derived for backpropagating the error and updating adjustable parameters within this layer that is consistent with existing supervised learning techniques. The developed technique is applied to the identification of a dynamic system and compared with conventional feedforward artificial neural network identifier. Results exhibit an improvement in the quality of the identification model and an increase in the rate of convergence. As a practical application, a prior information is utilized for identification of switched reluctance motor characteristics on the basis of experimental measurements. The results further demonstrate that artificial neural networks employing a priori information converge faster, require fewer adjustable weights, and more accurately predict the system of interest.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

In order to create convincing doctored images, forged images are exposed to some linear transformations (like rotation and resizing) which involve a resampling step. In copy–paste image forgery, the pasted portion is rescaled in order to hide traces of malicious tampering. In this paper, an algorithm is proposed to detect the global resizing operation of the doctored image blindly based on features extracted using subsampling. Fisher criterion is employed in order to choose the relevant features and reduce the dimensionality of the statistical features. Support vector machine (SVM) and multi-layer feedforward artificial neural network (ANN) are used for classification. Experimental results using $$C_b$$ , $$C_r$$ and grayscale images demonstrate that the proposed method has good rescaling detection performance even when dealing with distortions like JPEG compression. The results indicate that SVM performs better compared to ANN classifier.

This paper aims to perform modeling of Taiwanese farm and ecotourism data using compositional models as a probabilistic approach and to compare its results with the performance of an artificial neural network approach. Authors use probabilistic compositional models together with the artificial neural network as a classifier and compare the accuracy of both approaches. The probabilistic model structure is learned using hill climbing algorithm, and the weights of multilayer feedforward artificial neural network are learned using an R implementation of H2O library for deep learning. In case of both approaches, we employ a non-exhaustive cross-validation method and compare the models. The comparison is augmented by the structure of the compositional model and basic characterization of artificial neural network. As expected, the compositional models show significant advantages in interpretability of results and (probabilistic) relations between variables, whereas the artificial neural network provides more accurate yet “black-box” model.

The purpose of this study is to develop new multilayer feed-forward artificial neural network (ANN)-based maximal oxygen uptake (VO2max) prediction models by using submaximal treadmill exercise and nonexercise data. Using 10-fold cross validation on the dataset, standard error of estimate (SEE) and multiple correlation coefficient (R) of the models are calculated. It is shown that the models including submaximal, standard nonexercise and questionnaire variables yield higher R and lower SEE than the ones including submaximal and standard nonexercise variables only. The results of ANN-based models are also compared with the ones obtained by Multiple Linear Regression (MLR) and Support Vector Machines (SVM). It is shown that ANN-based models perform better than MLR and SVM-based models for predicting VO2max.

Modulation type is one of the most important characteristics used in signal waveform identification for wireless communications. In this paper, a cepstral algorithm for Automatic Digital Modulation Recognition (ADMR) is proposed. This algorithm uses Mel-Frequency Cepstral Coefficients (MFCCs) to extract the features of the modulated signal and a multi-layer feed-forward Artificial Neural Network (ANN) to classify the modulation type and its order. The proposed algorithm is capable of recognizing the modulation scheme with high accuracy in the presence of Additive White Gaussian Noise (AWGN) over a wide Signal-to-Noise Ratio (SNR) range.

An approach based on artificial neural networks is proposed for the analysis of electromechanical modes in Taiwan power system. To evaluate the dynamic performance of a power system in system operation and planning, the dominant eigenvalues for the worst-damped electromechanical mode must be computed. A multilayer feedforward artificial neural network is developed. It is well known that eigenvalues are complicated functions of many system variables such as bus loads, line flows, generation schedule, bus voltages etc. An important procedure in neural network design is to select those features which most affect system eigenvalues. A clustering artificial neural network is thus designed for feature selection. To demonstrate the effectiveness of the approach, results from eigenvalue analyses of Taiwan power system are reported.

Baiu season is the most important precipitation period to the climate of east China. It is often raining day after day with mixed type of precipitation over large area and lasting almost one month. The estimation of 3hr mean areal precipitation over SHIGUANHE catchment during Baiu season by use of GMS IR technique is present here. For getting better estimation of heavy rain, a multilayer feed-forward Artificial Neural Network (ANN) is introduced by training sample pairs of TRMM PR radar and GMS image on raining points and simulating the network in a relative small area nearby and short time span. Parts of raining points estimated by GMS IR method are modified by the output of the network according to data quality. Case studies over one or three consecutive swaths of PR show that the standard deviations of rainfall estimation increase after heavy rain modification while the correlation coefficients between rainfalls estimated and measured by ground based radar keep almost the same before and after data correction.

&lt;div class="section abstract"&gt;&lt;div class="htmlview paragraph"&gt;Heat treated cast aluminum components like engine blocks and cylinder heads can develop significant amount of residual stress and distortion particularly with water quench. To incorporate the influence of residual stress and distortion in cast aluminum product design, a rapid simulation approach has been developed based on artificial neural network and component geometry characteristics. Multilayer feed-forward artificial neural network (ANN) models were trained and verified using FEA residual stress and distortion predictions together with part geometry information such as curvature, maximum dihedral angle, topologic features including node's neighbors, as well as quench parameters like quench temperature and quench media.&lt;/div&gt;&lt;/div&gt;

Material classification through neural networks