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.
Journal bearings are the most common type of bearings in which a shaft freely rotates in a metallic sleeve. They find a lot of applications in industry, especially where extremely high loads are involved. Proper analysis of the various bearing faults and predicting the modes of failure beforehand are essential to increase the working life of the bearing. In the current study, the vibration data of a journal bearing in the healthy condition and in five different fault conditions are collected. A feature extraction method is employed to classify the different fault conditions. Automatic fault classification is performed using artificial neural networks (ANN). As the probability of a correct prediction goes down for a higher number of faults in ANN, the method is made more robust by incorporating deep neural networks (DNN) with the help of autoencoders. Training was done using the scaled conjugate gradient algorithm and the performance was calculated by the cross entropy method. Due to the increased number of hidden layers in DNN, it is possible to achieve a high efficiency of 100% with the feature extraction method.
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.
This paper presents a new test method able to infer - in periods of less than 7 seconds - the refrigeration capacity of a compressor used in thermal machines, which represents a time reduction of approximately 99.95% related to the standardized traditional methods. The method was developed aiming at its application on compressor manufacture lines and on 100% of the units produced. Artificial neural networks (ANNs) were used to establish a model able to infer the refrigeration capacity based on the data collected directly on the production line. The proposed method does not make use of refrigeration systems and also does not require using the compressor oil.
The prediction of machined surface parameters is an important factor in machining centre development. There is a great need to elaborate a method for on-line surface roughness estimation [1-7]. Among various measurement techniques, optical methods are considered suitable for in-process measurement of machined surface roughness. These techniques are non-contact, fast, flexible and tree-dimensional in nature. The optical method suggested in this paper is based on the vision system created to acquire an image of the machined surface during the cutting process. The acquired image is analyzed to correlate its parameters with surface parameters. In the application of machined surface image analysis, the wavelet methods were introduced. A digital image of a machined surface was described using the one-dimensional Digital Wavelet Transform with the basic wavelet as Coiflet. The statistical description of wavelet components made it possible to develop the quality measure and correlate it with surface roughness [8-11]. For an estimation of surface roughness a neural network estimator was applied [12-16]. The estimator was built to work in a recurrent way. The current value of the Ra estimation and the measured change in surface image features were used for forecasting the surface roughness Ra parameter. The results of the analysis confirmed the usability of the application of the proposed method in systems for surface roughness monitoring.
In this paper an artificial neural network, which realizes a nonlinear adaptive control algorithm, has been applied in a control system of variable speed generating system. The speed is adjusted automatically as a function of load power demand. The controller employs a single layer neural network to estimate the unknown plant nonlinearities online. Optimization of the controller is difficult because the plant is nonlinear and no stationary. Furthermore, it deals with the situation where the plant becomes uncontrollable without any restrictive assumptions. In contrast to previous work  on the same subject, the number of neural networks has been reduced to only one network. The number of the neurons in a network structure as well as choosing certain design parameters was specified a priori. The computer test results have been presented to show performance of proposed neural controller.
Malignant melanomas are the most deadly type of skin cancer, yet detected early have high chances of successful treatment. In the last twenty years, the interest in automatic recognition and classification of melanoma dynamically increased, partly because of appearing public datasets with dermatoscopic images of skin lesions. Automated computer-aided skin cancer detection in dermatoscopic images is a very challenging task due to uneven sizes of datasets, huge intra-class variation with small interclass variation, and the existence of many artifacts in the images. One of the most recognized methods of melanoma diagnosis is the ABCD method. In the paper, we propose an extended version of this method and an intelligent decision support system based on neural networks that uses its results in the form of hand-crafted features. Automatic determination of the skin features with the ABCD method is difficult due to the large diversity of images of various quality, the existence of hair, different markers and other obstacles. Therefore, it was necessary to apply advanced methods of pre-processing the images. The proposed system is an ensemble of ten neural networks working in parallel, and one network using their results to generate a final decision. This system structure enables to increase the efficiency of its operation by several percentage points compared with a single neural network. The proposed system is trained on over 5000 and tested afterwards on 200 skin moles. The presented system can be used as a decision support system for primary care physicians, as a system capable of self-examination of the skin with a dermatoscope and also as an important tool to improve biopsy decision making.
Self-aligning roller bearings are an integral part of the industrial machinery. The proper analysis and prediction of the various faults that may happen to the bearing beforehand contributes to an increase in the working life of the bearing. This study aims at developing a novel method for the analysis of the various faults in self-aligning bearings as well as the automatic classification of faults using artificial neural network (ANN) and deep neural network (DNN). The vibration data is collected for six different faults as well as for the healthy bearing. Empirical mode decomposition (EMD) followed by Hilbert Huang transform is used to extract instantaneous frequency peaks which are used for fault analysis. Time domain and time-frequency domain features are then extracted which are used to implement the neural networks through the pattern recognition tool in MATLAB. A comparative study of the outputs from the two neural networks is also performed. From the confusion matrix, the efficiency of the ANN has been found to be 95.7% and using DNN has been found to be 100%.
Artificial neural networks are one of the modern methods of the production optimisation. An attempt to apply neural networks for controlling the quality of bentonite moulding sands is presented in this paper. This is the assessment method of sands suitability by means of detecting correlations between their individual parameters. The presented investigations were aimed at the selection of the neural network able to predict the active bentonite content in the moulding sand on the basis of this sand properties such as: permeability, compactibility and the compressive strength. Then, the data of selected parameters of new moulding sand were set to selected artificial neural network models. This was made to test the universality of the model in relation to other moulding sands. An application of the Statistica program allowed to select automatically the type of network proper for the representation of dependencies occurring in between the proposed moulding sand parameters. The most advantageous conditions were obtained for the uni-directional multi-layer perception (MLP) network. Knowledge of the neural network sensitivity to individual moulding sand parameters, allowed to eliminate not essential ones.
When an artificial neural network is used to determine the value of a physical quantity its result is usually presented without an uncertainty. This is due to the difficulty in determining the uncertainties related to the neural model. However, the result of a measurement can be considered valid only with its respective measurement uncertainty. Therefore, this article proposes a method of obtaining reliable results by measuring systems that use artificial neural networks. For this, it considers the Monte Carlo Method (MCM) for propagation of uncertainty distributions during the training and use of the artificial neural networks.
The article presents the prototype of a measurement system with a hot probe, designed for testing thermal parameters of heat insulation materials. The idea is to determine parameters of thermal insulation materials using a hot probe with an auxiliary thermometer and a trained artificial neural network. The network is trained on data extracted from a nonstationary two-dimensional model of heat conduction inside a sample of material with the hot probe and the auxiliary thermometer. The significant heat capacity of the probe handle is taken into account in the model. The finite element method (FEM) is applied to solve the system of partial differential equations describing the model. An artificial neural network (ANN) is used to estimate coefficients of the inverse heat conduction problem for a solid. The network determines values of the effective thermal conductivity and effective thermal diffusivity on the basis of temperature responses of the hot probe and the auxiliary thermometer. All calculations, like FEM, training and testing processes, were conducted in the MATLAB environment. Experimental results are also presented. The proposed measurement system for parameter testing is suitable for temporary measurements in a building site or factory.
The development of digital signal processors and the increase in their computing capabilities bring opportunities to employ algorithms with multiple variable parameters in active noise control systems. Of particular interest are the algorithms based on artificial neural networks. This paper presents an active noise control algorithm based on a neural network and a nonlinear input-output system identification model. The purpose of the algorithm is an active noise control system with a nonlinear primary path. The algorithm uses the NARMAX system identification model. The neural network employed in the proposed algorithm is a multilayer perceptron. The error backpropagation rule with adaptive learning rate is employed to update the weight of the neural network. The performance of the proposed algorithm has been tested by numerical simulations. Results for narrow-band input signals and nonlinear primary path are presented below.
The paper presents local dynamic approach to integration of an ensemble of predictors. The classical fusing of many predictor results takes into account all units and takes the weighted average of the results of all units forming the ensemble. This paper proposes different approach. The prediction of time series for the next day is done here by only one member of an ensemble, which was the best in the learning stage for the input vector, closest to the input data actually applied. Thanks to such arrangement we avoid the situation in which the worst unit reduces the accuracy of the whole ensemble. This way we obtain an increased level of statistical forecasting accuracy, since each task is performed by the best suited predictor. Moreover, such arrangement of integration allows for using units of very different quality without decreasing the quality of final prediction. The numerical experiments performed for forecasting the next input, the average PM10 pollution and forecasting the 24-element vector of hourly load of the power system have confirmed the superiority of the presented approach. All quality measures of forecast have been significantly improved.
Two low-cost methods of estimating the road surface condition are presented in the paper, the first one based on the use of accelerometers and the other on the analysis of images acquired from cameras installed in a vehicle. In the first method, miniature positioning and accelerometer sensors are used for evaluation of the road surface roughness. The device designed for installation in vehicles is composed of a GPS receiver and a multi-axis accelerometer. The measurement data were collected from recorded ride sessions taken place on diversified road surface roughness conditions and at varied vehicle speeds on each of examined road sections. The data were gathered for various vehicle body types and afterwards successful attempts were made in constructing the road surface classification employing the created algorithm. In turn, in the video method, a set of algorithms processing images from a depth camera and RGB cameras were created. A representative sample of the material to be analysed was obtained and a neural network model for classification of road defects was trained. The research has shown high effectiveness of applying the digital image processing to rejection of images of undamaged surface, exceeding 80%. Average effectiveness of identification of road defects amounted to 70%. The paper presents the methods of collecting and processing the data related to surface damage as well as the results of analyses and conclusions.
In the paper a method using active thermography and a neural algorithm for material defect characterization is presented. Experimental investigations are conducted with the stepped heating method, so-called time-resolved infrared radiometry, for the test specimen made of a material with low thermal diffusivity. The results of the experimental investigations were used in simulations of artificial neural networks. Simulations are performed for three datasets representing three stages of the heating process occurring in the investigated sample. In this work, the simulation research aimed to determine the accuracy of defect depth estimation with the use of the mentioned algorithm is descibed
This paper describes a Deep Belief Neural Network (DBNN) and Bidirectional Long-Short Term Memory (LSTM) hybrid used as an acoustic model for Speech Recognition. It was demonstrated by many independent researchers that DBNNs exhibit superior performance to other known machine learning frameworks in terms of speech recognition accuracy. Their superiority comes from the fact that these are deep learning networks. However, a trained DBNN is simply a feed-forward network with no internal memory, unlike Recurrent Neural Networks (RNNs) which are Turing complete and do posses internal memory, thus allowing them to make use of longer context. In this paper, an experiment is performed to make a hybrid of a DBNN with an advanced bidirectional RNN used to process its output. Results show that the use of the new DBNN-BLSTM hybrid as the acoustic model for the Large Vocabulary Continuous Speech Recognition (LVCSR) increases word recognition accuracy. However, the new model has many parameters and in some cases it may suffer performance issues in real-time applications.