Rockburst is a common engineering geological hazard. In order to evaluate rockburst liability in kimberlite at an underground diamond mine, a method combining generalized regression neural networks (GRNN) and fruit fly optimization algorithm (FOA) is employed. Based on two fundamental premises of rockburst occurrence, depth, σθ, σc, σt, B1, B2, SCF, Wet are determined as indicators of rockburst, which are also input vectors of GRNN model. 132 groups of data obtained from rockburst cases from all over the world are chosen as training samples to train the GRNN model; FOA is used to seek the optimal parameter σ that generates the most accurate GRNN model. The trained GRNN model is adopted to evaluate burst liability in kimberlite pipes. The same eight rockburst indicators are acquired from lab tests, mine site and FEM model as test sample features. Evaluation results made by GRNN can be confirmed by a rockburst case at this mine. GRNN do not require any prior knowledge about the nature of the relationship between the input and output variables and avoid analyzing the mechanism of rockburst, which has a bright prospect for engineering rockburst potential evaluation.
The petrographic composition of coal has a significant impact on its technological and sorption properties. That composition is most frequently determined by means of microscope quantitative analyses. Thus, aside from the purely scientific aspect, such measurements have an important practical application in the industrial usage of coal, as well as in issues related to the safety in underground mining facilities. The article discusses research aiming at analyzing the usefulness of selected parameters of a digital image description in the process of automatic identification of macerals of the inertinite group using neural networks. The description of the investigated images was based on statistical parameters determined on the basis of a histogram and co-occurrence matrix (Haralick parameters). Each of the studied macerals was described by means of a 20-element feature vector. An analysis of its principal components (PCA) was conducted, along with establishing the relationship between the number of the applied components and the effectiveness of the MLP network. Based on that, the optimum number of input variables for the investigated classification task was chosen, which resulted in reduction of the size of the network’s hidden layer. As part of the discussed research, the authors also analyzed the process of classification of macerals of the inertinite group using an algorithm based on a group of MLP networks, where each network possessed one output. As a result, average recognition effectiveness of 80.9% was obtained for a single MLP network, and of 93.6% for a group of neural networks. The obtained results indicate that it is possible to use the proposed methodology as a tool supporting microscopic analyses of coal.
Speech enhancement is fundamental for various real time speech applications and it is a challenging task in the case of a single channel because practically only one data channel is available. We have proposed a supervised single channel speech enhancement algorithm in this paper based on a deep neural network (DNN) and less aggressive Wiener filtering as additional DNN layer. During the training stage the network learns and predicts the magnitude spectrums of the clean and noise signals from input noisy speech acoustic features. Relative spectral transform-perceptual linear prediction (RASTA-PLP) is used in the proposed method to extract the acoustic features at the frame level. Autoregressive moving average (ARMA) filter is applied to smooth the temporal curves of extracted features. The trained network predicts the coefficients to construct a ratio mask based on mean square error (MSE) objective cost function. The less aggressive Wiener filter is placed as an additional layer on the top of a DNN to produce an enhanced magnitude spectrum. Finally, the noisy speech phase is used to reconstruct the enhanced speech. The experimental results demonstrate that the proposed DNN framework with less aggressive Wiener filtering outperforms the competing speech enhancement methods in terms of the speech quality and intelligibility.
In this paper, a new Multi-Layer Perceptron Neural Network (MLP NN) classifier is proposed for classifying sonar targets and non-targets from the acoustic backscattered signals. Besides the capabilities of MLP NNs, it uses Back Propagation (BP) and Gradient Descent (GD) for training; therefore, MLP NNs face with not only impertinent classification accuracy but also getting stuck in local minima as well as lowconvergence speed. To lift defections, this study uses Adaptive Best Mass Gravitational Search Algorithm (ABGSA) to train MLP NN. This algorithm develops marginal disadvantage of the GSA using the bestcollected masses within iterations and expediting exploitation phase. To test the proposed classifier, this algorithm along with the GSA, GD, GA, PSO and compound method (PSOGSA) via three datasets in various dimensions will be assessed. Assessed metrics include convergence speed, fail probability in local minimum and classification accuracy. Finally, as a practical application assumed network classifies sonar dataset. This dataset consists of the backscattered echoes from six different objects: four targets and two non-targets. Results indicate that the new classifier proposes better output in terms of aforementioned criteria than whole proposed benchmarks.
The aim of the studywas to find an effective method of ripple torque compensation for a direct drive with a permanent magnet synchronous motor (PMSM) without time-consuming drive identification. The main objective of the research on the development of a methodology for the proper teaching a neural network was achieved by the use of iterative learning control (ILC), correct estimation of torque and spline interpolation. The paper presents the structure of the drive system and the method of its tuning in order to reduce the torque ripple, which has a significant effect on the uneven speed of the servo drive. The proposed structure of the PMSM in the dq axis is equipped with a neural compensator. The introduced iterative learning control was based on the estimation of the ripple torque and spline interpolation. The structurewas analyzed and verified by simulation and experimental tests. The elaborated structure of the drive system and method of its tuning can be easily used by applying a microprocessor system available now on the market. The proposed control solution can be made without time-consuming drive identification, which can have a great practical advantage. The article presents a new approach to proper neural network training in cooperation with iterative learning for repetitive motion systems without time-consuming identification of the motor.
The paper presents the operation of two neuro-fuzzy systems of an adaptive type, intended for solving problems of the approximation of multi-variable functions in the domain of real numbers. Neuro-fuzzy systems being a combination of the methodology of artiﬁcial neural networks and fuzzy sets operate on the basis of a set of fuzzy rules “if-then”, generated by means of the self-organization of data grouping and the estimation of relations between fuzzy experiment results. The article includes a description of neuro-fuzzy systems by Takaga-Sugeno-Kang (TSK) and Wang-Mendel (WM), and in order to complement the problem in question, a hierarchical structural self-organizing method of teaching a fuzzy network. A multi-layer structure of the systems is a structure analogous to the structure of “classic” neural networks. In its ﬁnal part the article presents selected areas of application of neuro-fuzzy systems in the ﬁeld of geodesy and surveying engineering. Numerical examples showing how the systems work concerned: the approximation of functions of several variables to be used as algorithms in the Geographic Information Systems (the approximation of a terrain model), the transformation of coordinates, and the prediction of a time series. The accuracy characteristics of the results obtained have been taken into consideration.
Land surveyors, photogrammetrists, remote sensing engineers and professionals in the Earth sciences are often faced with the task of transferring coordinates from one geodetic datum into another to serve their desired purpose. The essence is to create compatibility between data related to different geodetic reference frames for geospatial applications. Strictly speaking, conventional techniques of conformal, affine and projective transformation models are mostly used to accomplish such task. With developing countries like Ghana where there is no immediate plans to establish geocentric datum and still rely on the astro-geodetic datums as it national mapping reference surface, there is the urgent need to explore the suitability of other transformation methods. In this study, an effort has been made to explore the proficiency of the Extreme Learning Machine (ELM) as a novel alternative coordinate transformation method. The proposed ELM approach was applied to data found in the Ghana geodetic reference network. The ELM transformation result has been analysed and compared with benchmark methods of backpropagation neural network (BPNN), radial basis function neural network (RBFNN), two-dimensional (2D) affine and 2D conformal. The overall study results indicate that the ELM can produce comparable transformation results to the widely used BPNN and RBFNN, but better than the 2D affine and 2D conformal. The results produced by ELM has demonstrated it as a promising tool for coordinate transformation in Ghana.
Prior any satellite technology developments, the geodetic networks of a country were realized from a topocentric datum, and hence the respective cartography was performed. With availability of Global Navigation Satellite Systems-GNSS, cartography needs to be updated and referenced to a geocentric datum to be compatible with this technology. Cartography in Ecuador has been performed using the PSAD56 (Provisional South American Datum 1956) systems, nevertheless it’s necessary to have inside the system SIRGAS (SIstema de Referencia Geocéntrico para las AmericaS). This transformation between PSAD56 to SIRGAS use seven transformation parameters calculated with the method Helmert. These parameters, in case of Ecuador are compatible for scales of 1:25 000 or less, that does not satisfy the requirements on applications for major scales. In this study, the technique of neural networks is demonstrated as an alternative for improving the processing of UTM planes coordinates E, N (East, North) from PSAD56 to SIRGAS. Therefore, from the coordinates E, N, of the two systems, four transformation parameters were calculated (two of translation, one of rotation, and one scale difference) using the technique bidimensional transformation. Additionally, the same coordinates were used to training Multilayer Artificial Neural Network -MANN, in which the inputs are the coordinates E, N in PSAD56 and output are the coordinates E, N in SIRGAS. Both the two-dimensional transformation and ANN were used as control points to determine the differences between the mentioned methods. The results imply that, the coordinates transformation obtained with the artificial neural network multilayer trained have been improving the results that the bidimensional transformation, and compatible to scales 1:5000.
This study investigates the estimated adsorption efficiency of artificial Nickel (II) ions with perlite in an aqueous solution using artificial neural networks, based on 140 experimental data sets. Prediction using artificial neural networks is performed by enhancing the adsorption efficiency with the use of Nickel (II) ions, with the initial concentrations ranging from 0.1 mg/L to 10 mg/L, the adsorbent dosage ranging from 0.1 mg to 2 mg, and the varying time of effect ranging from 5 to 30 mins. This study presents an artificial neural network that predicts the adsorption efficiency of Nickel (II) ions with perlite. The best algorithm is determined as a quasi-Newton back-propagation algorithm. The performance of the artificial neural network is determined by coefficient determination (R2), and its architecture is 3-12-1. The prediction shows that there is an outstanding relationship between the experimental data and the predicted values.
Noise control has gained a lot of attention recently. However, presence of nonlinearities in signal paths for some applications can cause significant difficulties in the operation of control algorithms. In particular, this problem is common in structural noise control, which uses a piezoelectric shunt circuit. Not only vibrating structures may exhibit nonlinear characteristics, but also piezoelectric actuators. In this paper, active device casing is addressed. The objective is to minimize the noise coming out of the casing, by controlling vibration of its walls. The shunt technology is applied. The proposed control algorithm is based on algorithms from a group of soft computing. It is verified by means of simulations using data acquired from a real object.
The article presents results of the influence of the GMDH (Group Method of Data Handling) neural network input data preparation method on the results of predicting corrections for the Polish timescale UTC(PL). Prediction of corrections was carried out using two methods, time series analysis and regression. As appropriate to these methods, the input data was prepared based on two time series, ts1 and ts2. The implemented research concerned the designation of the prediction errors on certain days of the forecast and the influence of the quantity of data on the prediction error. The obtained results indicate that in the case of the GMDH neural network the best quality of forecasting for UTC(PL) can be obtained using the time-series analysis method. The prediction errors obtained did not exceed the value of ± 8 ns, which confirms the possibility of maintaining the Polish timescale at a high level of compliance with the UTC.
The aim of this paper is to answer the question: Are the Łódź Hills useful for electrical energy production from wind energy or not? Due to access to short-term data related to wind measurements (the period of 2008 and 2009) from a local meteorological station, the measure – correlate – predict approach have been applied. Long-term (1979‒2016) reference data were obtained from ECWMF ERA-40 Reanalysis. Artificial neural networks were used to calculate predicted wind speed. The obtained average wind speed and wind power density was 4.21 ms⁻¹ and 70 Wm⁻¹, respectively, at 10 m above ground level (5.51 ms⁻¹, 170 Wm⁻¹ at 50 m). From the point of view of Polish wind conditions, Łódź Hills may be considered useful for wind power engineering.
The article shows a new model of Continuous Cooling Transformation (CCT) diagrams of structural steels and engineering steels. The modelling used artificial neural networks and a set of experimental data prepared based on 550 CCT diagrams published in the literature. The model of CCT diagrams forms 17 artificial neural networks which solve classification and regression tasks. Neural model is implemented in a computer software that enables calculation of a CCT diagram based on chemical composition of steel and its austenitizing temperature.
The three-dimensional (3D) coordinate measurement of radio frequency identification (RFID) multi-tag networks is one of the important issues in the field of RFID, which affects the reading performance of RFID multi-tag networks. In this paper, a novel method for 3D coordinate measurement of RFID multitag networks is proposed. A dual-CCD system (vertical and horizontal cameras) is used to obtain images of RFID multi-tag networks from different angles. The iterative threshold segmentation and the morphological filtering method are used to process the images. The template matching method is respectively used to determine the two-dimensional (2D) coordinate and the vertical coordinate of each tag. After that, the 3D coordinate of each tag is obtained. Finally, a back-propagation (BP) neural network is used to model the nonlinear relationship between the RFID multi-tag network and the corresponding reading distance. The BP neural network can predict the reading distances of unknown tag groups and find out the optimal distribution structure of the tag groups corresponding to the maximum reading distance. In the future work, the corresponding in-depth research on the neural network to adjust the distribution of tags will be done.
Skin cancer is the most common form of cancer affecting humans. Melanoma is the most dangerous type of skin cancer; and early diagnosis is extremely vital in curing the disease. So far, the human knowledge in this field is very limited, thus, developing a mechanism capable of identifying the disease early on can save lives, reduce intervention and cut unnecessary costs. In this paper, the researchers developed a new learning technique to classify skin lesions, with the purpose of observing and identifying the presence of melanoma. This new technique is based on a convolutional neural network solution with multiple configurations; where the researchers employed an International Skin Imaging Collaboration (ISIC) dataset. Optimal results are achieved through a convolutional neural network composed of 14 layers. This proposed system can successfully and reliably predict the correct classification of dermoscopic lesions with 97.78% accuracy.
In this paper, the application of the Artificial Neural Network (ANN) algorithm has been used for testing selected specification parameters of voltage-controlled oscillator. Today, mixed electronic circuits specification time is an issue. An analog part of Phase Locked Loopis a voltage-controlled oscillator, which is very sensitive to variation of the technology process. Fault model for the integrated circuit voltage control oscillator (VCO) in ring topology is introduced and the before test stage classificatory is designed. In order to reduce testing time and keep the specification accuracy (approximation) on the high level, an artificial neural network has been applied. The features selection process and output coding for specification parameters are described. A number of different ANN have been designed and then compared with real specification of the VCO. The results obtained gives response in short time with high enough accuracy.
In order to enhance the acoustical performance of a traditional straight-path automobile muffler, a multi-chamber muffler having reverse paths is presented. Here, the muffler is composed of two internally parallel/extended tubes and one internally extended outlet. In addition, to prevent noise transmission from the muffler’s casing, the muffler’s shell is also lined with sound absorbing material. Because the geometry of an automotive muffler is complicated, using an analytic method to predict a muffler’s acoustical performance is difficult; therefore, COMSOL, a finite element analysis software, is adopted to estimate the automotive muffler’s sound transmission loss. However, optimizing the shape of a complicated muffler using an optimizer linked to the Finite Element Method (FEM) is time-consuming. Therefore, in order to facilitate the muffler’s optimization, a simplified mathematical model used as an objective function (or fitness function) during the optimization process is presented. Here, the objective function can be established by using Artificial Neural Networks (ANNs) in conjunction with the muffler’s design parameters and related TLs (simulated by FEM). With this, the muffler’s optimization can proceed by linking the objective function to an optimizer, a Genetic Algorithm (GA). Consequently, the discharged muffler which is optimally shaped will improve the automotive exhaust noise.
The void fraction is one of the most important parameters characterizing a multiphase flow. The prediction of the performance of any system operating with more than single phase relies on our knowledge and ability to measure the void fraction. In this work, a validated simulation study was performed in order to predict the void fraction independent of the flow pattern in gas-liquid two-phase flows using a gamma ray 60Co source and just one scintillation detector with the help of an artificial neural network (ANN) model of radial basis function (RBF). Three used inputs of ANN include a registered count under Compton continuum and counts under full energy peaks of 1173 and 1333 keV. The output is a void fraction percentage. Applying this methodology, the percentage of void fraction independent of the flow pattern of a gas-liquid two-phase flow was estimated with a mean relative error less than 1.17%. Although the error obtained in this study is almost close to those obtained in other similar works, only one detector was used, while in the previous studies at least two detectors were employed. Advantages of using fewer detectors are: cost reduction and system simplification.
Reliable monitoring for detection of damage in epicyclic gearboxes is a serious concern for all industries in which these gearboxes operate in a harsh environment and in variable operational conditions. In this paper, autonomous multidimensional novelty detection algorithms are used to estimate the gearbox’ health state based on vectors of features calculated from the vibration signal. The authors examine various feature vectors, various sources of data and many different damage scenarios in order to compare novel detection algorithms based on three different principles of operation: a distance in the feature space, a probability distribution, and an ANN (artificial neural network)-based model reconstruction approach. In order to compensate for non-deterministic results of training of neural networks, which may lead to different network performance, the ensemble technique is used to combine responses from several networks. The methods are tested in a series of practical experiments involving implanting a damage in industrial epicyclic gearboxes, and acquisition of data at variable speed conditions.
At present, most of the existing target detection algorithms use the method of region proposal to search for the target in the image. The most effective regional proposal method usually requires thousands of target prediction areas to achieve high recall rate.This lowers the detection efficiency. Even though recent region proposal network approach have yielded good results by using hundreds of proposals, it still faces the challenge when applied to small objects and precise locations. This is mainly because these approaches use coarse feature. Therefore, we propose a new method for extracting more efficient global features and multi-scale features to provide target detection performance. Given that feature maps under continuous convolution lose the resolution required to detect small objects when obtaining deeper semantic information; hence, we use rolling convolution (RC) to maintain the high resolution of low-level feature maps to explore objects in greater detail, even if there is no structure dedicated to combining the features of multiple convolutional layers. Furthermore, we use a recurrent neural network of multiple gated recurrent units (GRUs) at the top of the convolutional layer to highlight useful global context locations for assisting in the detection of objects. Through experiments in the benchmark data set, our proposed method achieved 78.2% mAP in PASCAL VOC 2007 and 72.3% mAP in PASCAL VOC 2012 dataset. It has been verified through many experiments that this method has reached a more advanced level of detection.
The paper presents a methodology for parametric fault clustering in analog electronic circuits with the use of a self-organizing artificial neural network. The method proposed here allows fast and efficient circuit diagnosis on the basis of time and/or frequency response which may lead to higher production yield. A self-organizing map (SOM) has been applied in order to cluster all circuit states into possible separate groups. So, it works as a feature selector and classifier. SOM can be fed by raw data (data comes from the time or frequency response) or some pre-processing is done at first. The author proposes conversion of a circuit response with the use of e.g. gradient and differentiation. The main goal of the SOM is to distribute all single faults on a two-dimensional map without state overlapping. The method is aimed for the development stage because the tolerances of elements are not taken into account, however single but parametric faults are considered. Efficiency analyses of fault clustering have been made on several examples e.g. a Sallen-Key BPF and an ECG amplifier. Testing procedure is performed in time and frequency domains for the Sallen-Key BPF with limited number of test points i.e. it is assumed that only input and output pins are available. A similar procedure has been applied to a real ECG amplifier in the frequency domain. Results prove a high efficiency in acceptable time which makes the method very convenient (easy and quick) as a first test in the development stage.
In recent years, deep learning and especially deep neural networks (DNN) have obtained amazing performance on a variety of problems, in particular in classification or pattern recognition. Among many kinds of DNNs, the convolutional neural networks (CNN) are most commonly used. However, due to their complexity, there are many problems related but not limited to optimizing network parameters, avoiding overfitting and ensuring good generalization abilities. Therefore, a number of methods have been proposed by the researchers to deal with these problems. In this paper, we present the results of applying different, recently developed methods to improve deep neural network training and operating. We decided to focus on the most popular CNN structures, namely on VGG based neural networks: VGG16, VGG11 and proposed by us VGG8. The tests were conducted on a real and very important problem of skin cancer detection. A publicly available dataset of skin lesions was used as a benchmark. We analyzed the influence of applying: dropout, batch normalization, model ensembling, and transfer learning. Moreover, the influence of the type of activation function was checked. In order to increase the objectivity of the results, each of the tested models was trained 6 times and their results were averaged. In addition, in order to mitigate the impact of the selection of learning, test and validation sets, k-fold validation was applied.
Effects of infrared power output and sample mass on drying behaviour, colour parameters, ascorbic acid degradation, rehydration characteristics and some sensory scores of spinach leaves were investigated. Within both of the range of the infrared power outputs, 300–500 W, and sample amounts, 15–60 g, moisture content of the leaves was reduced from 6.0 to 0.1±(0.01) kg water/kg dry base value. It was recorded that drying times of the spinach leaves varied between 3.5–10 min for constant sample amount, and 4–16.5 min for constant power output. Experimental drying data obtained were successfully investigated by using artificial neural network methodology. Some changes were recorded in the quality parameters of the dried leaves, and acceptable sensory scores for the dried leaves were observed in all of the experimental conditions.