Backpropagation-type Neural Network Research Articles

Gear contact fatigue life prediction based on transfer learning

The gear contact fatigue test is characterised by extraordinarily high costs and a long period of testing. A transfer learning algorithm was applied to develop a gear contact fatigue life prediction strategy to fully use other low-cost fatigue test data, such as twin-disc test data. The twin-disc contact fatigue test data were incorporated into the neural network model to explore the combined effects of surface integrity and stress level on fatigue life, which was further transferred for gear contact fatigue life prediction. The results indicate that this method enables an effective gear fatigue life prediction under minor gear sample conditions. With a small sample amount of 7, the prediction error in the fatigue life of the model with transfer learning is 46.33%, while the typical back-propagation neural network (BPNN) model without transfer learning is 101.66%. To control life prediction error to within 50%, the model without transfer learning requires at least 17 gear samples, while the transfer model calls for only 7 gear samples. The proposed data transfer learning framework can be used to predict the gear contact fatigue S-N curve under minor sample conditions and is straightforwardly applied to extended applications.

Effectiveness of predicting tunneling-induced ground settlements using machine learning methods with small datasets

Prediction of tunneling-induced ground settlements is an essential task, particularly for tunneling in urban settings. Ground settlements should be limited within a tolerable threshold to avoid damages to aboveground structures. Machine learning (ML) methods are becoming popular in many fields, including tunneling and underground excavations, as a powerful learning and predicting technique. However, the available datasets collected from a tunneling project are usually small from the perspective of applying ML methods. Can ML algorithms effectively predict tunneling-induced ground settlements when the available datasets are small? In this study, seven ML methods are utilized to predict tunneling-induced ground settlement using 14 contributing factors measured before or during tunnel excavation. These methods include multiple linear regression (MLR), decision tree (DT), random forest (RF), gradient boosting (GB), support vector regression (SVR), back-propagation neural network (BPNN), and permutation importance-based BPNN (PI-BPNN) models. All methods except BPNN and PI-BPNN are shallow-structure ML methods. The effectiveness of these seven ML approaches on small datasets is evaluated using model accuracy and stability. The model accuracy is measured by the coefficient of determination ( R 2 ) of training and testing datasets, and the stability of a learning algorithm indicates robust predictive performance. Also, the quantile error (QE) criterion is introduced to assess model predictive performance considering underpredictions and overpredictions. Our study reveals that the RF algorithm outperforms all the other models with the highest model prediction accuracy (0.9) and stability (3.02 × 10 −27 ). Deep-structure ML models do not perform well for small datasets with relatively low model accuracy (0.59) and stability (5.76). The PI-BPNN architecture is proposed and designed for small datasets, showing better performance than typical BPNN. Six important contributing factors of ground settlements are identified, including tunnel depth, the distance between tunnel face and surface monitoring points (DTM), weighted average soil compressibility modulus (ACM), grouting pressure, penetrating rate and thrust force.

Reference evapotranspiration prediction using neural network method

Water is the most vital resource for life on earth, At present we know that irrigation systems have currently acquired great importance due to the scarcity that is affecting worldwide, since there is no awareness about this important resource, however, for years we have worked to try to solve this problem. The objective of this research work was to develop a Feedforward Backpropagation type neural network algorithm with three layers: in the input layer include the operating factors such as the maximum temperature (°C), the minimum temperature (°C), the average temperature (°C) and solar radiation (mm / day) and the hidden layer three neurons and yet in the output layer only one neuron, this algorithm has been trained by the Levenberg-Marquardt algorithm to predict the evapotranspiration. The results were satisfactory because the algorithm was able to predict the reference evapotranspiration with a correlation coefficient of 99.99% and with an error of 0.0001. Therefore, this technique can be considered to automate the online irrigation system by monitoring plant transpiration and soil evaporation.

Diagnosis of Defective Rotor Bars in Induction Motors

This paper proposes a diagnosis method, combining signal analysis and classification models, to the rotor defect problems of motors. Two manufacture technologies, nonmagnetic high-temperature resistant ceramic adhesive and electrical discharge machining (EDM), are applied to make testing samples, including blowhole and perforation defects of rotor bars in this study. The typical multiresolution analysis (MRA) model is used to analyze acquired source current signals of motors. The features are extracted from the signals of each column of MRA-matrix, including maximum, mean, standard deviation, root-mean-square, and summation. The typical back-propagation neural network (BPNN) model is used to diagnose the rotor bar defects of motors, and then the various signal-to-noise ratio (SNR) of white Gaussian noise (WGN), 30, 25, and 20 dB, are added to the signals to verify the robustness of the proposed method. The results show the availability of the proposed method to diagnose the rotor bar defects of motors.

Study on Waste Type Identification Method Based on Bird Flock Neural Network

According to the waste type identification requirement in waste classification, a waste type identification method based on a bird flock neural network (BFNN) was proposed. The problem of obtaining the feature dataset of waste images was considered, and color histogram and texture feature extraction techniques were used. The local optimum problem of a typical backpropagation neural network (BPNN) was considered, and a bird flock optimization (BFO) algorithm was proposed. The accuracy problem of the typical BPNN was considered, and a new online weight adjustment method of neurons was proposed. The number of hidden layer neurons (nodes) of the typical BPNN was considered, and an online adjustment method was proposed. The experimental results show that the recyclables (paper, plastic, glass, and cloth) and nonrecyclables can effectively be identified by the waste type identification method based on the BFNN, and the recognition accuracy is 81% which meets actual needs.

Hybrid fuzzy neural network versus backpropagation neural network: An application to predict the Ibex-35 index stock

The use of neural networks has been extended in all areas of knowledge due to the good results being obtained in the resolution of the different problems posed. The prediction of prices in general, and stock market prices in particular, represents one of the main objectives of the use of neural networks in finance. This paper presents the analysis of the efficiency of the hybrid fuzzy neural network against a backpropagation type neural network in the price prediction of the Spanish stock exchange index (IBEX-35). The paper is divided into two parts. In the first part, the main characteristics of neural networks such as hybrid fuzzy and backpropagation, their structures and learning rules are presented. In the second part, the prediction of the IBEX-35 stock exchange index with these networks is analyzed, measuring the efficiency of both as a function of the prediction errors committed. For this purpose, both networks have been constructed with the same inputs and for the same sample period. The results obtained suggest that the Hybrid fuzzy neuronal network is much more efficient than the widespread backpropagation neuronal network for the sample analysed.

Solving polynomial systems using a fast adaptive back propagation-type neural network algorithm

This paper proposes a neural network architecture for solving systems of non-linear equations. A back propagation algorithm is applied to solve the problem, using an adaptive learning rate procedure, based on the minimization of the mean squared error function defined by the system, as well as the network activation function, which can be linear or non-linear. The results obtained are compared with some of the standard global optimization techniques that are used for solving non-linear equations systems. The method was tested with some well-known and difficult applications (such as Gauss–Legendre 2-point formula for numerical integration, chemical equilibrium application, kinematic application, neuropsychology application, combustion application and interval arithmetic benchmark) in order to evaluate the performance of the new approach. Empirical results reveal that the proposed method is characterized by fast convergence and is able to deal with high-dimensional equations systems.

Discrimination of neutrons and γ-rays in liquid scintillator based on Elman neural network**Supported by National Natural Science Foundation of China (11275134,11475117)

In this work, a new neutron and γ (n/γ) discrimination method based on an Elman Neural Network (ENN) is proposed to improve the discrimination performance of liquid scintillator (LS) detectors. Neutron and γ data were acquired from an EJ-335 LS detector, which was exposed in a 241Am-9Be radiation field. Neutron and γ events were discriminated using two methods of artificial neural network including the ENN and a typical Back Propagation Neural Network (BPNN) as a control. The results show that the two methods have different n/γ discrimination performances. Compared to the BPNN, the ENN provides an improved of Figure of Merit (FOM) in n/γ discrimination. The FOM increases from 0.907 ± 0.034 to 0.953 ± 0.037 by using the new method of the ENN. The proposed n/γ discrimination method based on ENN provides a new choice of pulse shape discrimination in neutron detection.

Comparison of Back Propagation Neural Network and Genetic Algorithm Neural Network for Stream Flow Prediction

Comparison of stream flow prediction models has been presented. Stream flow prediction model was developed using typical back propagation neural network (BPNN) and genetic algorithm coupled with neural network (GANN). The study uses daily data from Nethravathi River basin (Karnataka, India). The study demonstrates the prediction ability of GANN. The statistical tests show that GANN model performs much better when compared to BPNN model.

Neural Networks Based Modeling of Viscosity for Facilitating Transportation of Magnetite Orewater Slurry

Transportation of ores from mines to distant processing plants can be done in different ways. One method, which is found to be suitable in certain situations, is to carry magnetite through a pipeline in the form of solid-liquid slurry. For smooth flow of slurry, maintaining appropriate slurry viscosity along with flow pressure and flow velocity is essential. The Neural Networks (NN) algorithm can be used to find out the condition to guarantee smooth transportation process of magnetite-water slurry. In the present work, experiments are conducted to determine the effect of solids concentration, particle size and additive on the apparent viscosity of magnetite ore water slurry. The back propagation type Neural Networks (NN), constructed based on experimental results, is utilized to estimate apparent, or dynamic, viscosity of magnetite-water slurry under different parametric conditions. The NN model outputs show good prediction of apparent viscosity of the slurry, and hence, the applicability of the system adopted.

Selection of optimum cutting condition of cobalt-based superalloy with GONNS

Machining of new superalloys is challenging. Automated software environments for determining the optimal cutting conditions after reviewing a set of experimental results are very beneficial to obtain the desired surface quality and to use the machine tools effectively. The genetically optimized neural network system (GONNS) is proposed for the selection of optimal cutting conditions from the experimental data with minimal operator involvement. Genetic algorithm (GA) obtains the optimal operational condition by using the neural networks. A feed-forward backpropagation-type neural network was trained to represent the relationship between surface roughness, cutting force, and machining parameters of face-milling operation. Training data were collected at the symmetric and asymmetric milling operations by using different cutting speeds (V c), feed rates (f), and depth of cuts (a p) without using coolant. The surface roughness (Raasymt, Rasymt) and cutting force (Fxasymt, Fyasymt, Fzasymt, Fxsymt, Fysymt, Fzsymt) were measured for each cutting condition. The surface roughness estimation accuracy of the neural network was better for the asymmetric milling operation with 0.4% and 5% for training and testing data, respectively. For the symmetric milling operations, slightly higher estimation errors were observed around 0.5% and 7% for the training and testing. One parameter was optimized by using the GONNS while all the other parameters, including the cutting forces and the surface roughness, were kept in the desired range.

Prediction of impact sensitivity of nitro energetic compounds by neural network based on electrotopological-state indices

A quantitative structure–property relationship (QSPR) model was constructed to predict the impact sensitivity of 156 nitro energetic compounds by means of artificial neural network (ANN). Electrotopological-state indices (ETSI) were used as molecular structure descriptors which combined together both electronic and topological characteristics of the analyzed molecules. The typical back-propagation neural network (BPNN) was employed for fitting the possible non-linear relationship existed between the ETSI and impact sensitivity. The dataset of 156 nitro compounds was randomly divided into a training set (64), a validation set (63) and a prediction set (29). The optimal condition of the neural network was obtained by adjusting various parameters by trial-and-error. Simulated with the final optimum BP neural network [16–12–1], the results show that most of the predicted impact sensitivity values are in good agreement with the experimental data, which are superior to those obtained by multiple linear regression (MLR) and partial least squares (PLS). The model proposed can be used not only to reveal the quantitative relation between impact sensitivity and molecular structures of nitro energetic compounds, but also to predict the impact sensitivity of nitro compounds for engineering.

Radial Basis Function Neural Network for Modeling Rating Curves

The establishment of a rating curve is an important problem in hydrology. Generally, a regression approach is applied to establish the relationship between stage and discharge. However, this approach fails in the cases where hysteresis is present in the data. The aim of the study is to investigate the potential of employing radial basis function (RBF) type neural networks for modeling stage-discharge relationships at gauging stations and to compare different types of networks. The results are promising and suggest that the neural network approach is highly viable. A comparison of the RBF models with backpropagation type neural networks reveals that the former is superior in performance for rating curves exhibiting hysteresis.

Acid deposition in the eastern United States and neural network predictions for the future

Back-propagation type neural networks were trained on total sulphur dioxide emissions from power plants and measured field data on precipitation chemistry. These trained networks were then able to predict seasonal changes in sulphate, hydrogen, nitrate, and ammonium ion concentrations caused by projected decreases in sulphur dioxide emissions from power plants in the eastern United States. Results showed that by 2010 the proposed reductions in sulphur dioxide emissions by the U.S. electric power utilities would just be sufficient to reduce acid rain conditions to the levels where human health problems are avoided. However, pollution from acid rain would still be impacting considerable regions of the north-eastern United States and south-eastern Canada causing other environmental damage such as loss of fish in acidic lakes. Key words: acid rain, desulphurization, modelling, neural networks, pollution.

オンライン学習機能を持つ視覚追跡システムの提案

In order to keep visual tracking systems with color segmentation technique running in real environment, it should be developed on-line learning method to update models for adapting them to dynamic changes of surroundings.To deal with this problem, we propose an on-line visual learning method for color image segmentation and object tracking in dynamic environment. Our method utilizes Fuzzy ART architecture which is a kind of neural network for competitive learning. The mechanism of this neural network is suitable for on-line learning and different from that of backpropagation type neural network. In order to use Fuzzy ART architecture for color segmentation on-line, we transform the color signal that the framegrabber used yields to a particular color space called Yrθ space. To show validity of our method, we present some results of experiments using sequences of real images.

Feasibility of neural networks in modelling radio propagation for field strength prediction

A typical back-propagation neural network (BPN) model is developed for modelling radio propagation for field strength prediction based on data measurements of propagation loss (in decibels) with terrain information taken in an urban area (Athens region) in the 900 MHz band. The feasibility of the BPN model is checked against the performance of a conventional semiempirical reference model. The performance of both models is quantified by statistical methods. The evaluation is done by comparing their prediction error statistics of average absolute, standard deviation and root mean square and by comparing their percentage accuracy and correlation of predicted values relative to true data measurements. © 1998 John Wiley & Sons, Ltd.

Neuro-fuzzy reasoning for occluded object recognition

To tackle the problem of occluded object recognition first we give a new interpretation to the multidimensional fuzzy reasoning and then realize that new interpretation through backpropagation-type neural network. At the learning stage of the neural network, fuzzy linguistic statements are used. Once learned, the nonfuzzy features of an occluded object can be classified. At the time of classification of the nonfuzzy features of an occluded object we use the concept of fuzzy singleton. An effective approach to recognize an unknown scene which consists of a set of occluded objects is to detect a number of significant (local) features on the boundary of the unknown scene. Thus the major problems fall into the selection of the appropriate set of features (local) for representing the object in the training stage, as well as in the detection of these features in the recognition process. The features should be invariant to scale, orientation and minor distortions in boundary shape. The performance of the proposed scheme is tested through several examples.

Neural Network Modeling of Anisotropic Aggregate Behavior from Repeated Load Triaxial Tests

Determining horizontal specimen response in a repeated load triaxial test is essential to properly characterize the directional dependency of unbound aggregate resilient behavior under anisotropic loading conditions. Recent research has applied artificial neural networks (ANNs) for predicting, in the absence of lateral deformation data, the anisotropic stiffness properties of granular materials from standard AASHTO tests. Feed-forward backpropagation-type neural networks were successfully trained with two triaxial stresses (confining pressure and applied deviator stress), measured vertical deformation, and two aggregate properties (compacted dry density and crushed particle percentage) used as input variables. The output variables were the horizontal and shear moduli for which the actual (target) values were derived and computed from test results. The ANN models predicted the two moduli, with mean errors of less than 3 percent compared with those computed by using experimental stresses and strains. Both the applied stress state and the aggregate properties were found to affect the generalization and thus the prediction ability of the ANN models.

Two-dimensional shape recognition using oriented-polar representation

To deal with such a problem as object recognition of position, scale, and rotation invariance (PSRI), we utilize some PSRI properties of images obtained from objects, for example, the centroid of the image. The corresponding position of the centroid to the boundary of the image is invariant is spite of rotation, scale, and translation of the image. To obtain the information of the image, we use the technique similar to Radon transform, called the oriented-polar representation of a 2-D image. In this representation, two specific points, the centroid and the weighted mean point, are selected to form an initial ray, then the image is sampled with N angularly equispaced rays departing from the initial rays. Each ray contains a number of intersections and the distance information obtained from the centroid to the intersections. The shape recognition algorithm is based on the least total error of these two items of information. Together with a simple noise removal and a typical backpropagation neural network, this algorithm is simple, but the PSRI is achieved with a high recognition rate.

Neuro Fuzzy Approach to Pattern Recognition

To tackle the pattern recognition problems first we give a new interpretation to the multi-dimensional fuzzy reasoning and then realize that new interpretation through a backpropagation type neural network. At the learning stage of the neural network fuzzy linguistic statements have been used. Once learned, the nonfuzzy features of a pattern can be classified. At the time of classification of the nonfuzzy features of a pattern we use the concept of fuzzy singleton. The performance of the proposed scheme is tested through synthetic data. Finally, we use the proposed scheme for the vowel recognition problem of three Indian languages. Copyright © 1996 Elsevier Science Ltd.