Neural Network Learning Procedure Research Articles

Fault Identification and Classification of Asynchronous Motor Drive Using Optimization Approach with Improved Reliability

This article aims to provide a technique for identifying and categorizing interturn insulation problems in variable-speed motor drives by combining Salp Swarm Optimization (SSO) with Recurrent Neural Network (RNN). The goal of the proposed technique is to detect and classify Asynchronous Motor faults at their early stages, under both normal and abnormal operating conditions. The proposed technique uses a recurrent neural network in two phases to identify and label interturn insulation concerns, with the first phase being utilised to establish whether or not the motors are healthy. In the second step, it discovers and categorises potentially dangerous interturn errors. The SSO approach is used in the second phase of the recurrent neural network learning procedure, with the goal function of minimizing error in mind. The proposed CSSRN technique simplifies the system for detecting and categorizing the interturn insulation issue, resulting in increased system precision. In addition, the proposed model is implemented in the MATLAB/Simulink, where metrics such as accuracy, precision, recall, and specificity may be analysed. Similarly, existing methods such as Adaptive Neuro-Fuzzy Inference System (ANFIS), Recurrent Neural Network (RNN), and Salp Swarm Algorithm Artificial Neural Network (SSAANN) are utilised to evaluate metrics such as Root mean squared error (RMSE), Mean bias error (MBE), Mean absolute percentage error (MAPE), consumption, and execution time for comparative analysis.

Being Friends Instead of Adversaries: Deep Networks Learn from Data Simplified by Other Networks

Amongst a variety of approaches aimed at making the learning procedure of neural networks more effective, the scientific community developed strategies to order the examples according to their estimated complexity, to distil knowledge from larger networks, or to exploit the principles behind adversarial machine learning. A different idea has been recently proposed, named Friendly Training, which consists in altering the input data by adding an automatically estimated perturbation, with the goal of facilitating the learning process of a neural classifier. The transformation progressively fades-out as long as training proceeds, until it completely vanishes. In this work we revisit and extend this idea, introducing a radically different and novel approach inspired by the effectiveness of neural generators in the context of Adversarial Machine Learning. We propose an auxiliary multi-layer network that is responsible of altering the input data to make them easier to be handled by the classifier at the current stage of the training procedure. The auxiliary network is trained jointly with the neural classifier, thus intrinsically increasing the 'depth' of the classifier, and it is expected to spot general regularities in the data alteration process. The effect of the auxiliary network is progressively reduced up to the end of training, when it is fully dropped and the classifier is deployed for applications. We refer to this approach as Neural Friendly Training. An extended experimental procedure involving several datasets and different neural architectures shows that Neural Friendly Training overcomes the originally proposed Friendly Training technique, improving the generalization of the classifier, especially in the case of noisy data.

DESIGN AND IMPLEMENTATION OF A PORTABLE AUTOMATED AUDIOMETER FOR HEARING CLASSIFICATION USING MACHINE LEARNING APPROACHES

Audiometric tests can identify the hearing loss at specific frequencies using the audiogram. The aim and objectives of the study were (i) to develop an automated audiometer for self-diagnosing the hearing ability of the patient; (ii) to extract the features from the acoustic signals and to classify the normal and profound hearing loss patients using different machine learning algorithms; (iii) to validate the hearing loss classification using six-frequency average (6-FA) method based on simple linear regression analysis and machine learning algorithms. The study is conducted among 150 patients, including 75 patients with normal hearing ability and 75 patients with profound hearing loss. The total population of 150 underwent audiometric test both in the soundproof audiometric room and in the normal field environment. Based on the patient response, the intensity and frequency are changed automatically, and the audiogram is plotted by the principle of Artificial Neural Network learning procedures. The overall accuracy produced by classification of normal and profound hearing loss patients using Support Vector Machine (SVM), k-Nearest Neighbor classifier, and Naïve Bayes classifier is 97%, 96%, and 95%, respectively. The results indicated that the SVM classifier outperforms the other two classifiers well. The preliminary audiometric test can be performed remotely and then consulted with an audiologist. Thus, the patient could operate the developed prototype independently and get a consultation from trained medical personnel.

Nearly Optimal Tunable MPC Strategies on Embedded Platforms

We present an embeddable optimization-free application of a near-optimal MPC implementation with continuous tuning capabilities. We propose a strategy combining the advantages of explicit model predictive control with tunable properties that is implementable on embedded platforms with limited memory and computational resources. We consider a neural network (NN) learning procedure to mimic the control actions of an MPC strategy. While acknowledging limited guarantees on the constraints satisfaction with just the NN-based controller, we introduce an optimization-based corrector of the mimicked control action. Such a corrector then steers the control authority of the mimicked controller such that constraints on manipulated and process variables are enforced. To demonstrate the efficacy of the proposed control strategy, a case study implemented on an embedded platform is shown.

An Incremental Learning Framework for Human-Like Redundancy Optimization of Anthropomorphic Manipulators

Recently, the human-like behavior on the anthropomorphic robot manipulator is increasingly accomplished by the kinematic model establishing the relationship of an anthropomorphic manipulator and human arm motions. Notably, the growth and broad availability of advanced data science techniques facilitate the imitation learning process in anthropomorphic robotics. However, the enormous dataset causes the labeling and prediction burden. In this article, the swivel motion reconstruction approach was applied to imitate human-like behavior using the kinematic mapping in robot redundancy. For the sake of efficient computing, a novel incremental learning framework that combines an incremental learning approach with a deep convolutional neural network is proposed for fast and efficient learning. The algorithm exploits a novel approach to detect changes from human motion data streaming and then evolve its hierarchical representation of features. The incremental learning process can fine-tune the deep network only when model drifts detection mechanisms are triggered. Finally, we experimentally demonstrated this neural network's learning procedure and translated the trained human-like model to manage the redundancy optimization control of an anthropomorphic robot manipulator (LWR4+, KUKA, Germany). This approach can hold the anthropomorphic kinematic structure-based redundant robots. The experimental results showed that our architecture could not only enhance the regression accuracy but also significantly reduce the processing time of learning human motion data.

RETRACTED: An efficient grasshopper optimization with recurrent neural network controller-based induction motor to replace flywheel of the process machine

This paper proposes a convenient power electronic circuitry with a control approach for the flywheel replacement of an induction motor. The proposed control approach is the joined execution of grasshopper optimization algorithm and recurrent neural network based on duty ratio controller and hence the proposed work is named grasshopper optimization with recurrent neural network. The main contribution of this work is, the power electronic circuitry gets the input voltage samples and limits the deviation to appraise the instantaneous torque demand. The required voltage for the instantaneous torque demand is produced by the proposed control technique. In the proposed grasshopper optimization with recurrent neural network technique, the grasshopper optimization algorithm is a meta-heuristic population-based algorithm, which works from the perspective of the swarming behavior of grasshoppers in nature. In the proposed system, the recurrent neural network learning procedure is improved by the grasshopper optimization algorithm in the perspective of the minimum error objective function. the proposed grasshopper optimization with recurrent neural network technique optimizes the inverter switching states by limiting the error between the setpoint torque and the demand torque regarding objective function. With this proposed technique, the unbalance between demand torque and generated torque is found with high precision and the quicker execution to pull back out the torsional pulsation insensitive load linked transmission systems. By utilizing the proposed methodology, the extreme fluctuation of load torque due to peaky loads in an induction motor will be detected accurately. Also, the proposed technique reduces the torsional vibrations, weakness in components and minimizes the outages of uninterrupted production leading to higher profits. The proposed strategy is actualized in the MATLAB/Simulink platform and evaluated their performance. The performances are appeared differently compared with the existing methods.

Pre-Synaptic Pool Modification (PSPM): A supervised learning procedure for recurrent spiking neural networks.

Learning synaptic weights of spiking neural network (SNN) models that can reproduce target spike trains from provided neural firing data is a central problem in computational neuroscience and spike-based computing. The discovery of the optimal weight values can be posed as a supervised learning task wherein the weights of the model network are chosen to maximize the similarity between the target spike trains and the model outputs. It is still largely unknown whether optimizing spike train similarity of highly recurrent SNNs produces weight matrices similar to those of the ground truth model. To this end, we propose flexible heuristic supervised learning rules, termed Pre-Synaptic Pool Modification (PSPM), that rely on stochastic weight updates in order to produce spikes within a short window of the desired times and eliminate spikes outside of this window. PSPM improves spike train similarity for all-to-all SNNs and makes no assumption about the post-synaptic potential of the neurons or the structure of the network since no gradients are required. We test whether optimizing for spike train similarity entails the discovery of accurate weights and explore the relative contributions of local and homeostatic weight updates. Although PSPM improves similarity between spike trains, the learned weights often differ from the weights of the ground truth model, implying that connectome inference from spike data may require additional constraints on connectivity statistics. We also find that spike train similarity is sensitive to local updates, but other measures of network activity such as avalanche distributions, can be learned through synaptic homeostasis.

Solidification analysis by non-equilibrium phase field model using thermodynamics data estimated by machine learning

A multi-phase field method (MPFM) using the finite interface dissipation model is applied to simulate the solidification microstructure evolution of a stainless-steel composition, including the delta-ferrite to gamma austenite peritectic transformation. The calculation is performed for a quinary system of engineering steel in a two-dimensional field. Thermodynamics calculations using the CALPHAD database in this MPFM are replaced by machine learning prediction to reduce the numerical time. Neural network methodology is introduced for machine learning in this study. The Gibbs free energy and chemical potential values estimated from the CALPHAD database coupling results are inputted into the neural network learning procedure, together with the composition and temperature values. The microstructure evaluated using the obtained neural network parameter is in good agreement with that directly coupled with the CALPHAD database. This calculation is approximately five times faster than direct CALPHAD calculation.

A genetic fuzzy radial basis function neural network for structural health monitoring of composite laminated beams

In this paper, a new neural network learning procedure, called genetic fuzzy hybrid learning algorithm (GFHLA) is proposed for training the radial basis function neural network (RBFNN). The method combines the genetic algorithm and fuzzy logic to optimize the centers and widths of the RBFNN, and the linear least-squared method is used to adjust the neural network connection weights. The modal frequencies of a glass/epoxy laminates beam with varying assumed delamination sizes and locations were computed using finite element method and fed into the genetic fuzzy RBF neural network to predict the delamination location and its extent. The simulation demonstrates that the neural network based on GFHLA is robust and promising.

On Adaptive Learning Rate That Guarantees Convergence in Feedforward Networks

This paper investigates new learning algorithms (LF I and LF II) based on Lyapunov function for the training of feedforward neural networks. It is observed that such algorithms have interesting parallel with the popular backpropagation (BP) algorithm where the fixed learning rate is replaced by an adaptive learning rate computed using convergence theorem based on Lyapunov stability theory. LF II, a modified version of LF I, has been introduced with an aim to avoid local minima. This modification also helps in improving the convergence speed in some cases. Conditions for achieving global minimum for these kind of algorithms have been studied in detail. The performances of the proposed algorithms are compared with BP algorithm and extended Kalman filtering (EKF) on three bench-mark function approximation problems: XOR, 3-bit parity, and 8-3 encoder. The comparisons are made in terms of number of learning iterations and computational time required for convergence. It is found that the proposed algorithms (LF I and II) are much faster in convergence than other two algorithms to attain same accuracy. Finally, the comparison is made on a complex two-dimensional (2-D) Gabor function and effect of adaptive learning rate for faster convergence is verified. In a nutshell, the investigations made in this paper help us better understand the learning procedure of feedforward neural networks in terms of adaptive learning rate, convergence speed, and local minima.

Nitriding parameters analized by neural network and genetic algorithm

The surface hardness and hardness profile of a nitrided workpiece depend on the chemical composition of the steel, nitriding temperature and time, and on type of the nitriding process (i.e. atmosphere). An issue in this approach was to test how the statistical analysis, artificial neural network, genetic algorithm and genetic programming may be used for determination of nitriding time and surface hardness, in case when the chemical composition of steel, nitriding temperature and required thickness of nitrided layer are known. In the neural network learning procedure datasets of results were used, after nitriding 5 different steel grades. Different combinations of time, temperature, surface hardness and thickness of plasma and gas nitriding layer are compiled from the experiments and industrial experience and also from the literature. The static multi-layer feed-forward neural network is proposed. To accelerate the convergence of the proposed static error-back propagation learning algorithm, the momentum method is applied. The mean error between experimental data of nitriding time and data predicted using a neural network, and also the standard deviation for both the learning and the testing dataset is found to be small and acceptable. The determination of time by genetic algorithm gives greater standard deviation than by using neural network. Determining surface hardness after nitriding by the use of neural network gives less reliable results due to relatively imprecise input data and a narrow learning dataset. Due to nitriding data insufficiency, the network was tested only with the learning dataset.

Comparative analysis of spatial and seasonal variability: Austrian precipitation during the 20th century

AbstractThe purpose of this investigation is to demonstrate the usability of objective methods to study the variability of precipitation and hence to contribute to a better understanding of spatial and seasonal variability of Austria's precipitation climate during the 20th century.This will be achieved by regionalizing the intra‐annual variability of seasonal precipitation distributions during three non‐overlapping 33 year samples (1901–33, 1934–66, 1967–99). Monthly precipitation totals were extracted at 31 Austrian stations from a homogenized long‐term climate dataset provided by the Austrian weather service. Three statistical techniques, namely cluster analysis (CLA), rotated empirical orthogonal functions (REOFs) and an unsupervised learning procedure of artificial neural networks (ANNs), were utilized to find homogeneous precipitation regions.The results of summer (June, July, August (JJA)) and winter (December, January, February (DJF)) seasons are presented. The resulting homogeneous precipitation regions depend on season, period and method in this order. Hence, differences introduced by using different methods are small compared with those inferred by investigating different episodes and especially with those related to the seasons.During winter, three homogeneous precipitation regions are found, independent from the period considered. These regions can be assigned to different airflows dominating Austria's climate and triggering precipitation events during the cold season. The situation during summer is more complicated. Thus, at least four clusters are necessary to record the circumstances, which are caused by spatially inhomogeneous convective events such as thunderstorms. Copyright © 2003 Royal Meteorological Society

Effects of Learning Parameters on Learning Procedure and Performance of a BPNN

We examined the effects of changing learning parameters on the learning procedure and performance of back-propagation neural networks used to pick seismic arrivals. The results show that such change mainly affects the speed of convergence of the learning procedures, and does not affect the BPNN structure and its overall performance. A relationship between the learning parameters and iteration number is obtained. This relationship may be used as a guide to check the convergence of the learning procedure and the BPNN performance. We also use a weight map of BPNN structure to analyze its interior and performance. Two BPNNs used to pick seismic arrivals from three-component and single-component seismograms have similar weight patterns and operate in a similar way, although they have different structures and trained by different training dataset.

Unsupervised neural network learning procedures for feature extraction and classification

In this article, we review unsupervised neural network learning procedures which can be applied to the task of preprocessing raw data to extract useful features for subsequent classification. The learning algorithms reviewed here are grouped into three sections: information-preserving methods, density estimation methods, and feature extraction methods. Each of these major sections concludes with a discussion of successful applications of the methods to real-world problems.

MODELLING UNKNOWN STRUCTURAL SYSTEMS THROUGH THE USE OF NEURAL NETWORKS

This paper explores the potential of using neural networks to identify the internal forces of typical systems encountered in the field of earthquake engineering and structural dynamics. After formulating the identification task as a neural network learning procedure, the method is applied to a representative chain-like system under deterministic and stochastic excitations. The neural network based identification method provides very good results for general classes of multi-degree-of-freedom structural systems. The range of validity of the approach is demonstrated, and some application issues are discussed for (a) partially known multi-degree-of-freedom systems and (b) completely unknown systems.

The Target Switch Algorithm: A Constructive Learning Procedure for Feed-Forward Neural Networks

We propose an efficient procedure for constructing and training a feed-forward neural network. The network can perform binary classification for binary or analogue input data. We show that the procedure can also be used to construct feedforward neural networks with binary-valued weights. Neural networks with binary-valued weights are potentially straightforward to implement using microelectronic or optical devices and they can also exhibit good generalization.

Novel “On-Line” Identification Procedure using Artificial Neural Network

A novel neural network learning procedure which has faster training features suitable to real-time implementation is proposed. Then, improved structure backpropagation and radial basis function networks by using the proposed learning scheme are developed. Simulations are performed to confirm the feasibility of these neural networks for real time applications. Lastly, these are experimentally applied to machine tool breakage detection problems resulting in excellent fault diagnosis methods.

Identification of the internal forces of structural systems using feedforward multilayer networks

This paper explores the potential of using parallel distributed processing (neural network) approaches to identify the internal forces of typical systems encountered in the field of structural dynamics. Such problems arise in the context of structural control and the “health monitoring” of dynamic systems. After formulating the identification task as a neural network learning procedure, the method is applied to a representative example problem under a variety of excitations. The range of validity of the approach is demonstrated, and some application issues are discussed.

Neural Network Approaches Versus Statistical Methods In Classification Of Multisource Remote Sensing Data

Neural network learning procedures and statistical classificaiton methods are applied and compared empirically in classification of multisource remote sensing and geographic data. Statistical multisource classification by means of a method based on Bayesian classification theory is also investigated and modified. The modifications permit control of the influence of the data sources involved in the classification process. Reliability measures are introduced to rank the quality of the data sources. The data sources are then weighted according to these rankings in the statistical multisource classification. Four data sources are used in experiments: Landsat MSS data and three forms of topographic data (elevation, slope, and aspect). Experimental results show that two different approaches have unique advantages and disadvantages in this classification application.

Learning algorithms for neural networks with the Kalman filters

Based on various approaches, several different learing algorithms have been given in the literature for neural networks. Almost all algorithms have constant learning rates or constant accelerative parameters, though they have been shown to be effective for some practical applications. The learning procedure of neural networks can be regarded as a problem of estimating (or identifying) constant parameters (i.e. connection weights of network) with a nonlinear or linear observation equation. Making use of the Kalman filtering, we derive a new back-propagation algorithm whose learning rate is computed by a time-varying Riccati difference equation. Perceptron-like and correlational learning algorithms are also obtained as special cases. Furthermore, a self-organising algorithm of feature maps is constructed within a similar framework.